Track L: miscellaneous.

Biomed Tech 2014; 59 (s1) © 2014 by Walter de Gruyter • Berlin • Boston. DOI 10.1515/bmt-2014-5011

S758

Supervised Speech Enhancement Nasser Mohammadiha, Simon Doclo Dept. of Medical Physics and Acoustics and Cluster of Excellence Hearing4all University of Oldenburg, Germany

Introduction In many applications, such as hearing aids and cochlear implants, reducing interference from noisy microphone recordings is very beneficial to improve the speech intelligibility and quality. When information about the acoustic environment is available, supervised speech enhancement methods can be used to obtain a higher-quality signal compared to the unsupervised speech enhancement methods. In this contribution, we present an overview of recently developed supervised methods for speech enhancement.

Methods In supervised speech denoising methods, e.g., based on hidden Markov models (HMM) and nonnegative matrix factorization (NMF), a model is considered for both the speech and the noise signals and the model parameters are estimated using training samples from those signals. The speech and the noise models are then combined to construct a model for the noisy observations, using which the noise reduction task is carried out. An advantage of these supervised methods is that there is no need to estimate the noise power spectral density using a separate algorithm.

Results Under matched training and testing conditions, supervised denoising methods can significantly outperform unsupervised methods. For example, a recently developed NMF-based noise reduction method outperformed a state-of-the-art unsupervised method by 0.2 MOS in perceptual evaluation of speech quality (PESQ), averaged over different noise types. This performance gap becomes smaller if there is a mismatch between training and testing data, where online learning methods can be used to improve the performance.

Conclusion Supervised noise reduction methods can be used to reduce highly non-stationary noise, such as babble noise, given that a noise-dependent model can be learned a-priori. If the speech signal is degraded with a different noise type than the training noise type, the enhancement performance might degrade substantially. Online model learning paradigms may then provide a solution to reduce the effect of the mismatch.

Unauthenticated Download Date | 4/21/17 2:48 PM


S759

EEG to go? Towards auditory BCIs in daily life situations M. De Vos, M. Bleichner and S. Debener, Methods in neurocognitive psychology, University of Oldenburg, Oldenburg, Germany & Cluster of Excellence Hearing4all, [email protected]

Introduction Research on Brain Computer Interfaces (BCI) is focused on developing technology to use brain signals to steer computerized devices, aiming to repair or enhance human cognitive or sensory-motor functions. BCIs largely rely on noninvasively recording brain signals with Electroencephalography (EEG) technology. However, traditional wired EEG technology does not tolerate substantial body or equipment motion, resulting in BCI experiments being performed in highly artificial laboratory environments. Our goal is to develop a mobile EEG system that reduces these heavy limitations on the validity and applicability of the results for real-life behavior and that can ultimately be used for auditory BCIs, e.g. to control hearing aid signal processing.

Methods We developed a mobile EEG system and assessed its performance as follows. We recorded 20 subjects performing a previously introduced auditory BCI experiment during seated and walking conditions. Subjects were requested to pay attention to only 1 of 3 different tones. P300 single-trial analysis, reflecting a neural correlate of auditory attention, was performed with regularized stepwise linear discriminant analysis to determine from the recorded brain signals if a single tone was attended to or not.

Results Above chance-level classification accuracies was obtained for most participants (19 out of 20 for the seated, 16 out of 20 for the walking condition), with mean classification accuracies of 71% (seated) and 64% (walking). Moreover, the resulting information transfer rates for the seated and walking conditions were comparable to a recently published laboratory auditory brain-computer interface (BCI) study.

Conclusion We demonstrated the feasibility to record correlates of auditory attention in humans freely walking around. This holds great potential for the further development of auditory BCI systems.



S760



S761

Should neural nets have ears? The role of auditory features and deep learning in automatic speech recognition Angel Mario Castro Martinez1, Niko Moritz2, Bernd T. Meyer1 1 Medical Physics, Cluster of Excellence Hearing4all, University of Oldenburg, Germany 2 Fraunhofer IDMT - Hearing, Speech and Audio Technology, Oldenburg, Germany {angel.castro, niko.moritz, bernd.meyer}@uni-oldenburg.de

Abstract Current systems for automatic speech recognition (ASR) perform well in close-talk or noise-free situations but do not function in complex acoustic scenes as well as required by assistive devices for listeners with a mild or moderate hearing loss in their everyday life. This study investigates two approaches to speech research that have the potential of substantially improving the robustness ASR: Speech features related to signal processing strategies in the healthy auditory system have been shown to increase the robustness of ASR systems in the past, and similarly, the use of deep neural networks was found to outperform traditional ASR approaches in many conditions. Since neural nets have the potential to learn the task-relevant features from a conventional filter bank representation, we investigate if the combination of auditory features and deep learning should be preferred over self-learned patterns. Specifically, noise-robust Gabor features and Amplitude Modulation Filter-Bank features, highly invariant against reverberation, are used as input to a state-of-the-art ASR system incorporating neural net processing. On a speech recognition task with a vocabulary of 5000 words, both standard features and a state-of-the-art baseline are outperformed in most acoustic conditions by using auditory processing, yielding average relative improvements of up to 69% over standard features and 21% over the state-of-the-art system. This highlights the mutual benefit of auditory signal processing and recent advances in machine learning.



S762

Design Space Exploration of Hardware Architectures for Hearing Aid Devices G. Payá-Vayá, J. Hartig, L. Gerlach, and H. Blume Institute of Microelectronic Systems, Leibniz Universität Hannover, Hannover, Germany {guipava,hartig,gerlach,blume}@ims.uni-hannover.de

Introduction By 2015, about 16% of the human population will suffer from hearing problems and will be potential users of hearing aid systems. The continuous research in signal processing algorithms, like detection and recognition of speakers in complex acoustic scenes, does not only extremely improve the hearing ability, but also increases the processing performance requirement stringently. This performance requirement in combination with the limited power consumption makes the research field of hardware architectures for digital hearing aid systems challenging.

Methods In order to meet the mentioned design goals (high processing performance at limited power consumption), different hardware architectures can be implemented. On the one hand, a dedicated hardware approach would provide the lowest power consumption by reducing the required silicon area. However, the implemented algorithms remain fixed after chip manufacturing. On the other hand, there is a trend to customize the instruction-set of a processor architecture for processing common hearing aid algorihtms efficiently. The resulting Application-Specific Instruction-Set Processor (ASIP) provides the necessary programmability and enough performance for executing future hearing aid algorithms, though requiring more power and silicon area.

Results A design space exploration was performed to measure the gap between dedicated hardware and ASIP implementations. Early implementation results show that for the same hearing aid algorithm and target semiconductor technology (TSMC Low-Power 40nm), our proposed ASIP is about 2 times bigger than the dedicated implementation, while consuming up to 15 times more power. In case of using a commercial Tensilica LX4 ASIP, the required silicon area and power consumption is about 5 and 60 times bigger than the dedicated implementation, respectively.

Conclusion The use of ASIP architectures provides the necessary programmability to execute new hearing aid algorithms without requiring to manufacture new devices whenever the algorithm changes. In terms of performance and power consumption, ASIPs offer a reasonable trade-off. The proposed ASIP architecture requires less than 1 mm2 silicon area and consumes up to 1 mW.



S763

Personalization of audio playback using intuitive self-fitting interfaces J. Rennies1, A.M. Kubiak1, S. Doclo1,2 1

Fraunhofer IDMT / Hearing, Speech and Audio Technology, Oldenburg, Germany, {rns, kbk}@idmt.fraunhofer.de Universität Oldenburg, Department Medizinische Physik und Akustik, Signalverarbeitung, Oldenburg, Germany, [email protected]

2

Introduction Personal hearing preferences are known to differ considerably between listeners, even when they are audiologically normally hearing. The reasons are not yet fully understood and, consequently, a personally fitted sound playback for enhanced listening comfort is not yet realized in practical applications such as consumer electronics. Additional need for sound personalization arises from the fact that less than 25% of people with hearing impairment are actually treated with medical hearing aids. This group of users would greatly benefit from sound personalization both with respect to listening comfort and accessibility of modern communication systems.

Methods In order to provide individualized sound to the listeners, different user interfaces (2D-touchscreen, rotary knobs, pairwise comparison, loudness-based method) for self-fitting hearing-support algorithms were tested with hearing-impaired (speech signals) and normally hearing users (music signals). The the fitting duration, test / re-test accuracy as well as usability issues for users aged 50+ were investigated.

Results The data confirm the expected large interindividual spread of preferred settings both in normally hearing and hearingimpaired listeners. This concerns not only the preferred volume settings, but also the preferred frequency shaping or frequency dependent dynamic range compression. The individual factors cannot be easily related to basic auditory parameters such as the audiogram. The investigated self-fitting interfaces seem to be highly user friendly, easy to handle, relatively quick, and accurate, especially the 2D-touchscreen and the rotary knobs.

Conclusion It is possible to meet the considerable need for audio playback individualization by means of self-fitting without expert guidance. Especially the investigated 2D interface and the haptic knobs are highly accepted also by older users and therefore provide proimising opportunities for further research and applications.



S764

Investigation of laser-tissue-interaction for optical cochlea stimulation N. Kallweit1,3, H. Maier2,3, M. Schultz1,3, N. Tinne1, A. Krüger1,3, A. Kral2,3, T. Ripken1,3 Laser Zentrum Hannover e.V., Hollerithallee 8, 30419 Hannover, Germany; 2Institute of Audioneurotechnology, Department of Otolaryngology, Hannover Medical School, Feodor-Lynen-Str. 35, 30625 Hannover, Germany; 3Cluster of Excellence "Hearing4all" e-mail: [email protected]

1

Introduction Optical stimulation of the cochlea is discussed as a therapeutic alternative for conventional treatment of sensorineural hearing loss with cochlea implants. The proposed method could potentially achieve a better intracochlear spatial delimitation resulting in an improved frequency resolution. There is, however, an ongoing debate whether pulsed laser stimulation is based on an optoacoustic effect on intact hair cells or on direct neuronal stimulation without functional hair cells. Pressure measurements of pulsed lasers were conducted for understanding the fundamental physics of the interaction.

Methods The dependence of pressure amplitudes on pulse peak power and pulse duration was investigated. Therefore, two laser systems were used in order to analyze the difference of optical stimulation between thermal and stress confinement conditions. Pulse duration of the first laser was fixed at 5ns and the pulse width of the second system was variable between 10µs and 20ms. The light was coupled into an optical fiber placed in a water tank. For pressure measurements a conventional hydrophone was placed orthogonal to the fiber. In addition, a laser vibrometer was used for a custom built hydrophone-setup.

Results The data of the conventional hydrophone measurements has shown an increasing pressure amplitude which correlates with an increasing pulse peak power. At constant pulse peak power the pressure amplitude remained stable. The custom built hydrophone has shown a higher sensitivity for the laser induced sound than the conventional hydrophone. The measurements with the conventional as well as the custom built hydrophone have shown distinct signals at laser onset and offset depending on the pulse duration. Below a specific threshold duration the split-up into on- and offset of the signal merged.

Conclusion Results speak in favour of the optoacoustic effect as the mechanism in laser stimulation of the cochlear. Furthermore, findings are consistent to formerly described compound action potential data in hearing guinea pigs.



S765



S766

Towards personalised noise reduction and directional processing in hearing aids Tobias Neher, Cluster of Excellence “Hearing4all”, Medical Physics Section, Oldenburg University Recently, there has been growing interest in the personalisation of medical interventions. In this presentation, we will give an overview of research activities conducted as part of the Cluster of Excellence “Hearing4all” that are aimed at individualising signal processing in hearing aids. In particular, we will report on a number of studies that explored potential avenues for individualising noise reduction and directional processing. Benefit from these two types of hearing aid technology can vary widely among hearing-impaired listeners, and so it is of interest to identify ways of tailoring them better to the abilities and needs of the individual. In our studies, we carried out measurements with well-matched groups of elderly hearing-impaired listeners in complex speech-in-noise situations under different aided conditions. We considered the potential influence of user characteristics such as hearing loss, cognitive function and noise sensitivity on outcomes related to speech intelligibility, listening effort and overall preference. Our analyses suggest an influence of some user characteristics in some outcome domains. We will discuss these findings with a view towards developing more effective fitting strategies that are expected to lead to greater satisfaction with hearing aids.



S767

Cocktail parties and binaural hearing aids: How Hearing technology gets us connected Birger Kollmeier and Volker Hohmann, Cluster of Excellence „Hearing4all“ and Medizinische Physik, Universität Oldenburg, D26111 Oldenburg

Even though substantial progress has recently been made in the development of hearing aids and cochlea implants, the degree of auditory rehabilitation achieved is still limited. Special problems occur for hearing impaired listeners (approx. 18% of our population) in socalled “cocktail party situations” where they have difficulties understanding the desired speech in a mixture of voices and noise. But even in quiet the “distortion component” of hearing loss may lead to a strong limitation in the perceived sound quality and speech intelligibility. In order to reach a substantial improvement, auditory models come into play – in a similar way as for MP3 audio coding. The modeling of the “effective” processing during the acquisition and processing of acoustic signals (such as music or speech) allows for the construction of “intelligent” hearing instruments that support speech intelligibility not only in cocktail party situations. The lecture provides an overview of the activities in the cluster of excellence „Hearing4all“ that spans across biophysical principles in hearing impairment, clinical applications in auditory diagnostics and rehabilitation up to assistive listening devices in daily life. A focus of the talk is the model-based compensation of the “distortion component” and the binaural (i.e., listening with two ears) compensation of degraded speech perception in cocktail party situations. The advantage of such binaural algorithms will be demonstrated using sound examples. By using field tests with wearable prototype hearing aids, these new procedures are evaluated with patients. This work on binaural hearing aids has recently been awarded the German Presidents Award for Technology and Innovation (Deutscher Zukunftspreis 2012).



S768

Exploring the effect of mastoid obliteration to the output of electromechanical transducers Martin Grossöhmichen, Burkard Schwab, Rolf Salcher, Thomas Lenarz, Hannes Maier Dept. of Otolaryngology, Hannover Medical School, [email protected]

Introduction Electromechanical transducers of implanted active middle ear devices or Direct Acoustic Cochlear Implants are normally surrounded by air of the middle ear cavity. In cases with obliterated radical mastoid cavities, the transducer is embedded in fat or scar tissue of unknown elastic properties. Here the elastic properties of these tissues and the impact on transducer output were investigated experimentally in vitro.

Methods An indentation test method for the elasticity of small (Ø3–5 mm) tissue samples was developed. Elasticity of fatty tissue samples from fresh and revision surgeries was determined and the Young’s moduli were estimated from the tests results. A phantom material having equal elasticity was identified. In a plastic mold, simulating the radical mastoid cavity, the output of transducers surrounded by the phantom material was measured with Laser Doppler Velocimetry and compared to the unloaded output.

Results The elasticity of phantom materials with different compositions was equal to the elasticity of fresh and matured human fat tissue samples. Comparisons of the output signals of embedded and not embedded transducers showed minor differences in amplitude and resonance frequency. The Young’s moduli estimated for fresh human abdominal fat were in a range comparable to published values for human breast fat.

Conclusion The developed technique simulating an obliterated radical mastoid cavity by a plastic mold and a phantom material with elasticity equal to human abdominal fat is easy to handle and adequate to estimate the effect of obliteration to the output of electromechanical transducers in vitro. Our results demonstrate that the expected decrease in output of electromechanical transducers implanted in obliterated mastoid cavities is minor and not relevant in clinical indication.



S769

Finite-element analysis evaluating the influence of middle ear and cochlear implants on the travelling wave in the cochlea Johannes Baumgart*, Matthias Bornitz#, Thomas Zahnert#, Mario Fleischer# *

Max Planck Institute for the Physics of Complex Systems, Dresden, Germany Technische Universität Dresden, Department of Medicine Carl Gustav Carus, Clinic of Otorhinolaryngology, Dresden, Germany #

The mammalian hearing organ is a remarkable biophysical systems which converts the incoming sound into a neural signal with a high resolution in frequency. The main components in the signalling chain are the middle ear, the basilar membrane, and the organ of Corti. The middle ear matches the impedance of the liquid filled inner ear to the surrounding air. Along the basilar membrane a travelling wave builds up for an oscillatory sound and peaks at a frequency dependent location due to a gradient in stiffness and geometry. Atop of the basilar membrane resides the organ of Corti wherein the mechanotransduction process takes place. By this the efferent nerves are excited tonotopically near the peak amplitude of the travelling wave. Implants in the hearing organ can rehabilitate to some degree hearing of pathological ears. They are implanted as active devices in the middle ear to amplify the mechanical signal and as cochlea implants to excite directly the nerves. These devices affect the morphometry and mechanical properties of the middle and the inner ear. Here we study the effect on the mechanics of hearing aids by a detailed three-dimensional finite-element model. The model is based on geometry data from μCT-scans of a human ear and the mechanical properties of the basilar membrane are adjusted to provide the right tonotopic map and phase decay of the travelling wave. This model allows us to compare the difference in the pressure field on the basilar membrane if a cochlear implant with variable length is inserted or a floating mass transducer at different positions in the middle ear is added. Further, the model provides a tool to predict changes in the tonotopic mapping and sensitivity.



S770

Objective assessment of cochlear-implant outcome by means of electrophysiology P. Sandmann, Department of Neurology, Cluster of Excellence “Hearing4all”, Hannover Medical School, Hannover, Germany, [email protected]

Introduction Cochlear implants (CI) can partially restore hearing, but there is a high inter-individual variability in terms of the CI outcome. This variability may be related to differences in the capacity of the central auditory system to adapt to the new CI input after implantation. Electroencephalography (EEG) may help better understand the experience-related cortical changes after implantation, and how they relate to individual differences such as for duration of deafness and age at implantation. Moreover, EEG may provide an objective measure of auditory processing in CI users and thus can help evaluate the success of cochlear implantation.

Methods There is an increasing number of studies which have used EEG to compare auditory event-related potentials (ERPs) between CI users and normal-hearing controls. Auditory processing was examined in active and passive listening conditions and at different time points after the initial CI processor setup.

Results CI users showed rapid ERP changes in the ipsilateral and contralateral auditory cortex after implantation. Despite these remarkable changes, however, auditory ERPs of postlingually deafened adults remained smaller compared with normalhearing listeners, indicating reduced cortical activation in CI users and confirming poorer auditory discrimination ability with a CI. Moreover, previous studies showed an inverse relationship between the duration of deafness and auditory ERPs, suggesting limited cortical adaptation in particular after long duration of auditory deprivation.

Conclusion ERPs provide an objective measure of the extent of restored hearing in CI users and they can be of substantial clinical value by indicating the reorganization of the auditory system after implantation. ERPs are promising in particular in terms of assessing auditory performance in infants and other populations who cannot be tested with regular speech recognition tasks.



S771

Experimental Assessment versus Audiometric Evaluation Jae Hoon Sim University Hospital Zurich, Zurich, CH-8091 Switzerland

Audiometric tests of hearing thresholds are widely used to assess outcomes of the surgical reconstructions clinically. Experimental assessments can be combined with the audiometric tests, for more comprehensive analysis of the surgical reconstructions including factors affecting the functional performance. For the experimental assessment to be reliable, it necessarily reflects clinical data from the patients. Our established methods for experimental assessments, which measure objective quantities in middle and inner ears, were applied to assess surgical reconstructions of (1) myringoplasty, (2) stapes surgery, and (3) surgery with a transcutaneous bone conduction implant. The results were compared with the relevant clinical data of corresponding patients’ audiometric tests. In the case that there exists difference between the two quantities, analysis was made to identify factors causing the difference. While the results from the experimental assessments generally showed agreement with clinical data of the patients, some difference in a specific frequency range was observed. Experimental assessments on stapes surgery showed difference from air-bone gaps of the patients around 2 kHz, where the patients have depression of bone conduction preoperatively. Incomplete postoperative recovery of the depressed of bone conduction was identified as the reason of the difference. Pre-clinical data on surgery with a transcutaneous bone conduction implant also showed difference from the corresponding patients’ data in the frequency range above 2 kHz. The difference was due to placement of the device using steel-band/head-band in pre-clinical measurements, with which the transferred force is different from the implant condition. When the transferred force was calibrated and the results were calibrated by the transferred force, pre-clinical and clinical data showed similar results. The experimental methods, which provide objective and pre-clinical assessment of surgical reconstruction, should be verified in comparison with clinical data, prior to their application to evaluation of the surgical reconstruction.



S772

The use of air conduction and bone conduction in audiology as a diagnostic tool

Prof. Dr.sc. techn. Dr.med. Martin Kompis University hospital of ENT, Head and Neck surgery, Inselspital, University of Bern, 3010 Bern, Switzerland

The human ear is adapted for the reception of acoustic stimuli reaching the inner ear through the external auditory canal and the middle ear. This acoustic path is called air conduction (AC). There is, however, another path, called bone conduction (BC), in which vibrations travel through the skull to both inner ears. BC is substantially less energy efficient, the difference lying in the order of magnitude of 50 dB. As a consequence, bone conduction plays only a minor role in the everyday life of normal hearing people, with one notable exception: one’s own voice is transmitted partially through BC, which results in a substantially different subjective impression for the speaker than for all other listeners. Hearing thresholds measurements using AC and BC are probably the most frequently used tests in clinical audiology. Combining AC and BC thresholds allows the clinician not only to assess the severity of a given hearing disorder as a function of the frequency, but also to distinguish between different kinds of hearing disorders. The two main categories are conductive and sensorineural hearing loss. This differentiation is essential for patients and physicians alike, as the therapeutic options differ between the two types. For instance, surgical procedures to improve sound transmission through the middle ear are only useful in certain types of conductive hearing losses, but not in sensorineural hearing loss. Measuring bone conduction thresholds for both ears separately is not trivial, as the sound level at both inner ears is rather similar as soon as sound is transmitted by bone conduction. The different characteristics of air conduction and bone conduction are also employed in much simpler screening tests as e.g. some tuning fork test. These tests give the clinician a rough overview over the hearing status of a patient in just a few seconds time.



S773

Experimentelle Audiologie: Druckmessungen in der Gehörschnecke zur Untersuchung der Schallübertragung. Christof Stieger1,2,3, Defne Abur4, Julie P. Merchant1, Kourosh Roushan2, Rosemary B Farahmand11, J Rosowski1, Hideko Heidi Nakajima1 1

Massachusetts Eye and Ear Infirmary, Harvard Medical School, USA HNO Klinik, Universitätsspital Basel, Schweiz 3 HNO Klinik, Universitätsspital Bern, Schweiz 4 Smith College, Boston, USA 2

Grundlage: Die Dimensionen und physiologischen Bewegungen im menschlichen Ohr sind sehr klein. Der der sogenante Steigbügel ist ein Mittelohrknochen und mit rund 5 mm der kleinste menschliche Knochen. Dessen Vibrationsamplitude beträgt bei einem gerade hörbaren Sinustonton von 1000 Hz weniger als 10-10 m. Physiologische Experimente im Felsenbein stellen somit messtechnisch hohe Anforderungen. So sind heute keine käuflichen Drucksensoren erhältlich, welche klein genug sind um in die Gehörschnecke (Cochlea) einzuführen und gleichzeitig physiologische Schalldruckpegeln messen können. Elisabeth Olson hat 2004 neuartige fiberoptische Mikrosensoren vorgestellt [1], welche in aufwändiger Handarbeit produziert werden, dafür aber die Anforderungen für physiologisch Schalldruckmessungen in der Cochlea erlauben. Diese Drucksensoren haben einen ähnlichen Durchmesser (160um) wie ein menschliches Haar. Wir präsentieren hier den Einsatz solcher Drucksensoren für die Erforschung der Übertragungseigenschaften des hörbaren Schalls (20-20‘000 Hz) zur menschlichen Cochlea. Methode: Eine Methode wurde entwickelt, mit welcher in humanen postmortalen Präparaten (Felsenbeinen) der Druck innerhalb der Cochlea gemessen werden kann [2,3]. Zuerst muss ein Zugang zum Innenohr mit einem Knochenfräser erstellt werden, ohne die knöchernen Strukturen des Mittelohres und die flüssigkeitsgefüllten Strukturen der Cochlea zu beschädigen. Danach wird das Mittelohr geflutet und die in der Cochlea zwei sehr feine Löcher (ca. 200um) erstellt, je ein Drucksensor eingeführt, mit einem Dentalmaterial (Jeltrade) abgedichtet und letztlich die Flüssigkeit aus dem Mittelohr wieder abgesogen. Diese Prozedur minimiert das Risiko eines Lufteinschlusses in der Cochlear während der Einführung der Sensoren. Die Sensoren liegen nun auf beiden Seiten der sogenannten Basilarmembran, welche die drucksensitiven Haarzellen beherbergt. Es ist bekannt, dass der Druckabfall über der Basilarmembran korreliert mit neurophysiologischen Messungen des Gehörs [4] Somit können wir mit den Drucksensoren den cochleären Input messen. Nun wurde drei verschiedene Anregungsarten des Ohres (1) Lautsprecher im Gehörgang, (2) mechanische Stimulation des sogenannten runden Fensters, (3) Vibration des gesamten



S774

Knochens bei physiologische und pathologischen Zustände des Ohres (i) normales Mittelohr (ii) blockiertes Mittelohr, (iii) partiell fehlendes Mittelohr, (iv) zusätzliches Loch im Innenohr untersucht. Resultate: Die Messungen zeigen, dass die mechanische Impedanz des Mittelohrs eine wesentlichen Einfluss auf die Druckabfall über der Cochlea. Drei klinisch oder empirische bekannte Phänomene konnten experimentell nachgewiesen oder mindestens teilweise erklärt werden: A) sogenannte Carhart Senke, B) die Wirkungsweise von implantierbaren Hörsystemen bei der Ankopplung am runden Fenster und C) Hypersensitivität der Knochenleitung bei einem zusätzliche Loch im Innenohr. Schlussfolgerung: Intracochleäre Druckmessungen erlauben die Untersuchung unterschiedlichster klinisch relevanter Phänomene der Übertragungseigenschaften des hörbaren Schalls. Sie sind aber mit heutigen Messmethoden sehr aufwändig und es existieren keine käufliche Sensoren.

Funding: NIH/NIDCD R03DC011158 and R01DC013303 [1] Olson ES., 1998. Observing middle and inner ear mechanics with novel intracochlear pressure sensors. J Acoust Soc Am: 103: 3445-3463. [2] Nakajima HH, Dong W, Olson ES et al., 2009. Differential intracochlear sound pressure measurements in normal human temporal bones. J Assoc Res Otolaryngol: 10: 23-36. [3] Stieger C, Rosowski JJ, Nakajima HH., 2013. Comparison of forward (ear-canal) and reverse (round-window) sound stimulation of the cochlea. Hear Res: 301: 105-114. [4] Dancer, A., Franke, R., 1980. Intracochlear sound pressure measurements in guinea pigs. Hear. Res. 2, 191-205.



S775

Multi-Frequency Acquisition of DPOAE Input-Output Functions for Auditory-Threshold Estimation D. Zelle, J. P. Thiericke, A. W. Gummer, E. Dalhoff Dept. of Otolaryngology, Eberhard-Karls-University Tübingen, Tübingen, Germany [email protected]

Abstract Distortion product otoacoustic emissions (DPOAEs) evolve as a byproduct of the nonlinear amplification process of two stimulus tones in the cochlea and comprise a nonlinear-generation and a coherent-reflection component. Wave interference between these components may limit the diagnostic validity of DPOAEs for assessing the function of the inner ear. Utilizing the extracted nonlinear-generation component from DPOAE signals acquired with pulsed f2-tones increases the accuracy of auditory threshold estimates computed from semi-logarithmic input-output (I/O) functions. However, the acquisition of pulsed DPOAEs required up to now a considerably longer measurement time to attain sufficient signal-to-noise ratio. The measurement time may be decreased by shortening both the f2-pulse and also the acquisition blocks for ensemble averaging. However, undersized blocks bear the risk that slowly decaying time responses interfere with the DPOAE signal in the subsequent block, resulting in reduction of the accuracy of the auditory-threshold estimate. Here, to further decrease the measurement time, we introduce a new method for presenting pulse-train stimuli for quasi-simultaneous acquisition of multiple I/O-functions. DPOAE I/O-functions were acquired from 16 normal-hearing subjects with the new acquisition technique for a set of frequencies with f2 = 1, 1.5, 2, 3 and 4 kHz. Estimated distortion product thresholds (EDPTs) derived from I/O-functions were compared with thresholds obtained by Békésy audiometry. EDPTs correlate with Békésy thresholds (r2 > 0.36) with high significance (P < 0.001). The standard deviation of the auditory-threshold estimates was 5.2 dB SPL for the pooled data. The average measurement time was 13.6 ± 3.7 min. This algorithm enables auditory-threshold estimation for a set of frequencies with high accuracy and reasonable cost of measurement time; it represents a promising diagnostic tool for objective auditory-threshold assessment.

1

Introduction

Presenting two stimulus tones to the inner ear evokes distortion-product otoacoustic emissions (DPOAEs) as a byproduct of the nonlinear cochlear amplification process which may be used to assess the function of the outer hair cells objectively [1]. According to a widely accepted model, DPOAEs consist of two source components, a nonlinear-generation and a coherent-reflection component [2]. The presence of two source components can be visualized during the onset and offset of the DPOAE response when the second stimulus tone is pulsed [3]. Wave interference between both components limits the diagnostic validity of DPOAEs and is the major reason for the high standard deviation of auditory-threshold estimation [4] by means of semi-logarithmic DPOAE input-output (I/O) functions [5]. Because the generation site of the nonlinear-generation component is close to the tonotopic place of the frequency f2 of the second stimulus tone, this component is the signal of interest when estimating auditory threshold on the basis of DPOAEs. Using a pulsed f2-tone, the nonlinear generation component can be extracted without significant interference from the coherent-reflection component by a technique called onset-decomposition (OD), which exploits the different latencies of both source components and enables the quantification of the nonlinear-generation

component in the time domain [6]. Extracting the nonlinear-generation component with f2-pulses of 100 ms duration enhances the accuracy of auditory-threshold estimates derived from semi-logarithmic I/O-functions but at the cost of increased measurement time [4]. Recently, we introduced a refined measurement paradigm, called short-pulse DPOAE acquisition, reducing the length of the f2-tone to 8 ms [7]. This method enables auditory-threshold estimation by means of OD while decreasing the measurement time considerably compared with f2-tones of 100-ms length, due to shortening of the acquisition blocks used for ensemble averaging. However, undersized blocks bear the risk that slowly decaying time responses interfere with the DPOAE signal in the subsequent block resulting in an accompanying reduction of signal fidelity. In this paper, we present a new technique incorporating pulse-train stimuli for quasi-simultaneous acquisition of multiple I/O-functions in order to decrease the measurement time while retaining accuracy of the auditory-threshold estimates.

2

Methods

2.1

DPOAE Signal Acquisition

DPOAEs were recorded from 16 normal-hearing subjects (age: 30 ± 6.4 yr.) with auditory thresholds better than 20 dB HL, using six primary-tone level pairs according to



the scissor paradigm L1 = 0.4L2 + 39 dB SPL to account for the different compression of the traveling waves evoked by the stimulus tones in the cochlea [8]. L2 varied from 25 to 65 dB SPL. DPOAEs were recorded for a set of frequencies with f2 = 1, 1.5, 2, 3, and 4 kHz and a constant ratio of f2/f1 = 1.2. DPOAE acquisition was performed unilaterally using an ER-10 C DPOAE probe system (Etymotic Research, Elk Grove Village, IL) connected to an IBM-compatible PC equipped with a 16-bit analog output card and a 24-bit signal acquisition card (NI PCI 6733 and NI PCI 4472, National Instruments, Austin, TX). The sampling frequency was 102.4 kHz. Both stimulus generation and signal acquisition were controlled by measurement software implemented in LabVIEW (Ver. 12.0, National Instruments, Austin, TX). The f1-tone comprised sequenced pulses each of 35 ms steady-state length and cosine-shaped rising and falling edges of 2.5 ms. The Hanning-shaped f2-pulses started 10 ms after the onset of their corresponding f1-pulse and had variable half-widths, THW = c/f, with c = 13.07 estimated from the results of Vetešník et al. [6], to account for the frequency-dependent latencies of the DPOAE source components. The total length of a single acquisition block was 200 ms. Data acquisition was terminated if a signal-to-noise ratio (SNR) of 10 dB was reached at f2 = 1 kHz, or at a maximum acquisition of 800 blocks. Consecutive phase shifts of the f1- and f2-tones by 90° and 180°, respectively, enabled cancellation of the primary tones in the ensemble-averaged signal [9]. Blocks with excessive noise were discarded from the averaging by a suitable artifact threshold. Post-processing was done in Matlab (Ver. 8.2, MathWorks, Natick, MA). Because a single recording contained multiple DPOAE signals associated with multiple stimulus frequencies, we call this measurement paradigm “multi-frequency acquisition”.

S776

Figure 1: Short-time Fourier transforms (STFT, N = 1024) of the f1-tones (yellow), f2-tones (red), and the corresponding band-pass filtered DPOAE signals (blue) with normalized amplitude. For visualization purposes, values below a certain threshold are not displayed. The x-y plane shows the contour plots of the associated STFTs. (Subject S001; L2 = 45 dB SPL). Extracted DPOAE signals with a SNR ≥ 10 dB were accepted for the computation of I/O-functions. A minimum of three data points was necessary to extrapolate distortion product thresholds from the semi-logarithmic I/Ofunctions. DPOAEs showing saturation behavior at high stimulus levels (open red circle in Fig. 2c) were excluded from the computation of the regression lines by an iterative algorithm which maximizes the correlation coefficient r of the I/O-functions [4].

2.2 Computation of I/O-functions Consecutive stimulus pairs were arranged in such a way as to provide sufficient distance in the frequency domain to allow band-pass filtering and to avoid overlap between stimulus tones and DPOAEs (see Fig. 1). Zero-phase bandpass filtering with auto-regression extrapolation to reduce filter-edge effects enabled the extraction of the cubic distortion products at frequencies fdp = 2f1-f2 from the multi-frequency recordings. Circular expansion resulted in DPOAE signals without discontinuities even in cases where the signal exceeds the end of the acquisition block. DPOAE primary-source extraction was achieved by OD [6], i.e. sampling the envelope of the DPOAE signal at a time instant before wave interference occurs (see Fig. 2b). The envelope was computed as the absolute value of the Hilbert transform of the DPOAE signal |H{p(t)}| and sampling instants were commensurate to T HW. Semilogarithmic I/O-functions (Fig. 2c) were derived from the extracted primary-source components to compute the estimated distortion product threshold (EDPT) by means of linear regression [5].

Figure 2: (a) DPOAE signal p(t) extracted from a multifrequency recording by band-pass filtering (Subject S004; f2 = 4 kHz, L2 = 45 dB SPL). (b) |H{p(t)}| of the DPOAE signal (blue) shown in (a) and the OD value (red dot) used to quantify the amplitude of the primary-source component POD. (c) Semi-logarithmic I/O-function obtained from the amplitudes of the extracted primary-source components. The intersection of the extrapolated regression line with the abscissa yields the EDPT. Triangles represent noise values.



The computed EDPT values were accepted for auditorythreshold estimation if the corresponding I/O-functions satisfied the objective evaluation criteria r2 ≥ 0.8 and σEDPT ≤ 10 dB SPL [5]. Accepted EDPTs were compared to thresholds obtained by Békésy audiometry. Békésy thresholds (BTHs) were computed as the mean of three successively recorded measurements at frequencies f2 and neighboring frequencies within the bandwidth (B = 1/THW) of the f2-pulse and with frequency spacing of Δf = 20 Hz. Correction for outliers and averaging over frequency yielded BTHs which correspond to responses to stimulation of the basilar membrane with short f2-pulses [4].

3

Results

The multi-frequency method yields I/O-functions from which reliable auditory-threshold estimates are obtainable. Highly significant correlations (P < 0.001) of the BTH with the EDPT are obtained for the pooled data and for f2 ≥ 3 kHz (Fig. 3a). Fig. 3b shows the standard deviation of the auditory-threshold estimate assuming a linear relationship between the BTH and the EDPT with either a constant slope, s, of 1.18 taken from a study by Boege and Janssen [5], or variable slope for f2 = 3 kHz and 4 kHz because at these frequencies correlations with r2 > 0.42 were obtained. The corresponding values are s = 0.88 (t14 = 6.27; P = 10-5) and 0.70 (t14 = 3.10; P = 0.004). The value s = 1.18 can be assumed to be reliable due to the high threshold range in their study and the significant correlation (r2 = 0.42). The standard deviation σ of the pooled data was 5.2 dB SPL. σ varies with stimulus frequency, exhibiting a minimum of 2.6 dB SPL at f2 = 3 kHz. Fig. 3b also plots the standard deviation of the BTH measurements (gray bars).

Figure 3: (a) Squared correlation coefficient r2 of BTH with EDPT as function of f2 and for the pooled data (“all”). (b) shows the corresponding standard deviation σ of the auditory-threshold estimate. Filled bars emphasize data exhibiting a highly significant correlation (P < 0.001). Gray bars in (b) depict the standard deviations of the BTH measurements. Fig. 4 shows the BTH as a function of the sound pressure level of the EDPT for the pooled data (Fig. 4a), f2 = 3 kHz (Fig. 4b), and 4 kHz (Fig. 4c). For comparison, the black circles in Fig. 4a show data acquired with continuous stimulus tones in the frequency range 1.5 ≤ f2 ≤ 2.5 kHz from a recent study [4]. The estimated thresholds obtained

S777

with the conventional measurement paradigm, i.e. continuous stimulation, exhibits pronounced scatter and high standard deviation of 10.4 dB SPL. The agreement of the red regression lines with the measurement data indicates a linear relationship between the BTHs and the EDPT values computed from the I/O-functions when the extracted non-linear generation component is used. For some I/O-functions the a priori determined sampling instant for onset-decomposition was insufficient to extract the primary-source component, causing an exclusion rate of 3.8% (3/80) according to the objective evaluation criteria for I/O-functions, as well as outliers in the auditory-threshold estimate. Figure 5 shows the median values of the slopes of the semi-logarithmic I/O-functions (red) (a) and the normalized slopes (b) as function of f2. The gray area covers 68.3% of the data computed as the 15.85%- and 84.15%-quantiles. The magnitude of the slopes varies inter-individually but exhibits a similar frequency dependence for all subjects with a maximum slope at f2 = 1.5 kHz, except for three subjects with peak slopes at 1 and 2 kHz. Above frequency of the maximum the slope decreases monotonically with 6dB/octave.

Figure 4: Auditory threshold acquired with Békésy audiometry as a function of the sound pressure level of the estimated threshold LEDPT (red dots) gained from DPOAE I/O-functions for pooled data (a) and for frequencies f2 = 3 kHz (b) and 4 kHz (c). Black circles in (a) represent data acquired with continuous stimulus tones from a recent study [4] exhibiting a standard deviation of 10.4 dB SPL. Black and red lines depict the corresponding regression lines with slope s = 1.18 (a), 0.88 (b), and 0.70 (c). Values for the squared correlation coefficient are r2 = 0.74 (f2 = 3 kHz) and r2 = 0.69 (f2 = 4 kHz).



The finding of maximum slope at 1.5 kHz is in good agreement with middle-ear transfer functions, which show on average an amplitude maximum at 1.2 kHz [10]. It is known that the slope of the I/O-function depends on the forward and reverse transmission properties of the middle ear [11]. The average measurement time to acquire all EDPTs per subject was 13.6 ± 3.7 min. Because the termination condition for data acquisition was applied to the DPOAE presumably exhibiting the smallest SNR, i.e. for f2 = 1 kHz, the signal quality is considerably better for higher frequencies as a result of the large number of additional acquisition blocks for those frequencies.

Figure 5: (a) Red line depicts the median values of the slope for the pooled data while the gray area shows the coverage of 68.3% of the data. (b) Same as (a), but with slope values normalized to the maximum of each subject.

4

Conclusion

Multi-frequency acquisition of short-pulse DPOAE I/Ofunctions enables quasi-simultaneous auditory-threshold estimation for a set of frequencies with high accuracy and reasonable cost of measurement time. The sequential arrangement of the primary tones allows shortening of the DPOAE stimulus to 40 ms while the analysis window increases to the combined length of all f1-tones, i.e. 200 ms in this study. Thus, the I/O-functions exhibit a reduced susceptibility to long DPOAE decay times, for instance due to multiple internal reflections [12] or the synchronization of spontaneous otoacoustic emissions to the distortion product [13]. The presented method achieves an auditory-threshold estimate with low standard deviation covering a clinically relevant frequency range between f2 = 1 and 4 kHz. For f2 ≥ 3 kHz the standard deviation of the auditorythreshold estimates is only 1.5 dB SPL higher than the corresponding standard deviation of the BTH measurements. Furthermore, the consistency of the normalized slope values over subjects (Fig. 5b) suggests that the slope of the DPOAE I/O-functions can be used as an additional diagnostic parameter to differentiate between conductive and sensorineural hearing loss [4]. The multi-frequency acquisition of DPOAE I/Ofunctions allows objective estimation of the auditory threshold with lower standard deviation than hitherto possible and with reasonable expenditure of time. Therefore, this method represents a promising diagnostic tool for objective auditory-threshold assessment, especially

S778

for the newborn hearing screening and in the field of pediatrics.

4

References

[1] Lonsbury-Martin B, Martin GK. Otoacoustic emissions. Curr Opin Otolaryngol Head Neck Surg 2003; 11: 361-366. [2] Shera C. Mechanisms of mammalian otoacoustic emissions and their implications for the clinical utility of otoacoustic emissions. Ear Hear 2004; 25: 86-97. [3] Talmadge CL, Long GR, Tubis A, Dhar S. Experimental confirmation of the two-source interference model for the fine structure of distortion product otoacoustic emissions. J Acoust Soc Am 1999; 105: 275-292. [4] Dalhoff E, Turcanu D, Vetešník A, Gummer AW. Two-source interference as the major reason for auditory-threshold estimation error based on DPOAE input-output functions in normal-hearing subjects. Hear Res 2013; 296:67-82. [5] Boege P, Janssen T. Pure-tone threshold estimation from extrapolated distortion product otoacoustic emission I/O-functions in normal and cochlear hearing loss ears. J Acoust Soc Am 2002; 111: 1810-1818. [6] Vetešník A, Turcanu D, Dalhoff E, Gummer AW. Extraction of sources of distortion product otoacoustic emissions by onset-decomposition. Hear Res 2009; 256: 21-38. [7] Zelle D, Gummer AW, Dalhoff E. Extraction of otoacoustic distortion product sources using pulse basis functions. J Acoust Soc Am 2013; 134: EL6469. [8] Kummer P, Janssen P, Hulin P, Arnold W. Optimal L1-L2 primary tone level separation remains independent of test frequency in humans. Hear Res 2000; 146: 47-56. [9] Whitehead ML, Stagner BB, Martin GK, LonsburyMartin BL. Visualization of the onset of distortionproduct otoacoustic emissions, and measurement of their latency. J Acoust Soc Am 1996; 100: 16631679. [10] Aibara R, Welsh JT, Puria S, Goode RL. Human middle-ear sound transfer function and cochlear input impedance. Hear Res 2001; 152: 100-109. [11] Kummer P, Schuster EM, Rosanowski F, et al. The influence of conductive hearing loss on DPOAE threshold. The effect of an individually optimized stimulation. HNO 2006; 6: 457-467. [12] Dhar S, Talmadge CL, Long GR, Tubis A. Multiple internal reflections in the cochlea and their effect on DPOAE fine structure. J Acoust Soc Am 2002; 112: 2882-2897. [13] van Dijk P, Wit HP. Synchronization of spontaneous otoacoustic emissions to a 2f1-f2 distortion product. J Acoust Soc 1990; 88: 850-856.



S779

How to measure CI performance objectively by using EEG M. Finke Cluster of Excellence "Hearing4all", Department of Otolaryngology, Medical University of Hannover Hannover Germany [email protected] P. Sandmann Cluster of Excellence "Hearing4all", Department of Neurology, Medical University of Hannover Hannover Germany [email protected] A. Büchner Cluster of Excellence "Hearing4all", Department of Otolaryngology, Medical University of Hannover Hannover Germany [email protected]

Introduction Speech tests and psychophysical measures always rely on the participants’ response (via a verbal or motor response), but Electroencephalography (EEG) can provide important insights to the neural processing of stimuli. The event-related potential (ERP) P3 peaks around 300 ms after stimulus presentation. It represents a higher level of neural processing and can be related to the time a person needs to classify stimuli. Additionally, earlier ERPs (N1-P2 complex, or the mismatch-negativity; MMN) can give insights on the automatic processing of the stimuli.

Methods CI users and normal hearing participants were presented with different speech and non-speech sounds. In case of bilateral implantation, the better site was selected. Age-based normal hearing was assured by pure tone audiometry prior to the EEG recording. We used a 82 EEG channel set-up (nose reference). Eye blinks and CI artifacts were reduced using independent/principle component analysis. Then, data were offline filtered (0.1-30 Hz). Individual ERPs were extracted for each experimental condition, and grand averages (GA) were computed separately for both groups. Response time (RT) in ms and hit rate (HR) in percent were calculated as performance measures.

Results We will report our results of differences in amplitude and latency between CI users and NH participants. Further, we will show the relationship between these ERP data, the behavioural discrimination ability and the speech recognition ability obtained from the clinical speech tests.

Conclusion Our EEG paradigms offer an objective measure of CI performance across different patient groups as well as in comparison to NH participants. The high time resolution of EEG enables us to investigate different processing stages of acoustic stimuli in individuals with a CI.



S780

Objective metric parameters in Cochlear Implant imaging in comparison to hearing performance Würfel W, Büchner A, Lenarz T Department of Otorhinolaryngology, Head and Neck Surgery, Hannover Medical School, Germany, email: [email protected]

Introduction: Cochlear Implants (CIs) are medical devices used to restore the sense of hearing in patients with profound hearing loss and deafness. In children they enable the possibility of speech development. A crucial topic is the assessment of the individual metrics in the cochlear size and the position of the CI array in a clinical setting. Measurement of the cochlear length (CL) is not standardized. However, insertion angle determination of the CI array is the gold standard in postoperative CI imaging. Nevertheless, this is a descriptive procedure and is often not possible to be objectified in terms of metric insertion depth evaluation.

Methods 116 patients who were implanted either with a Med-EL Flex 20 (n=27), Flex 24 (n=23), Flex 28 (n =46) or MedEL standard electrode (n=20) were retrospectively evaluated for CL in preoperative cone beam computed tomography (CBCT). Postoperatively, insertion parameters as Cochlear Coverage (CC) and insertion success have been determined.

Results Statistical significant differences can be shown between CC of the evaluated electrode groups (p < .001). Flex 20: Mean = 0.56 ± .03; Flex 24: Mean = 0.7 ± .04; Flex 28: Mean = 0.79 ± .06; Standard: Mean = 0.87 ± .05. Data shows positive correlation between CC and speech performance (p < .04).

Discussion It could be shown that objective parameters in cochlear size and postoperative insertion parameters can be collected and that there are significant differences between the electrode array groups. Not only different electrode array lengths, but also different CL have a significant influence to CC. This is the first time that this issue can be objectified. The CC is a crucial parameter for individualized CI treatment and may be a predictor for speech performance. This can be seen in the correlation between CC and hearing performance.



S781

Consideration of Temporal Masking in Cochlear Implant Speech Processing Strategies: the TPACE Strategy E. Kludt, T. Lenarz, A. Büchner Department of Otolaryngology, Medical University Hannover, Hannover, Germany [email protected]

Introduction Cochlear implant (CI) patients have extensive problems to understand speech in challenging environments as with background noise or reverberation. One factor that restricts speech intelligibility is the limited interface between CI electrodes and the auditory nerve. Identification of relevant speech signal components and concentration on their presentation by the sound coding strategy may improve speech intelligibility for CI patients. The novel speech coding strategy TPACE uses a temporal masking model to reduce successive stimulations of the same CI electrodes. These stimulations are assumed to be superfluous due to electrophysiological properties of the auditory nerve fibres. In this study we evaluate the speech intelligibility with TPACE and the relationship of electrophysiological measures to the performance with TPACE.

Methods Acute streaming experiments using the Nucleus Implant Communicator research environment were performed. The strength of TPACE temporal masking is described by the Temporal Masking Half-life T½. This time constant defines the time interval after which the strength of temporal masking has decreased to its half. In the first experiment, twelve CI subjects were tested for speech intelligibility in noise using TPACE with T½ of 0.5 and 1.1 ms as well as in a control condition without temporal masking. In the second experiment, the measurement of two additional time constants (0.4 and 0.8 ms) and the individual recovery functions were included into the experimental protocol. Up to date, five CI subjects were measured in the second experiment.

Results A statistically significant increase in performance was found for TPACE with T½ = 0.5 ms compared to the control condition without temporal masking. No correlation between the optimal T½ and the recovery function was observed within the preliminary dataset of the second experiment.

Conclusion It seems that the consideration of temporal masking might improve speech intelligibility for CI patients.



S782

Effect of dynamic range and location on temporal pitch perception of cochlear implant users RT Penninger1, CJ Limb2,3, I Dhooge4 and A Büchner1 1 Cluster of Excellence "Hearing4all", Medical University Hannover, Germany, [email protected] 2 Department of Otolaryngology, Johns Hopkins School of Medicine, Baltimore, MD, USA 3 Peabody Conservatory of Music, Baltimore, MD USA 4 Department of Otolaryngology, University Ghent, Ghent, Belgium

Abstract The coding of pitch in modern cochlear implant devices is challenging for recipients as it is mainly based on place pitch, which is subject to the location of a finite number of electrodes on an array. The coding of pitch via temporal cues in the stimulation may help to improve pitch perception since most cochlear implant subjects are able to perceive rate changes on a single electrode up to around 300 Hz. Some optimally performing subjects are able to perceive temporal pitch up to 1000 Hz. However, performance varies highly between subjects and depends on the selected electrodes. The first objective of this study was to quantify the effect of electrical dynamic range on temporal pitch perception. Therefore pitch ranking was measured on electrodes with the widest and the narrowest dynamic ranges. The second objective was to investigate the effect of the location of the electrodes on pitch ranking and therefore a basal, a middle and an apical electrode were additionally tested for each subject. The results show a high degree of intersubject variability. However, the electrical dynamic range was found to correlate with the pitch ranking score. No consistent effect was observed based on location on the implant array. Improving CI performance based on individual subject and electrode performance might be the key to better understanding of the mechanisms behind temporal pitch perception and it could lead to improved language and music perception.

1

Introduction

For cochlear implant (CI) users music can sound unpleasant mainly due to a lack of accurate pitch perception [1]. Pitch is an important attribute of many types of music and just noticeable differences (JND) in pitch for CI subjects are a lot worse compared to normal hearing (NH) control groups [2, 3]. Additionally CI subjects frequently confuse the direction of pitch change and this phenomenon is known as pitch reversal [4]. These pitch perception problems arise from the way the CI encodes pitch. The placement of the electrodes makes use of the tonotopic arrangement of the cochlea (place pitch) which varies from the basal end which responds optimally to high frequencies, to the apical end which responds optimally to low frequencies [5]. The pitch range in the low frequency region (apical end) is cut off by the insertion depth of the electrode [6]. Changing the time structure of the electrical stimulation can also lead to changes in pitch perception and this is referred to as temporal pitch. Temporal pitch can be created by sinusoidal amplitude modulation (SAM) of a fixed carrier rate on a single electrode [7]. There is an upper rate pitch limit of around 300 Hz on most [8, 9] but not all subjects [10]. The electrical dynamic range (DR) of an electrode refers to the current level between that which is required to elicit a just noticeable sensation and that which is perceived as being loud but tolerable. In the fitting of a CI to a recipient the DR is individually adjusted for each electrode. The size of the DR determines the depth of modulation for temporal pitch based on SAM. The perception of SAM pitch becomes exponentially weaker as the modulation depth decreases until it is perceptually similar in pitch to that of an unmodulated pulse train [11].

Stimulation of apical electrodes with SAM below 1 kHz could result in better pitch discrimination compared to basal electrodes due to a better match between temporal and place pitch. Performance seems to depend heavily on the selection of the appropriate electrode for each subject. Kong et al. observed that subjects’ ability to discriminate rate differences varied significantly depending on the electrode site stimulated [10]. Similar results have been reported by by Zeng and by Baumann and Nobbe [8, 12]. The first goal of the present study was to evaluate the effect of DR on temporal pitch perception. We hypothesized that electrodes with a wide dynamic range (WDR) may perform better than NDR electrodes. Our second goal was to evaluate the effect of location on pitch ranking performance. Our second hypothesis was that performance might gradually improve as the site of stimulation on the implant array is moved towards the apex due to a better match between SAM and the place of stimulation. Knowledge of the characteristics and locations of electrodes where subjects perform well on temporal pitch tasks has the potential to improve pitch processing strategies, which in turn may benefit the perception of music and speech intonation.

2

Methods

Ten CI users participated in the experiment. All subjects used Cochlear devices (Cochlear Ltd., Sydney, Australia) and relevant details are shown in Table 1. All subjects had more than six months of experience with their implant system. The experiments were performed at the Department of Otolaryngology, Head and Neck Surgery of the University Hospital Ghent, Belgium and at the Medical University Hannover, Germany under a research protocol approved



by their Ethical Committees. Written consent was obtained from all participants. Sub- Age Processor ject (years) Type S1 71 CP 810 SP S2 35 CP 810 SP S3 73 Freedom SP S4 30 CP 810 SP S5 69 Freedom SP S6 76 CP 810 SP S7 73 Freedom SP S8 57 CP 810 SP S9 55 CP 810 SP S10 63 CP 810 SP

CI experience (months) Implant type 17 CI512 25 CI512 39 CI24R (CA) 8 CI512 62 CI24R (CA) 7 CI512 52 CI24R (CA) 18 CI512 87 CI512 16 CI24R (CA)

Coding Strategy ACE ACE ACE ACE MP3000 ACE ACE ACE ACE ACE

Table 1 shows details about subject demographics. All stimuli were delivered via the L34 research sound processor using Nucleus Implant Communicator software (Cochlear Ltd.). The software for the experiment was written locally using MATLAB R2009b (The MathWorks Inc., Natick, MA, USA) and run on a personal computer. All stimuli were biphasic pulse trains of 300 ms duration and were based on a carrier rate of 5000 pps. To convey pitch information SAM was applied to a carrier pulse train using the equation 3 SAM ( t ) f ( t ) d sin( 2 f m t ) 2 where f(t) was the unmodulated pulse train at 5000 pps presented at threshold level and d was the depth of the modulation. The factor fm was the modulation frequency and it had a starting phase of 3 /2. The maxima and minima of the SAM corresponded to the subjects’ maximum comfortable (MC) level and the threshold (T) level of an equal amplitude unmodulated 5000 pps pulse train. To quantify the effect of DR on pitch ranking performance, the DR of each electrode was estimated during a standard clinical follow-up session. Then a WDR and a NDR electrode were selected based on the widest and the narrowest DR. These two electrodes were on a different location on the implant array for each subject. To evaluate if different locations on the implant array had an influence on performance, three further electrodes were selected. These were on the same location on the implant array for each subject. A basal electrode (electrode 4), a middle electrode (electrode 11) and an apical electrode (electrode 18) were tested. Monopolar stimulation was used in all cases. This involved current flow between an electrode on the implant array and two extracochlear returns in parallel: one on the case of the receiver stimulator and one on a ball electrode placed under the temporalis muscle. The experiment started with loudness balancing using the method described by Landsberger and McKay [13]. Initially new T and MC levels had to be defined for an equal amplitude pulse train at 5000 pps for each participating electrode. To equalize loudness levels across experimental conditions, presentation levels of all SAM stimuli were loudness balanced to a reference level of 262 Hz on each electrode. Prior to this the amplitude of 262 Hz was verified to be comfortable for each subject. First the baselines were balanced with each other, then the signals were

S783

loudness balanced with the corresponding baselines (see Table 2). Baseline Signal (26% higher)

Modulation Rates (Hz) 131 185 262 370 165 233 330 466

523 660

693 880

Table 2 shows all modulation rates presented in the pitch ranking task. The electrical signals investigated in this study were SAM electrical pulse trains delivered to a single electrode position. The modulation frequencies of the six baseline stimuli were between 131 Hz (musical note C3) and 693Hz (musical note F5). Each baseline stimulus had a corresponding signal stimulus with a modulation frequency which was four semitones (26%) higher. Pitch ranking was performed using a twointerval-two-alternative-forced-choice test (2I2AFC) “mixed-block” procedure described by Kong et al. [10]. All five electrodes were tested in the same block. For most of the subjects, 1500 responses were collected (50 per frequency per electrode). Subject (S) 3 did 900 trials (30 per frequency and electrode) and S9 and S10 only did 600 trials (20 per frequency and electrode). In each trial, the order of the signal and the baseline stimuli was randomized and the subjects were asked to choose the stimulus with the higher pitch. No feedback was given as to whether the responses were correct. Subjects were permitted a short training block where feedback was provided after each response to confirm that they had understood the task.

3

Results

Our experiment used forced choice procedures, therefore the results can be analyzed with the binomial probability distribution. We were initially interested in ascertaining whether the subjects had been merely guessing. For our 2I2AFC procedure, we consider the null hypothesis that the probability of success on each trial was 50%. If the resulting probability (p) is less than the criterion value = 0.05 that is generally accepted for statistical significance and we can conclude that it is unlikely that the null hypothesis is true. That indicates that the subjects were likely to have used some cue in the stimuli as a basis for their responses. Scores significantly above chance level mean that the subjects ranked the pitch correctly. Most subjects (S1, S2, S4 and S7-10) scored significantly above chance level on at least one of the tested electrodes. Performance significantly below chance level indicated that the subjects’ responses were pitch inverted and this was found in four of the ten subjects (S3, S5, S6 and S8). The individual results from the subjects tested in this study shows disparity in their pitch ranking performance. Averaged across all modulation frequencies the following results were obtained for each subject: S1 and S9 scored significantly above chance level on all tested electrodes while S5, S6 and S8 performed significantly below chance level on all electrodes. S4 performed significantly above chance level on the basal, the middle and the apical electrode while S2 performed significantly above chance level only on the WDR electrode. S3 per-



020 131

Percent Correct 185 262 370 523 Baseline Frequency (Hz)

693

020 131

100

100

185 262 370 523 Baseline Frequency (Hz)

693

080

080


693


693


693


693

060 040

020

020

020 131

020 131

020 131


693

Percent Correct

Percent Correct

100

100

080

080 060 040

040

040

040

693

060

060

060


Percent Correct

693

080

020

020

020 131

020 131

020

020


693

020 131


693


693

100

100

080

080

Percent Correct

Percent Correct

Percent Correct

080

080

060

060

060

060

040

040

040

Percent Correct

100

100

040

020

020

020

020


693

100

020 131

020 131


693


693

020 131

100

100

080

080

Percent Correct

Percent Correct

080

060

060

060

040

060

040

040

020

Percent Correct

100

080

040

020

020

020


693

020 131

080

080

020 131


693


693

Percent Correct

100

Percent Correct

100

060

060

040

040

020 020 131

020

020

Percent Correct


080

020 131

040

040

020

100

020 131

060

060

040

040

100

020 131

080

Percent Correct

020 131

080

060

060

020

Apical

100

Percent Correct

Percent Correct

080

Percent Correct

100

080

040

Middle

wide dynamic range

100

060

Basal

100

Percent Correct

narrow dynamic range

nificantly correlated with each other (p = 0.58). The baseline frequency and the percent correct score showed a tendency towards correlation but this was not found to be significant (p = 0.09). Figure 2 shows details about the performance for each subject in part two of the experiment.

Percent Correct

formed significantly below chance level on the basal, the middle and the NDR electrode and S10 scored significantly above chance level on the middle and on the apical electrode. S7 performed significantly above chance on the WDR electrode and significantly below chance on the basal, the apical and the NDR electrode. The Kolmogorov-Smirnov test showed that only the DR was normally distributed (p > 0.05). Nonparametric analysis with Spearman’s rho showed that DR was significantly positively correlated with score (p < 0.05) and location (p < 0.01). Details about the performance for each subject in part one of the experiment are shown in Figure 1.

S784

020


693

020 131

Figure 1: Performance based on dynamic range. It shows the mean percent correct score for each of the subjects on the electrodes with wide and narrow dynamic range. Dynamic range was positively correlated with the percent correct score. For the second part of the experiment the following results were acquired: Electrode location and score were not sig-


693

020 131

Figure 2: Performance based on location on the implant array. It shows the mean percent correct score for each subject on a basal, a middle and an apical electrode. Location and percent correct score were not significantly correlated with each other.

4

Conclusion

This study investigated the pitch ranking ability of CI listeners in response to SAM stimulation on different electrodes. We found that DR and score were significantly correlated with each other. We also analyzed the effect of three positions along the electrode array but we did not find a significant effect for electrodes from the locations that we tested.



4.1 Effect of DR on pitch ranking performance

[2]

The electrical DRs observed in CI subjects are usually small and they can vary across different electrodes. Our first hypothesis was that SAMs with wider modulation depths are easier to perceive than SAMs with narrower modulation depths. This hypothesis was confirmed by our study where subjects showed a positive correlation between DR and performance on pitch ranking. A previous study by McKay et al. investigated a similar effect. They decreased the DR of SAM stimuli and matched the decreased SAM to the pitch of an equal amplitude pulse train. They started with zero modulation depth which matched the pitch of the carrier rate. As they increased the modulation depth the matched rate came closer to the value of the modulation frequency [11]. NDR electrodes are linked with poor electrode discrimination, poor placepitch perception and poor speech recognition [14, 15].

[6]

4.3 Performance based on modulation frequency

[7]

The SAM frequencies that were tested in this study were below 1000 Hz. Such low frequencies are tonotopically located in the apical part of the cochlea. If place and rate pitch perception are linked the expectation would be that apical electrodes should perform best, followed by middle and basal electrodes. However in the present study no significant correlation was observed between location of stimulation on the implant array and pitch ranking score. The lack of a consistent electrode effect on rate discrimination performance is similar to the findings of Kong et al. [10] and Baumann and Nobbe and Zeng [8, 12]. These studies found no significant difference between basal and apical electrodes for pulse rate modulation discrimination. In contrast to these findings a study of Middlebrooks and Snyder showed that intracochlear neurons with a low characteristic frequency had a higher “limiting” rate than intracochlear neurons with a high characteristic frequency [16]. Additionally Macherey et al. showed that stimulation at an apical site of the cochlea yields better rate discrimination at high rates when asymmetric waveforms are used. This is explained the extension of the focus of excitation beyond the most apical electrode towards the apex of the cochlea [17]. The length of the electrodes used in our study together with standard waveforms could explain why there was no effect of location on performance. The development of processing strategies directed towards a more accurate consideration of intersubject and inter-electrode variability for pitch discrimination may improve the ability of CI users to perceive pitch. Improving CI performance based on individual subject and electrode performance might be the key to better understanding the mechanisms behind temporal pitch perception. Acknowledgement:This work was supported by the DFG Cluster of Excellence EXC 1077/1 "Hearing4all".

5 [1]

References Limb CJ, Rubinstein JT. Current research on music perception in cochlear implant users. Otolaryngol Clin North Am 2012 Feb;45(1):129-40.

[3]

[4]

[5]

[8] [9] [10] [11]

[12] [13]

[14]

[15]

[16]

[17]

S785

Gfeller K, Turner C, Mehr M, Woodworth G, Fearn R, Knutson JF, et al. Recognition of familiar melodies by adult cochlear implant recipients and normal-hearing adults. Cochlear Implants Int 2002 Mar;3(1):29-53. Penninger RT, Chien WW, Jiradejvong P, Boeke E, Carver CL, Limb CJ. Perception of pure tones and iterated rippled noise for normal hearing and cochlear implant users. Trends Amplif 2013 Mar;17(1):45-53. Sucher CM, McDermott HJ. Pitch ranking of complex tones by normally hearing subjects and cochlear implant users. Hear Res 2007 Aug;230(12):80-7. Greenwood DD. A cochlear frequency-position function for several species--29 years later. J Acoust Soc Am 1990 Jun;87(6):2592-605. Rubinstein JT. How cochlear implants encode speech. Curr Opin Otolaryngol Head Neck Surg 2004 Oct;12(5):444-8. Laneau J, Wouters J, Moonen M. Improved music perception with explicit pitch coding in cochlear implants. Audiol Neurootol 2006;11(1):38-52. Zeng FG. Temporal pitch in electric hearing. Hear Res 2002 Dec;174(1-2):101-6. Shannon RV. Multichannel electrical stimulation of the auditory nerve in man. I. Basic psychophysics. Hear Res 1983 Aug;11(2):157-89. Kong YY, Deeks JM, Axon PR, Carlyon RP. Limits of temporal pitch in cochlear implants. J Acoust Soc Am 2009 Mar;125(3):1649-57. McKay CM, McDermott HJ, Clark GM. Pitch matching of amplitude-modulated current pulse trains by cochlear implantees: the effect of modulation depth. J Acoust Soc Am 1995 Mar;97(3):1777-85. Baumann U, Nobbe A. Pulse rate discrimination with deeply inserted electrode arrays. Hear Res 2004 Oct;196(1-2):49-57. Landsberger DM, McKay CM. Perceptual differences between low and high rates of stimulation on single electrodes for cochlear implantees. J Acoust Soc Am 2005 Jan;117(1):319-27. Pfingst BE, Holloway LA, Zwolan TA, Collins LM. Effects of stimulus level on electrode-place discrimination in human subjects with cochlear implants. Hear Res 1999 Aug;134(1-2):105-15. Donaldson GS, Nelson DA. Place-pitch sensitivity and its relation to consonant recognition by cochlear implant listeners using the MPEAK and SPEAK speech processing strategies. J Acoust Soc Am 2000 Mar;107(3):1645-58. Middlebrooks JC, Snyder RL. Selective electrical stimulation of the auditory nerve activates a pathway specialized for high temporal acuity. J Neurosci 2010 Feb 3;30(5):1937-46. Macherey O, Deeks JM, Carlyon RP. Extending the limits of place and temporal pitch perception in cochlear implant users. J Assoc Res Otolaryngol 2011 Apr;12(2):233-51.



S786

Development of a Model of the Electrically Stimulated Auditory Nerve 1

W. Nogueira1, W. Würfel1, A. Büchner1 Hearing4all, Hannover Medical School, Hannover, Germany, [email protected]

Structure:

1. Introduction / 2. Methods / 3. Results / 4. Conclusion

Abstract Cochlear Implants (CIs) are used to restore the sense of hearing in people with profound hearing loss. Some CI users can communicate over the phone and even understand speech with some background noise, however some other CI users do not obtain the same benefit from the device. One possible reason that might explain this variability, at least partially, is the individual differences in the interface created between the electrodes and the auditory nerve. For example the exact position of the electrodes in cochlea and the amount of functional auditory neurons might differ significantly for differnet CI users. In order to understand the electrode-nerve interface, a model of the auditory nerve activity stimulated by a cochlear implant has been developed. The model is based on existing models found in the literature. This paper presents the basic components of the model: Voltage distribution in the cochlea based on finite element method and auditory nerve model based on a multi-compartment Hodgkin Huxley equations. The model has been used to simulate 1) the extracellular voltage applied to the auditory nerve fibers and 2) the auditory nerve activity, produced by simple stimuli (single pulses) and complex stimuli (speech sounds) for different electrode configurations.

1

Introduction

Motivation Cochlear implants (CIs) have been successful in restoring hearing to a reasonable degree, provided that the auditory nerve is functional [1] [2]. In a cochlear implant current is directly applied to the auditory nerve stimulating an array of several electrodes in the cochlea. When an electrode is activated, the sensory nerves close to that electrode will start to send action potentials to the brain. One limitation in cochlear implants is the so-called spread of excitation, which produces excitation of auditory nerves beyond those close to a particular electrode. Spread of excitation causes a lack of specificity in the stimulation and has a negative impact in speech intelligibility and music perception for example. In this paper, a three dimensional finite element method (FEM) model coupled to a multicompartment auditory nerve model is presented. The goal is to simulate and assess the spread of excitation casued by cochlear implant electrode stimulation. The model is developed based on existing models in the literature ([2] [3] [4] and [8]) and it allows to be adapted to the cochlear dimensions of each individual based on clinical computer tomography (CT) data. The auditory nerve model can also be adjusted to reproduce the individual eCAPs of each CI users. The paper is organized as follows: Section 2 describes the construction of the geometry, the finite element method and the auditory nerve models. Section 3 shows the experimental results and simulations. Finally sections 4 presents the conclusions based on the results.

2

Methods

Geometry A general cochlea geometry has been constructed based on a histological dataset of a single human cochlea (Figure 1a). The shapes of the compartments (Scala Tympani, Scala Vestibuli, Scala Media, Reissener Membrane and Basilar Membrane) were approximated by polygons with a relative low number of points as done by Rattay et. al 2001. Next, every 30 around the vertical axis, a new plane containing all the compartments was repeated. The new plane was extruded and joined to the previous plane. In total 2.5 turns of the cochlea were modelled. Figure 1b) and c) present as an example the geometry and the triangulation for one of the cochleas created. The resulting geometry can be easily modified and adaptated to clinical CT data of each individual as presented in [5]. The electrode array inserted in the Scala Tympani was modelled inserting 16 ball electrodes separated by 1.5 mm. Different electrode positions were simulated. For example in a modiolar insertion, electrodes are placed close to the modiolus; In a midscalar insertion, electrodes are placed in the center of the scala tympani; and, in a lateral insertion, electrodes are placed close to the lateral wall of the cochlea. The auditory nerve was modelled drawing a single nerve fiber with 10 nodes (Figure 3a) all along the spiral ganglion over the two and half turns of the cochlea. In total 3000 spiral ganglions were modelled



( )[1

=

]

S787

( ) ,

=

, ,

with ( ) and ( ) being the voltage dependent opening and closing rates of the ion channels. For a more detailed description of the Equations see [4].

(a)

The HH model was adapted to human physiological data. The morphometric parameters of the nerve fibre model were based on Type I peripheral auditory nerve fibre as presented in [4].

(c)

(b)

Figure 1: a) Cochlea in saggital cross-section. 1: Basilar membrane; 2: Modiolus; 3: Spiral ligament; 4: Scala vestibule; 5: Scala tympani; 6: Lamina spiralis. This Picture has been extracted from Würfel, 2011. b) 3D geometrical model of the cochlea, c) Mesh after triangulation Finite Element Model (FEM) Finite element method was used to model the electric field in the spiral ganglion. The geometry was classified into domains (bone, nerve tissue, perilymph, endolymph, reissner membrane and basilar membrane), and each domain was assigned a material property in the form of resistivity. The values for the resistivities can be found in [6].

x 10

of the k-th compartement is described by the cable Equation: (

)

,

=

,

+

+

+

+

,

,

2 + 2 +

+

,

2 + 2

,

,

where is the axoplasmatic resistance to the neighbours at the k-th compartment and , gives the membrane capacitance.

Nerve Fibers position

-3

8

The nerve fibre cable model The human nerve fibre model is based on the cable model of Rattay. A type I auditory nerve fibre was modelled consisting of a drendrite, soma an axon. The dendrite and axon were divided into k cylindrical compartments. The axon had a diameter two times larger than the diameter of the dendrites. The soma was assumed to be spherical. Ranvier nodes were unmyelinated active axolemae with only the axon based on the general sensory nerve fiber at the centre model. The change in membrane potential

7 6

The myelinated internodes were simple double cable structures as modelled by [4].

5 y

4 3 2 1 0 -1 -4

(a)

-3

-2

(b)

-1

0

x

1

2

4

3

-3

x 10

(c)

Figure 3: a) Model of a singleauditory nerve with 10 nodes , b) 3D cochlea model with nerves inside, c) Node positions inside the volume for the XY plane Auditory Nerve Model The basic unit of the auditory neve model is the human ranvier node model which is based on the original Hodgkin Huxley (HH) model [4], [2]. The HH equations model the ionic membrane current as follows: =

where , conductances.

(

)+

,

are the

(

)+

,

(

),

, and leackage ion

The dynamics of the opening probabilities (m, h and n, respectively) of the ion channels are described by:

Individual model of degenerate nerve fiber Retrograde neural degeneration, in which the dendrites are lost but the somas and axons survive, occurs in persons with profound hearing loss. The degree of retrograde concerns only Type I auditory nerve fibers [4]. A degenerate version of the auditory nerve model similar to the one presented in [4] was used to simulate the effect of neural degeneration. This was modelled by removing the first four nodal and internodal sections of the auditory nerve fiber model. More severe neural degeneration could be modelled removing all dendritic nodes and internodes up to the soma. eCAP Model The electrode nerve interface of a CI user can be characterized using the recording capabilities of their device. Current CIs are able to record the small physiological potentials generated by stimulated auditory nerve fibers. These objective measurements are known as evoked compound action potentials (eCAPs). In this paper we present a model of the auditory nerve fiber activity which is able to simulate eCAP responses for different electrode positions, stimulation modes and degrees of auditory



nerve survival, and therefore allowing for a better understanding of the electrode-nerve interface.

S788

for different levels for a modiolar electrode insertion and a pulse width of 25 µs are presented in Figure 6. eCAP

where N is set to 10 nodes and J set to 3000 nerve fibers The eCAP can be recorded at the electrode positions by current commercial cochlear implant systems. To model this, we can assume that the eCAP is generated at the center of the geometry. Next, the FEM model presented before is used to simulate/recorded the voltage at the electrode positions. The coupling between the center of the geometry and the sensing electrode k is modelled as the transfer function . The final eCAP when electrode i is stimulated at the recording electrode k is then obtained as: ,

( )

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0

0.5

1

1.5 time

2

2.5

3 -3

x 10

Figure 6: Simulation of eCAP responses for different levels from 10 µA until 500 µA in steps of 20 µA. Each color represents a different current level. A biphasic pulse with a phase pulse width of 25 µs was used.

.

Simulation of Auditory Nerve Activity for Speech Sounds The simulation of the auditory nerve activity for complex sounds (such as speech) was accomplished by processing audio signals using a commercial strategy (Figure 4). The coding strategy computes the electrical currents applied to the cochlear implant electrodes. The audio signal is picked by the microphone, digitized and transformed into the frequency domain using a short time Fourier transform with overlap and add. Next, envelopes are extracted summing up the FFT bands that correspond to each critical band. Finally the bands are compressed and converted into current values adapting them to the threshold and most comfortable values of each cochlear implant user.

3.2 Simulation of an Amplitude Growth Function An amplitude growth function measure has been modelled measuring the voltage difference between the local maxima and local minima of the eCAP responses presented in Figure 6 for each stimulation level. The resulting amplitude growth in the neural response for increasing current levels is presented in Figure 7. Amplitude Growth Function 1 Modiolar Lateral Midscalar

0.9 0.8 0.7 Voltage [V]

( )=

1

Normalized Voltage [V]

The eCAP can be modelled summing the Voltage along each nerve j and node n ( , ( )) for each time epoch t as follows: eCAP (t) = , ( ),

0.6 0.5 0.4 0.3 0.2 0.1 0

Figure 4: Signal processing block diagram of a cochlear implant sound coding strategy The frame sequence of current amplitudes (eletrodogram) delivered by the speech processor is then used to compute the extracellular voltage using the FEM presented in the previous section.

3

Experiments

3.1 Simulation of the evoked action Potential The evoked action potential has been simulated presenting a biphasic pulse and measuring the overall neural response on all nodes (Equation 1) of. The eCAP responses

0

50

100

150

200 250 300 Current Level [uA]

350

400

450

500

Figure 7: Amplitude growth function for three different electrode positions (modiolar, midscalar and lateral) 3.3 Auditory Nerve Response Patterns to Speech Stimuli An audio signal containing the phoneme /a/ (Figure 8a) has been processed using a commercial cochlear implant strategy. The electrodogram obtained is presented in Figure 7b. The voltage distribution obtained at node 6 is presented in Figure 7c. It can be observed that the current spreads all along the cochlea. Figure 2d presents the auditory nerve activity obtained at node 6 along the spiral ganglion.



Wave Form /a/

Electrodogram

0.8

800

2 0.6

700

4 0.4

600

6 Electrodes

Amplitude

0.2

0

500 8 400 10 300

-0.2

12

-0.4

14

200 100 -0.6

16 0

0.02

0.04

0.06

0.08

0.1 Time

0.12

0.14

0.16

0.18

0.2

0.02

0.04

0.06

(a)

0.08 0.1 Time [s]

0.12

0.14

0.16

0

0.18

(b) 0.9

0.8 0.6

10

0.8

5

0.4

0.7

0.2

0.6

distance [mm]

Nerve Fibers

5

0 -0.2

10

0.5 0.4 15

15

0.3

-0.4

0.2

-0.6 20

20

0.1

-0.8 0

0.05

0.1 Time

-1

0.15

0.05

0

Figure 8: a) Audio waveform for the speech token /a/, b) Electrodogram, c) Extracellular voltage distribution along node 6 through the spiral ganglion, d) Auditory nerve response. Figures c) and d) have been normalized. The effect of neural degeneration on speech stimuli has been simulated for the same phoneme /a/. Figure 9a presents the extracellular voltage at node 6 when the dendrite nodes 10, 9, 8 and 7 are not present. Figure 9b presents the auditory nerve activity. Auditory Nerve Activity Node 6

Voltage distribution Node 6

1

0

1

0.9

0.8

[3] Hanekom, T. (2001). "Thesis - Cochlea modelling," in Faculty of Engineering, built Environment and Information Technology (University of Pretoria, Pretoria). [4] J. E. Smit, T. Hanekom and J.J. Hanekom (2008), Predicting action potential characteristics of human auditory nerve fibers through modification of the Hodgkin-Huxley equations, South African Journal of Science 104. July/August, 2008.

0.8

0.6 4

4

0.4 0.2

6

0 8

-0.2 -0.4

10

distance [mm]

Nerve Fibers

[2] Rattay, F., Leao, R. N., and Felix, H. (2001). A model of the electrically excited human cochlear neuron. II. Influence of the three-dimensional cochlear structure on neural excitability, Hear. Res. 153, 64-79.

2

2

0.7 0.6

6

0.5 8

0.4 0.3

10

-0.6 12

-0.8 0

0.05

0.1 Time

0.15

(a)

-1

0.2 12

0.1 0

0.05

0.1 Time

0.15

0

(b)

Figure 9: a) Extracellular voltage along node 6 through the spiral ganglion. b) Auditory nerve response. From Figures 9a and 9b it can be observed that neural degeneration produces a broader spread of excitation pattern compared to the spread of excitation observed in Figures 8c and 8d, probably explaining the poorer sound perception obtained by cochlear implants users having a longer period of deafness.

4

[1] Wilson, B. S., and Dorman, M. F. (2006). "Cochlear implants: A remarkable past and a brilliant future," in 9th International Conference on Cochlear Implants and Related Sciences (Elsevier Science Bv, Vienna, AUSTRIA), pp. 3-21.

0

0.15

0.1 Time

(d)

(c)

0

References

1

0

1

activity for different electrode positions and cochlear geometries. Examples of auditory nerve patterns generated by simple stimuli (single biphasic pulses) and complex stimuli (speech tokens) have been presented. The model can be used to understand better the auditory nerve activity of each CI user given that the electrode position and cochlear shape are known. Future investigations should include the validation of the model, comparing the simulations with real measurements in CI users.

4

Auditory Nerve Activity Node 6

Voltage distribution Node 6 0

S789

Conclusions

[5] W. Nogueira, W. Würfel, A. Büchner (2014), Development of a Model of the Electrically Stimulated Cochlea, Lecture Notes in Applied and Computational Mechanics, Springer, 2014 (Submitted). [6] J.J Briaire and J.H. Frijns, Field patterns in a 3D tapered spiral model of the electrically stimulated cochlea, Hear Res., vol. 148, pp. 18-30, Oct. 2000. [8] N. Nicoletti, Ch. Wirtz and W. Hemmert (2013), Modeling Sound Localization with Cochlear Implants, ISBN 978-3-642-37761-7, Springer, June. [9] W. Würfel (2011) Vergleichende Auswertung histologisher Schliffpräparation, fpVCT- und mCT –Bildgebung hinsichtlich der Eignung zur Darstellung und 3Drekonstruktion der Anatomie von Mittel- und Innenhor, Dissertation, Medical School Hannover.

The model presented in this paper is based on existing models. The geometrical model represents a typical cochlear shape that can be adapted to clinical CT data to the individual characteristics of each CI user. The electrical model has been coupled to a multi-compartment auditory nerve model which allows simulating the auditor nerve



S790

Effect of pulse rate on loudness growth functions in cochlear implant users Gunnar Geißler1, Matthias Hey2, Britta Böhnke2, Stefan Fredelake3, Caroline Frohne-Büchner3, Andreas Buechner1, Joachim Müller-Deile2 1 Hannover Medical School, Hanover, Germany, [email protected] 2 University Medical Center Schleswig-Holstein, Kiel, Germany 3 Advanced Bionics GmbH – European Research Center, Hanover, Germany Structure:

1. Introduction / 2. Methods / 3. Results / 4. Conclusion / 5. References

Abstract Aim of this evaluation was the comparison of loudness growth functions measured at two different pulse widths. The loudness growth functions were measured via a loudness scaling procedure with direct electrical stimulation. The stimulus was a pulse train presented on a single electrode with a defined amplitude and pulse width. 13 cochlear implant (CI) users with Advanced Bionics implants participated and for each of them the loudness scaling procedure was conducted with a constant pulse width of 33.2 µs and an individually determined minimal pulse width at which the upper dynamic range could be stimulated without reaching the compliance limit of the implant; the actual range was 18 µs to 40.4 µs. Our results indicate that for low loudness levels (threshold to about soft) the pulse width is significantly affecting the perceived loudness. When doubling the pulse width from 18 to 36 µs, the charge had to be on average 1.98 dB higher for “very soft” loudness levels. For higher loudness levels no influence of the pulse width was observed.

1

Introduction

The categorical loudness scaling is an essential tool in the diagnostics of hearing disorders and instrumentally for the fitting and verification of hearing aids. It is used to determine the dynamic range between “Inaudible” and “Too loud” with narrow band noise in an adaptive procedure in normal hearing as well as hearing impaired. Loudness scaling could also be a useful tool for setting of thresholds and MCL (Most Comfortable Level) in cochlear implant (CI) users. A CI is a surgically implanted prostheses allowing patients with severe to profound deafness to hear by electrical stimulation of the auditory nerve. In the speech processor program, the stimulation levels for threshold and MCL have to be set individually for each electrode contact. When using loudness scaling for this fitting, an important question is whether the scaling needs to be conducted at exactly the same pulse rate as used later in the speech processor program. The pulse rate may change during the fitting process. If the pulse rate has a strong effect on loudness scaling, the loudness scaling may need to be repeated when changing pulse rate. Therefore, the objective of this evaluation was to assess the influence of pulsewidth on the loudness growth function. In the adaptive procedure used here [1], single electrodes are stimulated with a defined amplitude and pulse width.

2

Methods

2.1

Stimuli

separated by 500 ms silence. Pulse width and amplitude were the parameters defined during the experiment. The stimulus was constructed in the same way as in a 15 channel sequential pulsatile strategy without any gaps between the stimuli, mimicking the Advanced Bionics HiRes 120 strategy [2]. Therefore, a change in pulse width was also resulting in a change of the pulse rate by the following equation, 1 𝑟𝑎𝑡𝑒 [𝑝𝑝𝑠] = 𝑃𝑊 [µμ𝑠] ∙ 2 ∙ 15 where PW is the pulse width. The subject had to judge the loudness on a scale ranging from “Inaudible” to “Too loud” with 11 stages. For further processing, these estimates were converted to an internal representation from 0 CatU (Categorial Units) to 50 CatU. The loudness scaling was performed for each subject with two different pulse widths (further explanation see 2.3 Conditions). The measurement equipment was realized using a research platform provided by Advanced Bionics. The stimuli to be presented were calculated in Matlab and the ratings were assessed via a graphical user interface on a touch pad.

2.2

Subjects

In this study, 13 postlingually deafened subjects participated. All were unilaterally or bilaterally implanted with C2- or HiRes90k implants. In bilateral subjects, only one ear – the leading one – was measured. The mean age was 55 years with a range from 19.2 to 75.4 years. The mean time of implant experience was 7.2 years ranging from 0.6 to 12.9 years.

For determination of the loudness growth function, single electrodes were stimulated with two 500 ms pulse trains,



Conditions

Each loudness scaling started with a stimulation level just below the MCL of the clinically used program. To allow comparison of stimulation levels across various pulse widths, the level was expressed in charge, measured in clinical units (CU). The relationship between pulse width, stimulation amplitude and charge is 𝑃𝑊 [µμ𝑠] 𝑄[𝐶𝑈] = 𝐼[µμ𝐴] ∙ 77.812 with Q as charge, 𝐼 as current and 𝑃𝑊 as pulse width. The charge is alternatingly increased and decreased until the boundaries of the dynamic range were reached. The upper boundary of the dynamic range was kept at the value where the subject rated “Too loud” for first time. To determine the lower boundary, the stimulation level was decreased until the stimulus was no longer perceived and then increased until it was rated as “Very soft” for the first time. After the determination of the dynamic range, the actual measurements of the loudness perception started following a pseudorandomized procedure. For clarification, an example of such a procedure is shown in figure 1, where in the 16th trial the upper boundary of the dynamic range was reached and in the 21st trial the lower boundary was determined.

The same measurements with the same two pulse widths were repeated during a second appointment approximately 6 weeks later.

2.4

Data analysis

The single estimates in CatU of one run were plotted against the current in µA or the charge in CU. A model function, consisting of two straight lines, was fitted with a least squares method (see figure 2). The interception point between the two straight lines was allowed to be in the range from 15 to 35 Categorical Units. 50 45 40

Loudness [Categorical Units]

2.3

S791

35 30 25 20 15 10 5

1000

0

900

I ( A)

600 500 35

400

25

30

40

45

45

45

45 45

50 30

45

40

30

20 15

300

10

10

5

10 200

5 5

100

5 0

0

5

10

0

0 15

0

0

0

0

20 Index

5

0

5

25

200

300

400

500 I [ A]

600

700

800

900

1000

10

30 15

100

Figure 2 Individual estimates of the example displayed in Fig. 1 with the model approximation (black solid line). The black dashed lines represent T-level and MCL from the clinical program, the red line indicates the compliance limit of the current source.

800 700

0

30

0

35

40

Figure 1 Example of one loudness scaling run; the consecutive stimuli are identified by the index, the respective stimulation level is indicated by the current on the y-axis and the subject`s loudness rating in CatU is shown as the number next to each circle. The red line shows the compliance limit of the current source. Two different pulse widths were investigated: first, a constant pulse width of 33.2 µs for all subjects and second, the minimal possible pulse width for each subject. Here, the measurements started with 18 µs. When the compliance limit of the current sources was reached before the loudness was judged as “Too loud”, the pulse width was increased and a new measurement started. This was repeated until the subject rated the loudness as “Too loud”, ensuring that the current source was operating within its compliance voltage. The measurements were performed on the electrodes 2 (more apical), 6, 10 and 14 (more basal) with the minimal pulse width. With the constant pulse width of 33.2 µs the measurements were performed on all active electrodes.

For further analysis, the required stimulation levels for each of the loudness steps were derived from the model function. For example in figure 2 a current of 145 µA would be necessary to obtain a loudness of 5 CU. For comparison of stimulation levels between electrodes, they were plotted across all electrodes, resulting in isoloudness contours (figure 3).

Figure 3 Example of iso-loudness curves. Dashed black lines represent MCL and threshold from the clinical program.



3

Results

The resulting pulse widths of the measurement with the individually determined minimal pulse width were in the range of 18 µs up to 40.4 µs. In one subject with very high MCL the resulting minimal pulse width was 70.7 µs. Thereby the constant pulse width of 33.2 µs was not sufficient to evoke loud sensations and a reasonable fitting of the model function was not possible. So this subject was not considered in the further analysis. To obtain insight in the influence of the pulse width on loudness, the required amount of charge for a given loudness obtained with the minimal pulse width was divided by the corresponding value of the measurement with the reference pulse width of 33.2 µs. This was calculated for each subject individually and all four electrodes were averaged. This manifests in the following equation 𝑄

.

(𝐿) =

1 4

, ,

,

𝑄(𝐸𝑙. = 𝑛, 𝐿[𝐶𝑎𝑡𝑈], 𝑃𝑊 = 𝑥 µμ𝑠)[𝐶𝑈] 𝑄(𝐸𝑙. = 𝑛, 𝐿[𝐶𝑎𝑡𝑈], 𝑃𝑊 = 33.2µμ𝑠)[𝐶𝑈]

where 𝑄 is the charge, 𝐸𝑙. is the electrode, 𝐿 is the loud-

S792

ness in categorical units and 𝑃𝑊 is the pulse width in seconds. The resulting 𝑄 . (𝐿) is the relative amount of charge with reference to the 33.2 µs condition needed to obtain equal loudness with the individually determined minimal pulse width. In figure 4, these values were plotted for all subjects at each appointment for the loudness levels of 5, 15, 25 and 35 CatU. For the loudness of 5 CatU (“Very soft”) a significant moderate correlation was found between the pulse width and required charge (Pearson, correlation coefficient r=0.442, p-value=0.045). When a wider pulse width was used, more charge was needed to obtain the same loudness. For a loudness of 15 CatU (“Soft”) only a slight, not significant correlation could be noticed (r=0.383, p=0.086). For louder values of 25 CatU (“Medium”) and 35 CatU (“Loud”) no correlation was observed (r=0.078, p=0.737 and r=0.016, p=0.944). The regression line for the loudness of 5 CatU had a slope of 0.0116 1/µs. That means, if a pulse width of 18 µs was doubled, the charge had to be increased by 1.98 dB on average to maintain the same loudness.

Figure 4 Normalised charge for loudness of 5 CU (top left), 15 CU (top right), 25 CU (bottom left) and 35 CU (bottom right). Plotted in red, a least squares fit regession line and in the upper right of each panel the p-value is shown.



In the given example in fig. 3 the profiles of the isoloudness curves, measured with the constant pulse width of 33.2 µs, are reasonable well aligned with MCL and threshold. This was also the case for 9 out of 13 subjects. For the other four subjects the MCL were for one or more electrodes below the 15 CatU iso-loudness courves. This is indicating, that the levels in the clinical program were not set high enough.

4

Conclusions

In the lower part of the dynamic range, i. e. between threshold and soft, an increase in pulse width (and thereby a decrease in pulse rate) requires an increase of charge to obtain equal loudness. In the upper part of the dynamic range no influence of the pulse width on the required charge was found. Several groups investigated the influence of stimulation rate on threshold and MCL without scaling in the whole dynamic range [3, 4, 5]. In these publications, a much larger range of pulse rates has been investigated than in our experiment. Green et al. [3] used an ascending method of adjustment for setting the threshold and the MCL with three repetitions. Kreft et al. [4] used also an ascending method of adjustment for setting the MCL, but a three-interval forced choice procedure for the thresholds. In [5], the subjects set levels of threshold and the MCL themselves. All of them found effects of the pulse rate on the threshold level as well as on the MCL. Even though significantly different methods were used they found similar results for the effect of pulse rate at threshold (1.07 up to 2.4 dB/doubling of PW). Similar to our results the effect at MCL was smaller than at threshold. However, it was still significant (0.54 up to 1.2 dB/doubling of PW). So the slope of 1.98 dB/doubling we have found for effects of pulse rate on threshold is in well accordance with these publications. Methodological differences between the studies may explain the remaining differences in outcome. In our present study some subjects showed a high variability of loudness ratings which need further investigation. They may be caused by adaptation effects to a large extend. When the loudness of a stimulus has to be estimated right after a very loud stimulus, it will often be rated as softer than in succession to a very soft stimulus. The effect will be attenuated by the use of the model function, but not completely cancelled. A second issue is the limited range of pulse widths tested. The overall range is smaller than in the cited experiements and for each subject only two pulse widths were examined. As these measurements and observations were only part of a larger study in which the influence of pulse rate on loudness was only a side aspect, the experiment was not particulary set on the influence of the pulse rate and/or pulse width. So it would be worthwhile to repeat the loudness scalings with a significantly increased range of pulsewidths and several pulse widths per subject. Still, one outcome of this study is that a loudness scaling should be done with the pulsrate that is used in the speech processor program.

5

S793

References

[1] Brand T., Hohmann V. (2002) An adaptive procedure for categorical loudness scaling. J. Acoust. Soc. Am. 112(4): 1597-1604. [2] Nogueira W., Litvak L., Edler B., Ostermann J., Büchner A. (2009) Signal Processing Strategies for Cochlear Implants Using Current Steering. EURASIP Journal on Advances in Signal Processing 2009:531213 [3] Green T., Faulkner A., Rosen S. (2012) Variations in Carrier Pulse Rate and the Perception of Amplitude Modulation in Cochlear Implant Users. Ear Hearing 33: 221230 [4] Kreft H. A., Donaldson G. S., Nelson D. A. (2004) Effects of pulse rate in threshold and dynamic range in Clarion cochlear-implant users. J. Acoust. Soc. Am. 115(5): 1885-1888 [5] Zhou N., Xu L., Pfingst B. E. (2012) Characteristics of detection thresholds and maximum comfortable loudness levels as a function of pulse rate in human cochlear implant users. Hear. Res. 284: 25-32



S794

Identification of a plant model for the control of arterial blood pressure during normothermic ex-vivo kidney perfusion M. Gransow, S. Koch, F. Tetschke, C. Thiele, H. Malberg, Institute of Biomedical Engineering, TU Dresden, Dresden, Germany, [email protected]

Introduction Normothermic ex-vivo perfusion of isolated organs is an innovative technique to maintain an organs viability and therefore extend storage times without lethal organ failure. For automating the process of ex-vivo perfusion, the control of renal arterial pressure is essential. Therefore a model is necessary that combines simplicity with an adequate simulation of the real plants behavior. In this work we derived such a plant model for the control of the arterial blood pressure during ex-vivo kidney perfusion.

Methods We modeled a perfusion circuit as a combination of generic electrical pathway analogies. Every line section is represented by a three-element RCL four terminal network. The plant model could be simplified to a four-element network, where only the kidneys parameters are unknown. The model was implemented to SIMULINK in state space representation. Additionally real ex-vivo perfusions of 8 porcine kidneys were performed. We applied steps of the blood pressure must values of one to 20 mmHg to both, the real and the simulated process by using an I-controller. Simulation parameters were adjusted to fit the real data and to evaluate the model.

Results We determined individual kidney parameter sets for 124 analyzable pressure steps, consisting of Resistance R, Compliance C and Inertance L. Under consideration of the renal arterial blood pressure and flow, the simulation could represent (𝑡) − 𝑦 (𝑡) )²⁄ ∑ 𝑦² (𝑡) ≥ 0.95. We observed a time variant behaviour of the real data with a 𝐹𝑖𝑡 = 1 − ∑ (𝑦 the kidney describing parameters over the duration of each kidney perfusion.

Conclusion Due to the observed time variant behaviour of process describing parameters, a control of the arterial blood pressure in a kidney perfusion system needs to be designed robustly or even parameteradaptive. By comparing simulated step responses to real measured data, we proved that our hydraulic plant model is able to represent the real perfusion system adequately in order to design blood pressure controllers.



S795

Online Gait Phase Detection with Automatic Adaption to Gait Velocity Changes Using Accelerometers and Gyroscopes Thomas Seel1 , Lucian Landgraf1 , Víctor Cermeño Escobar1 , Thomas Schauer1 Control Systems Group, TU Berlin, Germany, [email protected]

1

Abstract We consider real-time detection of gait events from inertial measurement data. Unlike previous approaches, we avoid the use of magnetometers and do not restrict the mounting of the sensor to certain locations or orientations. The proposed algorithm detects the toe-off and initial contact as well as the beginning and end of rest periods. We discuss suitable signals and criteria for the detection of these events and add algorithms for automatic adaption to changes in gait velocity. Gait experiments with healthy subjects and stroke patients are performed to evaluate reliability and robustness. The method is found to be suitable for a large variety of terrains and walking speeds.

1 Introduction The human gait can be divided into phases that are characterized by certain foot motions or levels of ground contact. Besides the classic optical motion analysis systems, more ambulatory sensor systems have been proposed, involving for example pressure soles and inertial measurement units (IMUs). These technologies are particularly helpful in active prostheses or rehabilitation systems, when a robotic force or the intensity of a functional electrical stimulation (FES) shall be manipulated synchronously to the gait. In such applications it is crucial that gait events, e.g. toe-off or initial contact, are detected in real-time. However, this task is more challenging than offline analysis, since event criteria may only use present and past data.

1.1 Brief Review of IMU-based Real-Time Gait Phase Detection Algorithms In previous literature, the definitions of gait phases and events vary as well as the employed sensor technologies and methods. A good review of ambulatory gait phase detection can be found in [4]. Since our goal is to provide a gait phase detection for synchronization of FES to gait in the Adaptive Peroneal Stimulator introduced in [7], we focus our review on real-time methods that employ only accelerometers or IMUs. In previous studies, accelerometers or IMUs were placed on the trunk [2], on the thigh [9], and on the shank [10], [1], often with promising results. In a feedback controlled peroneal stimulator, however, foot/shoe-mounted sensor units allow additional assessment of the stimulation outcome, e.g. foot-toground angles [8]. Rueterbories et al. suggest the use of a foot-mounted accelerometer to detect four gait phases from inflection points and curve extrema of the acceleration vector norm. However, the experimental evaluation was limited to constant gait velocity. In the following, we propose a threshold-based method that uses foot accelerations and angular rates and adapts itself to the subjects gait velocity.

Figure 1 Gait cycle modeled by finite state automaton. Transitions (gait events) are detected from inertial data.

Figure 2 Mean (solid) and standard deviation (bands) of characteristic signal trajectories in human gait.



S796

2 Gait Phase Detection Algorithm The 6d IMU is attached to the midfoot (or respective part of the shoe) in arbitrary position and orientation. The IMU provides real-time measurements of the foot acceleration a(t) 2 R3 and angular rate g(t) 2 R3 at a sample rate of fs = 100 Hz via a wireless link. Figure 1 shows the setup and the four considered gait phases. The upper plot of Figure 2 shows typical courses of the vector norms ||a(t)||2 , ||g(t)||2 2 R 0 .

2.1 Detection of Foot Rest For detection of foot-flat phases, we use the standard approach of defining rest bands with upper bounds arest , grest 2 R>0 . When both signals remain in their bands (i.e. ||g(t)||2 < grest and | 9.81 + ||a(t)||2 | < arest ) for at least nff 2 N>0 consecutive samples, the automaton switches to foot-flat phase regardless of the current state. On the contrary, it transitions from foot-flat to pre-swing, when the following holds for a single sample: |

9.81 + ||a(t)||2 | > ↵ arest _ ||g(t)||2 > ↵ grest , (1)

where ↵ 2 R>1 is a hysteresis factor that prevents chattering between foot-flat and pre-swing. Sensitivity of heeloff detection can be influenced by choosing ↵ appropriately. Furthermore, it is expected that choosing larger numbers for nff and small number for arest , grest leads to delayed detection of foot-flat, while doing the opposite increases the chances of false foot-flat detection during swing phase periods with low acceleration and rotation (cf. Figure 2).

2.2 Detection of Toe-Off The lower plot of Fig. 2 shows typical courses of the pitch rate pitch (t) 2 R and toe velocity ||v toe (t)||2 2 R 0 . While the pitch rate exhibits a sign change at toe-off, the toe velocity suddenly rises. When proper foot-flat detection is established, these signals can be computed as follows: Let t0 be the time instant at the end of the last foot-flat phase and restart an orientation strap-down integration [5] at each heel-off to obtain the rotation matrix R(t) from local sensor coordinates at moment t to local sensor coordinates1 at moment t0 . We then define Rt g(⌧ )d⌧ t , (2) R t0 pitch (t) := g(t) · || t0 g(⌧ )d⌧ ||2 Z t toe R(⌧ )atoe (⌧ ) a(t0 ) d⌧, (3) v (t) := t0

atoe (t) := a(t) + g(t) ⇥ (g(t) ⇥ o) + g(t) ˙ ⇥ o,

(4)

where the central dot denotes the scalar product and o 2 R3 is the approximate position of the forefoot in the local sensor coordinates, which can be estimated from 1 i.e. the local coordinates at t serve as a fixed reference coordinate 0 system into which local measurement vectors of following time instants t can be transformed via multiplication by R(t)

Figure 3 Mean (solid) and standard deviation (bands) of sensor and toe velocity during pre-swing and swing. The toe velocity, which can be calculated when the approximate local coordinates of the forefoot are known, allows a more precise detection of toe-off. pre-swing measurement data using the technique proposed in [6] or in [3]. Figure 3 shows the benefits of using this approach versus using o = (0, 0, 0)T , i.e. atoe (t) ⌘ a(t). While the IMU already moves during pre-swing, the forefoot starts to move just at toe-off. We will leave it as an option of the toe-off detection for the remainder of this article. Regarding the pitch rate, please note that (2) yields the rotation around the average axis of rotation since heel-off. Therefore, no prior knowledge about the sensor-to-foot orientation is required. Toe-off is detected when the current state is pre-swing phase and one of the following conditions is fulfilled: pitch (t) falls below a threshold 0 after having exceeded2 2 0 or ||v toe (t)||2 exceeds a threshold v0 .

2.3 Detection of Initial Contact Even for considerably low walking speed, the initial contact is characterized by a significant peak in the jerk ||da(t)/dt||2 2 R 0 , see Figure 2. However, we found that such peaks may also occur at toe-off, especially in paretic gait. Therefore, we propose two additional conditions: Either, a certain time Tsw,min since toe-off needs to have passed, before initial contact detection is enabled. IMU (t)||2 must have Or the horizontal sensor velocity ||vxy fallen below 75% of its maximum-since-toe-off value, i.e. IMU IMU (t)||2 < 0.75 max ||vxy (⌧ )||2 , (5) ||vxy ⌧ 2[t0 ,t] Z t IMU vxy (t) := R(⌧ )a(⌧ ) (a(t0 )T R(⌧ )a(⌧ ))a(t0 ) d⌧, t0

where t0 and R(t) are defined as in Section 2.2. The underlying assumption is that the horizontal velocity increases at toe-off, while it decreases (before or) at initial contact. If either this or the time criterion is fulfilled when jerk exceeds a threshold jhs , then the automaton switches to loading response. Please note that in paretic gait the foot might touch ground and move forward again multiple times before coming to rest. In such cases, the proposed method only detects the first (initial) ground contact and the only transition that can follow is the full contact as explained in Section 2.1. This is important for 2 i.e.

small (noise) peaks do not trigger transition



the Adaptive Peroneal Stimulator introduced in [7], since foot-to-ground angles are recorded during swing phase and used to adjust the stimulation intensity profile. These adjustments assume that the measured angles are a direct outcome of the stimulation. Therefore, loading response must be detected and the recording must stop at the very first ground contact after toe-off.

S797

(a)

2.4 Adaption to Gait Velocity Comparing the norm signal trajectories ||g(t)||2 , ||a(t)||2 for different subjects and different sensors at different walking speeds leads to the conclusion that the noise amplitudes at rest as well as the signal amplitudes during motion vary largely. In order to compensate the variations, we continuously determine the smallest standard deviations a , g of nstdev 2 N>0 subsequent samples of a(t), g(t), respectively, within the last Tobs > nstdev /fs seconds. The rest band widths are then continuously set to a multiple of these noise amplitudes, i.e. arest = k a , grest = k g , k 2 R>1 . Tobs can be used to influence the speed of adaption but must be larger than the longest duration of a step to guarantee that a , g characterize the signal noise amplitudes during foot-flat. To compensate the influence of gait velocity variations on toe-off and initial contact detection, we determine the maximum values of toe pitch , ||v (t)||2 , and ||da(t)/dt||2 for each step. In each foot-flat phase, we set the threshold values 0 , v0 , and jhs to 20% of the average of the last m stepwise maximum values, respectively. The speed of adaption can be increased by choosing smaller values for m, and vice versa. Finally, we also determine the time T¬ff between each two consecutive foot-flat phases and set Tsw,min to 20% of the average of T¬ff over the last m steps.

(b)

(c)

3 Stroke Patient Experiments The proposed gait phase detection algorithm with automatic gait velocity adaption is evaluated in real-time simulations3 using measurement data from a large number of gait experiments with both stroke patients and healthy subjects. Participants gave informed consent and the study was approved by the Local Research Ethics Committee at Charité Berlin. Patients and healthy subjects were asked to walk on a treadmill, on level ground, and on stairs; at self-selected speed as well as 25% slower and faster. Drop foot patients were asked to walk with ankle-foot orthosis or walking sticks and with a standard peroneal stimulator. The following parameter values were chosen: fs = 100 Hz, k = 4,

nff = 20, Tobs = 2 s,

↵ = 2, nstdev = 20,

m = 5.

Detection results were checked for irregularities and inspected visually step-by-step. Despite the large variety of subjects, terrains, gait velocities, and walking supports, more than 95% of all gait phases were correctly detected. 3 This means that the data was processed in an online manner, i.e. only current and previous data was used by the algorithm.

Figure 4 Experimental results for a drop foot patient walking on level ground at self selected speed with peroneal stimulator support. Staircase represents detection of foot-flat(0), pre-swing(1), swing(2), and loading response(3) phase. (a) Automatically adapted thresholds of acceleration and angular rate vector norm are used to detect foot rest. (b) Toe-off is detected by sign change in pitch rate and a sudden increase in toe velocity. (c) Initial contact is detected by a peak in the jerk norm after/during a decrease of horizontal foot velocity.



S798

detection algorithm. Furthermore, we would like to thank Cordula Werner and our research partners from Hasomed GmbH for performing gait experiments and recording valuable data. Finally, we thank Daniel Laidig for his help regarding plot design and Markus Valtin for his skillful support in automatic data evaluation. This work was partially conducted within the research project APeroStim, which is supported by the German Federal Ministry of Research and Education (FKZ 01EZ1204B).

5 References Figure 5 Comparison of gait phase durations for a drop foot patient walking on a treadmill and on level ground at self-selected speed. Step-to-step variance is significantly larger without additional support of the treadmill’s handrail and continuous motion. Figure 4 presents three steps from an experiment with a drop foot patient. It is visible in all three subplots that all signals vary from step to step and that repeatability is only given to a very limited extent. Therefore, it is important that the transition criteria extract only the most relevant features. The second step, for example, shows a remarkably different pre-swing phase with a fast twitch at heel-off. The loading response is almost nff samples short, which indicates that the foot touched ground with heel and toes simultaneously – in contrast to both other depicted steps. The obtained results can be used to compare gait phase duration features, as for example in Figure 5. The presented gait phase duration plots reveal that the patient exhibits a more regular gait when walking on a treadmill at (self-selected) constant speed and holding on to a handrail.

4 Conclusion and Future Work We proposed a new method for gait phase detection that employs a single 6d IMU on the foot/shoe and adapts its parameters automatically to the subject’s gait velocity. We demonstrated that the method can handle the challenge of irregularity in stroke patient’s gait and is suitable for a large range of scenarios. A more detailed experimental evaluation including assessment of timing with respect to a reference method will be subject of future research. Furthermore, the method might be extended to allow analysis of running motions.

Acknowledgements We would like to express our deep gratitude to the patients and the healthy subjects that participated in experiments. We are also indebted to Steffen Schäperkötter who developed and evaluated methods that contributed to and initiated the development of the current gait phase

[1] D. Kotiadis, H.J. Hermens, P.H. Veltink, "Inertial Gait Phase Detection for control of a drop foot stimulator inertial sensing for gait phase detection", Medical Engineering & Physics, vol. 32, no. 4, pp. 287-297, 2010. [2] A. Mansfield, G. M. Lyons, "The use of accelerometry to detect heel contact events for use as a sensor in FES assisted walking", Medical Engineering & Physics, vol. 25, pp. 879-85, 2003. [3] B. Mariani, S. Rochat, C. Büla, K. Aminian, "Heel and toe clearance estimation for gait analysis using wireless inertial sensors", IEEE Transactions on Biomedical Engineering, vol. 59, pp. 3162-68, 2012. [4] J. Rueterbories, E. G. Spaich, B. Larsen, O. K. Andersen, "Methods for gait event detection and analysis in ambulatory systems",Medical Engineering & Physics, vol. 32, pp. 545 - 552, 2010. [5] P.G. Savage, "Strapdown inertial navigation integration algorithm design part 1: attitude algorithms", Journal of guidance, control, and dynamics, vol. 21, pp. 19-28, 1998. [6] T. Seel, T. Schauer, J. Raisch, "Joint axis and position estimation from inertial measurement data by exploiting kinematic constraints", Proc. of the IEEE International Conference on Control Applications (CCA), pp. 45-49, 2012. [7] T. Seel, S. Schäperkötter, M. Valtin, C. Werner, T. Schauer, "Design and control of an Adaptive Peroneal Stimulator with inertial sensor-based gait phase detection", Proceedings of the 18th Annual International FES Society Conference, pp. 177-180, 2013. [8] T. Seel, D. Laidig, "Feedback Control of Foot Eversion in the Adaptive Peroneal Stimulator", submitted to 22nd Mediterranean Conference on Control and Automation, 2014. [9] Y. Shimada, S. Ando, T. Matsunaga, A. Misawa, T. Aizawa, T. Shirahata, E. Itoi, "Clinical application of acceleration sensor to detect the swing phase of stroke gait in functional electrical stimulation", Tohoku J. Exp. Med., vol. 207, pp. 197-202, 2005. [10] R. Williamson, B. J. Andrews, "Gait event detection for FES using accelerometers and supervised machine learning", IEEE Transitions on Rehabilitation Engineering, vol. 8, pp. 312-19, 2000.



S799

Model-based Control Approach for a CPAP-Device M. Scheel1,2 , A. Berndt2 , R. Kölln2 , A. Sievert3 , O. Simanski1 1 Hochschule Wismar, University of Applied Sciences - Technology, Business and Design, Wismar, Germany 2 HOFFRICHTER GmbH, Schwerin, Germany, scheel@ho↵richter.de 3 Institute of Automation, University of Rostock, Rostock, Germany

Abstract

In the field of sleep medicine it may lead to breathing interruptions, if the upper respiratory tracts collapse. The treatment of this sleeping sickness is mostly done with a CPAP(Continuous Positive Airway Pressure) therapy, in which the upper respiratory tracts of the patient are splinted by a positive pressure. The specification of such a medical device is to maintain the therapy pressure ascertained by the medical scientist, irrespective of the patient’s breathing. In doing so a natural respiration can be ensured.

1

Introduction

The obstructive sleep apnoea syndrome (OSAS) is a disease located in the field of sleep medicine. The OSAS is characterized by the collapse of the upper respiratory tract, while the musculature at the lower thorax is still active[1]. By these respites the blood-oxygen concentration is decreased. Below a critical level the patient is awakened by an alarm function of the human body [2].

reach the deep sleep phases because the wake up more often (figure 1b). Symptoms of sleep apnoea are fatigue, lack of concentration and high blood pressure [5]. This can lead to a higher risk of coronary thrombosis and apoplexy. To prevent these consequential diseases the apnoea has to be treated. The standard therapy to threat this sleeping sickness is the ”Continuous Positive Airway Pressure“ (CPAP) therapy. The patient is connected to the medical flow generator system through a breathing mask. A positive airway pressure is facilitated to splint the upper respiratory tracts to avoid a resulting sleep apnoea (s. figure 2).

Figure 1 Sleep cycles. a) healthy patient b) apnoea patient Figure 1 shows the categorization of di↵erent sleep cycles [3, 4]. The first sleep cycle is the REM sleep (rapid eye movement). This phase lasts only a few minutes. Blood pressure, heart rate and respiratory frequency are increased. A Characteristic of the REM sleep is the rapid eye movement. In NREM sleep (non rapid eye movement) the brain activity is decreased. The di↵erent NREM sleep phases describe the transition from falling asleep to light sleep to deep sleep. Especially the NREM sleep phases 3 and 4 in figure 1a are essential for a restful sleep. Apnoeapatients can not

Figure 2 CPAP-therapy: splint of the upper respiratory tracts [6] Di↵erent therapy devices are etabilshed in the global market. Devices in the clinical sector are from companies like Drägerwerk AG & Co. KGaA. Companies like HOFFRICHTER GmbH, ResMed and Respironics develop devices for the home care area. The main content of this paper focuses on the modeling of the breathing therapy system. In the following section the breathing therapy system is specified. Afterwards a model structure for a CPAP-therapy system is introduced. The fourth chapter focuses on a short description of a first



S800

control approach and some tests. After all further ideas are resistance Ra and the armature inductance La . The feedback proposed. from the mechanical part to the electrical cicuit is described trought the back electro-motive force constant Ke . The parameters for the mechanical part are the moment of 2 Process description Figure 3 shows the application of a CPAP-therapy device. inertia Jm and the constant of friction Kr . The motor constant Km couples the current to the drive torque. Most of the parameters were taken from the data sheet of the engine manufacturer. A simulation based optimization was used to determine unknown or uncertain parameters. The resulting model quality for the described electromechanical part is shown in figure 4. a) 1

Figure 3 Schematic of a CPAP-therapy system

current

turns 20 n [1000 rpm]

i [A]

0.8 0.6 0.4 0.2 0

0.1

0.6

18 16 14 12 10

1

0.1

0.6

1

——————————————————————b)

current

turns n [1000 rpm]

28 1 i [A]

The main component of the CPAP-device is a centrifugal blower. The therapy tube and the breathing mask couples the CPAP-device to the patient. An expiration valve is located at the mask to lighten the expiration. The basic idea is the generation of the therapy pressure in the mask Pmask to splint the upper respiratory tracts. The therapy pressure can be controlled by the device pressure Pdevice . This pressure can be controlled with the help of the centrifugal blower. The breathing interacts with the pressure in the mask. The main ambition of the CPAP-device is to achieve a defined control accuracy of the mask pressure. This is definded in the standard DIN EN ISO 17510-1. The maximum admissible pressure deviance in the mask is set to 0.5 hPa by therapy pressures from 4 to 10 hPa and 1 hPa by therapy pressures above 10 hPa. The available manipulating variable is the motor voltage Ua of the centrifugal blower. The measurement variables are: current ia , motorspeed n and the device pressure Pdevice .

0.5 0

0.1

0.6 t [s]

1

26 24 22 20 18

0.1

0.6 t [s]

1

Figure 4 Validation of the motor model. blue: model, green: real process a) for a voltage step from 30% to 50% This section describes the modeling of a CPAP-therapy b) for a voltage step from 50% to 70% device. The resulting model should be the basis for a following control design. The system in figure 3 consists of an electromechanical and a pneumatical part. The motor The model shows good results for a free-wheeling motor. belongs to the electromechanical part. The respiratory In the case of the CPAP-device the model needs to be system of the patient, the therapy tube and the mask expanded by the attached impeller. The impeller generates belong to the pneumatical part. The centrifugal blower a load which acts on the drive. The load ML on the engine ˙ A description is known is the connector between the electromechanical and the depences on the volume flow rate V. pneumatical part. from literatur [8]: In a first step the motor was modeled for a free-wheeling (2) ML = ⇢ V˙ ! Kload , case. The relationship between the motor voltage Ua and the motorspeed n can be described by a linear second order where ⇢ is the gas densitiy of the air and Kload is a collection state space model: of blower depended parameters. The model is valid for 3 " # "1# " # 2 Ra K e the assumptions of an incompressible, inviscid fluid and 77 ia i˙a 66 La 7 = 664 KLma + La Ua 5 Kr 7 an infinite number of fan blades. The compression for the ! ˙ ! 0 Jm Jm (1) used therapy pressures is slight and will be neglected. The 30 ! n= volume flow rate V˙ is derived from available measurement ⇡ variables and is assumed to be well known. An additional di↵erence between the free-wheeling The parameters for the electrical circuit are the armature

3

Process modeling



motor and the centrifugal blower is a modified moment of interia Jm . The extended motor model of the electromechanical part of the CPAP-device is given by: 3 " # " # " # 2 Ra Ke 1 666 L 777 ia i˙a L a a La U 7 = 664 Km + 7 ˙ a Kr +⇢ V Kload 5 ! ˙ ! 0 Jm Jm (3) 30 ! n= ⇡

expansion) equation 5 can be simplified to: dV =

dP =

current ia [A]

0.5 0 0.1

0.8

1

n [1000 rpm]

turns 20 18 16 14 12 10

(6)

Pmask Pmask dT + dm T m Pmask dVrespiratory Vrespiratory Pmask dVexpiration Vexpiration

Pmask dVtherapy Vtherapy (7)

This model is difficult to manage for a controller design, because of the unkown lung parameters and -states.

1 0

V V V dP + dT + dn P T n

m , where m is the weight and M is the molecular with n = M weight, equation 6 can be converted to dP. With regard to the process in figure 6 provides:

A simulation based optimization was used to determine the modified parameter Jm and the unknown parameter Kload . The model quality is shown in figure 5.

1.5

S801

0 0.1

4

Controller design

The description of the electromechanical part (S 1 ) is well known. However the pneumatical model (S 2 ) is difficult to manage. A familiar procedure to regulate a system with well known and uncertain parts is the cascaded control structure (see figure 7). 0.8

t [s] Figure 5 Validation of the extended motor model for a voltage step from 30% to 50%. blue: model, green: real process

1 Pmaskset +

C2

n set +

C1

S1

n

Pdevice S2

Pmask measurements estimator The pneumatic part consists of the respiratory system of the patient, the therapy tube, the mask and the centrifugal blower, shown in figure 6. Figure 7 Example of a double stage cascade

Figure 6 Schematic of the pneumatical part of a CPAP-system

In the inner cycle (C1 + S 1 ) a model-based controller (C1 ) was used. In the outer cycle (C2 + S 2 ) of the cascade an empirically adjusted controller (C2 ) was utilized. This is because of an uncertain model for the pneumatical part. Furthermore the pressure in the mask can not be measured and has to be estimated based on the available measurements.

The developed closed-loop pressure control has been A model can be derived from the thermic constitutive equation. The volume V is a function of temperatur T , subjected to a test to determine the control accuracy. In a first functional substantiation the pressure in the mask pressure P and amount of substance n. was disturbed by a sinusoidal volume flow rate. The flow V = f (P, T, n) (4) is generated from a reciprocating pump. The sinusoidal volume flow rate imitate the breathing of the patient. A The total di↵erential results in: frequency of 20 beats per minute (on average this is the maximum ventilation rate for most patients and thereby the @V @V @V dP + dT + dn (5) strongest influence on the pressure) and a tidal volume of dV = @P @T @n 500 mL have been set. Figure 8 shows the control results For an ideal gas (no volume compression, no volume for the described scenario.



complex and more difficult to handle. However, a first control approach based on a cascaded control structure was developed. One of the controller (C2 ) was designed heuristically. First results of the controller were shown in figure 8 and 9. Further steps are necessary to develop a model for the pneumatic part of the system with the aim to achieve a better control performance.

mask pressure

P [Pa]

430 400 370

0

2

4

6

S802

8

10

t [s]

Reference

[1] Groontoonk, Sandra : Beeinflussung des CPAP-Druckes bei obstruktivem Schlaf-ApnoeSyndrom durch Gabe von Theophyllin : Hohe Medizinische Fakultät der Ruhr-Universität Bochum : The closed-loop pressure control achieved a good result Dissertation 2003 within the specified control goal described in section 2. The controller could be tested on patients in a sleep laboratory. [2] Hübers, Dr. Ulrich : Schlaf, Schlafapnoe und The patient was under the surveillance by the specialised Schnarchen : sta↵. The quantitative estimation of the control quality is http://www.schlafapnoekieferorthopaedie.de/ : 12.03.13 shown in figure 9. A comparison of the mask pressure with [3] Tillack, Anna : Rätsel REM-Schlaf : and without regulation is made. http://www.br.de/radio/bayern2/sendungen/iq-wissenschaft-und-forschung/mensch/rem-schlafmask pressure traeume100.html : 12.03.13 1,000

P [Pa]

Figure 8 Control results for a therapy pressure of 400 Pa

[4] Nagel, Dr. rer. nat. Geraldine : Schlafphasen : http://www.onmeda.de/schlafen/schlafphasen.html : 28.08.13 700

[5] http://www.schlafapnoe-online.de

0

2

4

6

8

10

t [s] Figure 9 Control results. blue: with regulation, magenta: without regulation

5

Conclusion

[6] Honeger, Dr.med. Urs : Die nasale CPAP-Therapie : http://www.druhonegger.ch/mobile/mcpap.html : 20.08.13 [7] Elektrische Antriebe - Regelung von Antriebssystemen, Band 10 : Schröder, Dierk : Springer-Verlag, Berlin Heidelberg : 2009 [8] Bommes, Leonard : Fricke, Jürgen : Grundmann, Reiner : Ventilatoren 2.Auflage : Vulkan-Verlag GmbH, Essen : 2003

The main purpose of this paper was to looking for a [9] Scheel, Mathias : Umsetzung einer Druckregelung simple model for the CPAP-therapy system. A model for eines druckschlauchlosen Atemtherapiegerätes : the electromechanical part was developed. The modeling Master-Thesis Hochschule Wismar, Fachbereich of the pneumatic part of the CPAP-therapy system is more Elektrotechnik und Informatik, 2013



S803

Automation of medical systems – Current techniques, limitations and challenges Berno J.E. Misgeld1, Thomas Schauer2, Olaf Simanski3,4 1 Philips Chair for Medical Information Technology, RWTH Aachen University, Aachen, Germany, e-Mail: [email protected] 2 Contol Systems Group, Technische Universität Berlin, Berlin, Germany, e-Mail: [email protected] 3 Computational Engineeering and Automation Group, Hochschule Wismar - University of Applied Sciences: Technology, Business and Design, Wismar, Germany, e-Mail: [email protected] 4 Medical Control Group, Control Application Centre, University of Rostock, Rostock, Germany

Abstract Automation of medical systems is a growing multidisciplinary field, comprising techniques that range from modelling to feedforward or feedback control of physiological systems and medical devices. Although, the last decades showed a rapid increase of feedback control applications in aerospace, automotive, mechatronic or process industries, examples for product-available feedback-controlled medical devices are still few. This observation should receive special attention, since it is diametrically opposed to tremendous advances in sensor, actuator and computational technology. The aim of this contribution is to provide insight into modelling and feedback control problems for medical systems. A historical background to medical automation approaches is given firstly, followed by a rigorous structuring of feedback control strategies, needed to classify complexity of subsequently described examples. These examples are taken from currently unsolved control problems used to enlighten sensor, actuator or complexity related challenges. We conclude with the motivation for future research directions introduced to face the indentified challenges.

1

Introduction

The last century has seen a continuous growth and tremendous technological advancement in biomedical engineering, a term that was established in the late 1960s. Thus, by the end of the 20th century 11 technologies were identified by the Committee on Emerging Technologies of the EMBS (Engineering in Medicine and Biology Society) to have a revolutionary impact on the first decade of the 21st century [1]. Among those, a number of identified technologies have a strong connection to automation and control theory, like for example advanced computer modelling and simulation of physiological systems, implantable devices, artificial organs and assist devices and rehabilitation engineering. One decade later, and some more decades after the introduction of pacemakers, anaesthetic machines, artificial lungs, heart-lung machines, mechanical ventilators, or total artificial hearts, the number of successfully automated or feedback controlled applications in biomedical engineering that is product available is still low. However, one has to make careful use of the term automated or feedback controlled, since there already exists a number of devices employing automation or feedback controllers. Hence, differentiating between the type or structure of automation or feedback control, employed in a biomedical engineering scenario, is of fundamental importance and concepts for doing so will be introduced in the next section. Due to the technological contribution in terms of new smart actuators and sensors, but as well increadibly increased calculation power and possibilities for distributed

systems using internet access, the application of automation and control to the biomedical area opens vibrant possibilities for future research and products. These are equipped with the potential to have a significant impact on society even in problems that are still unsolved while being investigated by researchers and companies alike and over decades. From a historical perspective, the scientific field of control engineering and cybernetics was all along linked to biological or biomedical systems. First connections can be traced back to early work of Bernard and Cannon [2] who postulated dynamic regulation in organisms leading to equilibrium conditions (homeostasis). The work was picked up by Wiener [3] who introduced the term cybernetics, yet ultimately failed to establish it as successful science in the biological or biomedical context. Here again changes were made possible by recent advances in sensor development and miniaturisation introducing systems biology as a new field to science. Despite the focus on automation and control engineering in this paper, enlightening the techniques and limitations of sensors, actuators and algorithms, two none the less important practical aspects should be mentioned. Firstly, a dominating opinion in some medical areas seems to be that advanced feedback control strategies or algorithms deliver only low or moderate improvement of performance and quality over established techniques. The contrary is often hard to prove, since large studies involving animals or humans would be necessary which are expensive, complex and sometimes questionable. For example, a study on an automation of a complex clinical guideline [], would require a multicentric study involving highly trained doctors



Fig 1. Classification of feedback controlled medical systems. for the classical and automated approach. Furthermore, so called soft objectives exist in many biomedical problems that cannot be easily quantified by well-defined and reproducible measurement techniques or performance indices. Secondly, there exist a number of natural boundaries ranging from regulatory affairs to individual acceptance and even communicational problems between medical doctors and control engineers. For example regulatory approval might impose additional requirements on software development for feedback control and should thus be handled with suitable test scenarios. Here, the proof required for the approval of a device controlled by a feedback algorithm should rely on predefined protocols involving suitable disturbance or change of operating point tests.

2

Classification of medical feedback control systems

In order to adequately address limitations and challenges, associated with automated or feedback controlled medical systems, we classify the feedback control with regard to structure. We differentiate between three types of fundamental concepts for medical feedback systems [5]. All of the concepts are collected in Fig.1. TYPE I: The feedback control loop is located inside the medical device or system. Device internal sensors are used to control the device-internal process variable for which full information is assumed to available. Furthermore, there is no or negligible retroaction from the patient, thus no patient-device coupling exists. Examples for type I control systems are the control of gas concentration in an anaesthesia or intensive care device or the dosage control of a patient applied drug. TYPE II: In this structure the control loop has to be designed in a patient oriented way, since there exists a retro-

S804

action from the patient. The medical control system measures device internally, where patient-device coupling takes place inside or outside the medical device. Due to partial process uncertainty the feedback controller has to guarantee stability and performance for the coupled system. Thus additional complexity due to the partially involved physiological system of the patient has to be regarded. However, controller design can be conducted towards guaranteeing interaction stability of the feedback control system. Examples of type II control sytems are a pressure controlled ventilation device or an actuator used to generate a force or torque in a rehabilitation robotics scenario. TYPE III: The feedback control system is closed with the patient in the loop. This is a compensatory physiological control loop, where the control objective is sometimes physiological, rather than physical. This control loop structure comprises the highest level of complexity, since the physiological system of the patient, consisting usually of highly complex, uncertain and nonlinear system dynamics, is a full part of the loop. An example for a type III system where the physical quantity can be directly measured is the artificial pancreas. Here, a sensor measures blood-sugar concentration and insulin is applied by the feedback controller as a control input. An example for a type III system where the control objective cannot be directly measured is a left ventricular assist device (LVAD). The device is coupled to cardiovascular dynamics, which is usually handled by low level cascaded control algorithms. The overall objective of the control is to support the heart as needed and, if possible, guarantee a progressing recovery of heart function. For this control objective a non measurable value(s) have to be derived by the employed algorithm.

3

Artificial pancreas

To this day, the dysregulation of blood sugar level is critical in medical care of patients with type 1 diabetes. The associated problem is twofold. On the one hand, manual control of blood sugar by single shot injection often leads to hyperglycemia and subsequent complications, if not managed properly. The adjustment of blood sugar to the biological norm value, on the other hand, significantly increases the risk of severe hypoglycemia. This regulation, if taken over by feedback controlled system, is highly safety critical since hypoglycemic blood sugar level might lead to unconsciousness and even death. In the sense of a classical feedback control system, the artificial pancreas comprises three components visualized in Fig. 2 (Type III control system). These components are an insulin infusion pump, a continuous glucose sensor and an algorithm realised on a computer to continuously regulate the insulin dosage depending on sensor measurements. Connected to this classical setup a number of challenges and opportunities exist for the control engineer. Market available blood glucose sensors have relatively large sampling times and a timedelay, which is in addition to the low pass behaviour of subcutaneously measured glucose. If insulin is used sol-



S805

Fig 2. Classical artificial pancreas. edly as a control input, only positive values are allowed (saturation). Furthermore, and in addition to the complex nonlinear and patient individual dynamics of the endocrine system, the human body is subject to disturbances, like for example food intake and activity. This being said, it is highly questionable that a feedback controller can successfully guarantee the required performance (avoid hyper- or hypoglycaemic conditions) for this uncertain, nonlinear system with unknown (if not measured) disturbances. In order to resolve this problem, a number of research opportunities arise. With the development of new subcutaneous and intravenous glucose sensors [6] patient-individualised feedback control solutions can be developed. However, these have to rely on the successful inclusion of disturbance estimation techniques, for which new body-worn sensor techniques are available. Modern techniques from areas like robust, linear parameter-varying or modelpredictive control are most promising tools if applied in cascaded manner for feedback control and feedforward disturbance rejection. The first goal should be an application scenario, where the sensor and disturbance complications are manageable, for example an intensive care unit patient. Here, questions of medical approval, individualised models and communication between doctors and engineers can be addressed specifically.

volitional residual muscle activity [7]. Wireless miniature inertial measurement units (IMUs) are likely to become the most suitable approach to sense human limb movements under daily life conditions [8]. Based on these new sensor technologies, the lack of precise patient-indivual models might partially be tackled by using cascaded control schemes with fast internal EMG or acceleration feedback loops to directly regulate the desired contraction levels. Feedforward control will be prefered over feedback control when fast movement needs to be generated in a short time interval due to the slow muscle dynamics. As many human movements are cyclic, especially during walking, iterative learning control can be used to adjust the open-loop stimulation profiles from cycle to cycle (step to step). Regulation takes place in between cycles by analyzing the motion error of the last completed cycle and by updating the feedforward control law such that the error will consecutively decrease. Fig. 3 illustrates this idea for an adaptive dropfoot stimulator which is currently under development [9]. A sufficient dorsiflexion for the foot is maintained by updating the stimulation profile from step to step.

4

5

Functional Electrical Stimulation

Functional electrical stimulation (FES) is often used in the rehabilitation and treatment of stroke patients to support weak motor functions. Since the response of paretic muscles to electrical stimuli is highly individual and timevarying, the use of closed-loop control methods (type III) is required whenever precise movements shall be generated. However, existing stimulation systems only regulate the voltage or current amplitude of the generated stimulation pulses (Type II control system). This kind of control is required for example due to the patient indiviual electrical load. The stimulation effects, meaning the induced muscle contractions and the resulting limb movements, are usually not measured and regulated. Reasons therefore are diverse: inadequate sensor technology, too large effort in patient-individual modelling, and considerably short time periods for movement generation in contrast to a slow dynamics of the artificially stimulated muscles. However, recent advances in electromyography recording and signal processing will allow to assess the amount of muscle contractions and to differentiate FES-induced and

Fig 3. Adaptive drop-foot stimulator using iterative learning control.

Automatic drug delivery in anaesthesia

In clinical practice anesthesiologists have to observe and control a large amount of haemodynamic and respiratory variables as well as clinical signs for adequate hypnosis and analgesia. In neuro-, thoracic- and abdominal surgery a continuous neuromuscular block is needed to guarantee optimal surgical conditions. A neuromuscular blocking drug is administered in order to prevent reflex muscle movement. With the introduction of short-acting medications, continuous application was possible. The continuous application can be realised in two different ways. A first approach was developed with the so-called target controlled infusion systems (TCI systems). Here, the amount of the administered drug is calculated on the basis of implemented pharmacokinetic-pharmacodynamic models by specifying a target drug concentration. TCI systems work on the principle of action of a forward-control. The disadvantage of this type of control is the lack of measurement infomation.



In the event of a possible measurability of the subcomponents of anesthesia, such as the deep hypnosis, analgesia and the muscle relaxation can be closed control loops are developed . Following the feedback control classification of medical systems the proposed control structure, depicted in Fig. 4, results in a Type III system with the patient fully in the closed-loop. The main problems of such a scheme are in addition to the thoroughly tolerable measurement of selected parameters the almost unknown patient-internal cross-effects which are represented as Gij in Fig 4. Using the T1 measurement (muscle response to a stimulus, measured electromyographically) neuromuscular blockade can be determined relatively safe. For the measurement of the depth of hypnosis the BIS (bispectral)index established itself in recent years as a possible measurement, but which is not un-critical. Most difficult is the quantitative detection of analgesia. The evaluation of the heart and blood pressure variability are first approaches. The described obstacles complicate the design of a modelbased multivariable controller, so that initially a decentralised multivariable control system, as sketched in Fig 4., was implemented [10].

Fig 4. Medical feedback system for automatic drug delivery in anaesthesia.

6

Conclusions

Technological advances, like those seen in sensor and actuator technologies, are not fully developed, applied or used in potential medical feedback control systems of type III. Examples are blood sugar or the depth of hypnosis and analgesia, which need to be safely measured with minimal time-delay and measurement errors. This in turn hinders the successful application of feedback control even though there is a relatively large body of modern methods available, covering uncertain, coupled and nonlinear systems. The aim of a modern medical feedback control system of type II or type III should be a patientindivisualised approach. Thus by relying on patientindividual parametrised models, a balancing act between too much robustness on the one hand or controller over-

S806

individualisation on the other hand has to be made. This goal can only be achieved by a close interconnection of modelling and control design. In this context, methods that rely on system or disturbance a-priory knowledge, like for example feedforward, model-predictive, disturbance rejection or iterative learning control are suggested to reduce conservatism thereby improving performance and safety of the overall feedback control system. We acknowledge the increased complexity in regulatory approval due to functional safety requirements and point out that it is of fundamental importance to show medical and economic advantages in clinical studies. The success and acceptance of feedback control in biomedical systems is strongly depending on examples, which should involve the solution of a sensor and intra- or interpatient variability and complexity related problem, like for example the automatic drug delivery or the artificial pancreas.

7

References

[1] Nebeker, F.: Golden accomplishments in biomedical engineering, IEEE Engineering in Medicine and Biology Magazine, IEEE , 21(3), pp.17-47, 2002 [2] Cannon, W.B.: The Wisdom of the Body, W.W. Norton, 1932 [3] Wiener, N.: Cybernetics, or Control and Communication in the Animal and the Machine, John Wiley and Sons, 1948 [4] Luepschen, H., Konowalczyk, S., Meier, T., Pikkemaat, R. and Leonhardt, S.: Automatic lung recruitment maneuvers using robust closed-loop control of the oxygen saturation in mechanical ventilation, World Congress on Medical Physics and Biomedical Engineering, pp. 703-706, 2009 [5] Leonhardt, S. and Mersmann, S.: Automatisierungstechnik für die künstliche Beatmung – eine Standortbestimmung, at – Automatisierungstechnik, 55(5), pp. 244-254, 2009 [6] Lunze, K., Singh, T., Walter, M., Brendel, M. and Leonhardt, S.: Blood Glucose Control Algorithms for Type 1 Diabetic Patients: a Methodological Review, Biomedical Signal Processing and Control, 8, pp. 107-119, 2013 [7] Schauer, T. and Klauer, C.: Verwendung der Elektromyographie in der funktionellen Elektrostimulation, ORTHOPÄDIE-TECHNIK, 64(6), pp. 26–32, 2013 [8] Seel, T., Raisch, J. and Schauer, T.: IMU-based Joint Angle Measurement for Gait Analysis, Sensors (in review), 2013 [9] Seel, T., Valtin, M. and Schauer, T.: Neue Technologien für die Peroneusstimulation: Bessere Versorgung bei Fußheberschwäche. Deutsche Zeitschrift für klinische Forschung, 13(4), pp. 43–47. 2013 [10] Simanski, O., Janda, M., Schubert, A., Bajorat, J., Hofmockel, R. and Lampe, B.: Progress of automatic drug delivery in anaesthesia – „The Rostock assistance system for anaesthesia control (RAN)“ Int. J. Adapt. Control Signal Process, 23, pp. 504–521, 2009



S807

Linear affine lung mechanics model with emphasis on pleural dynamics. 1

1

2

2

Chuong Ngo , Berno Misgeld , Thomas Vollmer , Stefan Winter and Steffen Leonhardt 1 2

1

Chair of Medical Information Technology, Helmholz Institut for Biomedical Engineering, RWTH Aachen University Philips Research, Aachen, Germany

Abstract This paper introduces a novel mathematical model which simulates the pleural pressure of the respiratory system by using simple mechanical and electrical approaches. In context of lung mechanics and lung physiology, the lung-chest wall relation and the pleural space are discussed. The fluid accumulated in the pleural space appeals as a mechanical pressure coupling between the lung tissue and respiratory muscles. A mechanical model with telescope cannisters, tube and spring and an equivalent electrical model was derived which has a clear anatomical analogy to lung physiology. The frequency response shows the affine characteristics of the respiratory system. The simulation results show a high agreement with literatures values and physiological lung knowledge.

1

Introduction

A number of lung models have been introduced over the last decades. The simplest lung model consists of a resistance !!" , and a compliance !!" , which describes the flow resistance of the airway and the elastic recoil characteristic of lung tissue [7][2]. Later, multi-compartment models were introduced by studying the physiological structure of the respiratory system. Ottis [9] considered a left-right model where the lung are divided into two identical lung wings. Mead [8] introduced a 7-elements model with periphery resistance, inertance, mouth, bronchial, lung and chest wall compliance. Rideout [11] added a pressure source to simulate the respiratory muscles. These models’ approach is linear lumped element modeling, where the structure is simple and could be described using electrical equivalent circuits. However, these models seem not to be adequate for estimation of the pulmonary pleural pressure. To our knowledge, the pleural pressure !!" provides important information about the patient’s lung state, especially in lung diseases such as congestion, barotrauma or effusion. In this work, we introduce a simple model which describes the relationship between lung, chest wall and the pleural space.

2

which sucks air from the environment through the upper airway into the bronchial tree (see Fig. 1). According to Weibel [14], the bronchial tree could be divided in to 23 generation with generation 0 the trachea and generation 23 the alveoli. In spontaneous breathing, flow resistance and the tendencies of the lung and chest wall to recoil are overcome by the respiratory muscles’ effort. The respiratory muscles consist of the thorax and abdomen, which work together under the control of the brain, mostly unconsciously. The pleural space is the potential space that locates between the parietal pleura inside of the chest wall and the visceral pleural surrounding the lung. The lung is only suspended at the hilum from the mediastinum, and has no other attachment to the chest wall. The thorax cavity functions as a water bed in which the lung floats, surrounding by a thin layer of pleural fluid. The fluid, in normal case about 8-12 ml, excesses from the pulmonary capillaries to the pleural space and exits though the lymphatic system. This process is described by the Starling Equation and is normally kept in equilibrium [12].

The pleural space

Breathing is a mechanical process delivering fresh air into the lung alveoli, where oxygen and carbon-dioxide are exchanged. To guarantee the effective gas exchange, not only the blood-gas barrier and the breathing control center must remain intact, but also the mechanical functionality of the respiratory system must be maintained. In lung pathophysiology, such as obstructive and restrictive lung diseases, mechanical properties alter and cause some retrains in gas delivery and distribution. The lung expands and collapses elastically like a balloon,

Fig. 1. Physiological structure of the respiratory system



According to Guyton [3], !!" is sightly negative during the breathing cycle. At the end of a normal expiration, where lung volume is at functional residual capacity (FRC) level, negative pressure (about -5 cmH2O) is generated between the visceral and parietal pleura by the opposing elastic forces of the chest wall and lung. During inspiration, !!" decreases to -7.5 cmH2O, caused by the contraction of respiratory muscles, which expands the thorax cage and cre ate more negative pressure inside the pleural space. During expiration, the events are reversed. The pleura space has two main roles: firstly, mechanical coupling which transfers and regulates the pressure inside and outside of the lung. Secondly, its fluid serves as lubrication for the movement of the lung inside the thorax. In lung diseases such as restriction or pleural effusion, pressure changes occur in the thorax cage and may be observed in !!" . In our knowledge, except on the patients’ characteristics like height, weight and age, there are several factors which could have an influence on !!" : • Static: the positive end-expiration pressure (PEEP) applied by the ventilator device, the average pulmonary arterial and pulmonary capillary blood pressure, and the amount of accumulated air (barotrauma) or fluid (pleural effusion) in the thorax cavity. • Dynamic: breathing and heart beat. !!" is the pressure at the outer surface of the lung and the heart and inner surface of the thoracic cavity. During the activity of the respiratory and cardiovascular system, !!" changes along lung volume and heart beats.

3

Modeling of lung mechanics including the pleural pressure

1.1

The mechanical model

The mechanical model of respiratory mechanics is illustrated in Fig. 2. The idea of using a pair of telescoping canisters was introduced by Bates [1]. In this work, we use a structure of three cylindric telescoping canisters which could slide frictionlessly along each other (Fig. 2). The upside down cannister (CN1) are connected with other cannisters by springs (!! and !!" ). CN1 is opened to a pipe which is connected to the outside environment, so that air can be sucked into the system. The reference pressure of the environment is !! , the pressure at the beginning of the pipe !! . In normal breathing without PEEP, we have !! = ! !! . This model has a clear anatomical analogy to the human respiratory system. The pipe represents the conducting !. airway with flow resistance !!" and inertance The space between CN1 and CN2 represents the alveolar compartment, which can be inflated and deflated in the same way as the lung inspires and expires. The springs (!! and !!" ) represent the elasticity of lung tissue and chest wall. The space between CN2 and CN3 is filled with an enclosed amount of liquid, similar to pleural fluid. Due to

S808

Fig. 2. Mechanical model of lung and chest wall

the incompressibility of liquid, the lung (CN2) and chest wall (CN3) are mechanically coupled. The respiratory muscle is represented with an force !!"# , which applies on the chest wall (CN3). At the beginning of the pipe, we use another springcannister structure to simulate the compliance of the mouth and upper airway !! . The inertance ! and mouth compliance !! are only relevant at high frequency of forced breathing and cause noise in the impedance measurement with oscillation methods. Let !!,! and !!,!" be the natural lengths of !! and !!" , the ‘natural volumes’ of lung and chest wall, their volumes when the lung is separated from the thorax cavity, are !!,! = !!,! . ! and !!,!" = !!,!" . !, with ! the bottom surface of the cannisters. At the respiratory resting position, where lung volume is at FRC level and !!"# = !! = !! , the volumes of lung and chest wall are !!"#,! and !!"#,!" . From physiology, we know that !!,! < !!"#,! !!,!" > !!"#,!" which means !! is stretched and !!" is shrunk at FRC. An equilibrium between two forces occurs ∆!!"#,! + ∆!!"#,!" = 0!!!!!!!!!!!!!!!! !!"#,! −!!,! !!"#,!" −!!,!" +! ! = 0.!!!!!!!!(1) !! !!" In inspiration, !!"# is positive and the spring !!" becomes stretched. The tension generated by the stretched spring generates a negative pleural pressure (!!" ), which applies a stretching of !! . The tension caused by !! overcomes flow resistance !!" and inertance ! of the airway and drives air into the alveoli. At the end of an inspirtion, the muscle relaxes so that !! and !!" contract to their lengths at FRC level. The equation of motion for the respiratory system is given as follows !! ! − !!,! !!"# ! − !! = !!" . ! ! + + !! !!" ! − !!,!" !!!!!!!!!!!!!!!!+! !!" (2) If no pleural effusion occurs, lung volume can be approx-



imated as the volume inside the chest wall !!" ! = !! ! = ! ! , thus eq. (2) becomes 1 1 !!"# ! − !! = !!" . ! ! + +! . (! ! − !! ) !! !!" Replace !! by the relation given in eq. (1), we receive the well-known equation of motion for the respiratory system: 1 !!"# ! − !! = !!" . ! ! + . ! ! − !!"# , !!" where !!" is the respiratory compliance and given as 1 1 1 = +! !.!!!!!!!!!!!!!!!!!!!!!!!!!!!! 3 !!" !! !!"

1.2 Assumptions of the model In this model approach, we focus on the interaction between the lung and chest wall to simulate the pleural pressure. Thus, it should be noted that the respiratory system is very complex and our model obeys several assumptions: • The model is a linear model with lumped elements. It should be noted that the respiratory compliance is nonlinear [4] and should be linearized around FRC. The flow resistances are different between inhalation and exhalation [10] and could be divided into central and peripheral resistances which correspond to the upper airway and bronchial tree. • Though the lung can be ventilated by contraction of the diaphragm and elevation and depression of the ribs, one !!"# and no separation between ribs and diaphragm are considered for respiratory muscles. • The pleural pressure is not uniform along the thorax from dorsal to basal. The gradient is about 0.3 cm H20/cm vertical distance [5]. In this work, we consider no effect caused by the gravitation force.

1.2 The electrical model The mechanical model shown in Fig. 2 represents physiological analogs of the lung characterizing by pressure, flow and volume. However, it’s common practice to use electrical circuits in which the mathematics are also analog, though the variables are different. In electrical models, flow, volume, pressure are replaced respectively by current, charge and voltage. Fig. 3 illustrates the electrical model of the respiratory system. It’s parameters are the same as those in the mechanical model. The springs !! ,!!!" and flow resistance !!" are replaced by electrical capacitors and resistance. A voltage source !!"# represents the respiratory muscles. It should be noted that there are no such ‘natural length’ for electrical capacitors comparing with mechanical springs. Hence, we use a DC voltage source to model the prestress of the elastic lung and chest wall to ensure the relation in eq. (1). Analog to mechanical system, !! is positively and !!" negatively charged at FRC, and

S809

!!",!! !could be derived as !!",!! = ! ∆!!"#,! − ∆!!"#,!" !!!!!!!!!!!!!!!!!!!!!!!!

Fig. 3. Electrical equivalent model of the respiratory system. The model includes the lung, thorax, pleural space, respiratory mucsles and mouth compliance

!!"#,! −!!,! !!"#,!" −!!,!" −! !.!!!!!!!! !! !!" In adults, where !! ≈ !!" ≈ 2!!" , !!",!! becomes ∆!!"#,!" !!",!! = −2∆!!"#,!" = − !!!!!!!!!!!!!!(4) !!" where ∆!!"#,!" = !!"#,!" −!!,!" . Since the unloaded volume of the lung !!,! can not be measured in the practice, it may be adequate to approximate PDC,th through ∆PF RC,cw . According to Guyton, the pleural pressure is -5cmH20 at FRC, so !!",!! !is around +10cmH20. In spontaneous breathing, !! !and ! are neglectable due to the low respiratory rate. Let !!" ! = !!"# ! − !! (!) be the input of the system, the frequency response for the alveolar and pleural pressures are given as follow: !!!" !!" !!"# ! = !!" ! . 1 + !!!" !!" !!",!! !!! 1 + !!!" !! !!" ! = − +!!" ! . . 2 !! 1 + !!!" !!" =

where !!" is given in eq. (3) Note that −

!!",!! !

is a con-

stant term, thus, the correct representation of the pleural pressure is an affine system.

4

Simulation results

The model is implemented in MATLAB Simulink und MATLAB Simscape. Standard values for respiratory compliance in healthy subjects are between 0.07 and 0.1 cmH2O/l/s. In our simulation, we use a standard parameter set taken from [13] and shown in Tab. 1

!!!

!!"

!!

!!"

!

3 0,1 0,2 0,2 0,004 cmH20/l/s l/cmH2O l/cmH2O l/cmH2O cmH20/l/s2 Tab. 1. Parameter set for the respiratory system

In our simulation, we neglect the mouth compliance and estimate the respiratory muscles with a simple sinus signal at respiratory rate at 0.3Hz, an amplitude of 4cmH2O and same inspiratory and expiratory duration. The simulation result for !!"# and !!" are shown in Fig. 4. During a breathing cycles of a healthy object, the alveolar changes from positive (inspiration) to negative (expiration) with a certain phase delay to respiratory muscles. The pleural pressure is negative with a mean value of -



5cmH2O and also changes with the same frequency. Fig.4. Simulation result for P!"# and P!"

The behaviors of alveolar and pleural pressure shown in Fig. 4 are similar to experimental results in some physiology textbooks [3][6][15]. Thus, in our knowledge, our model is the first mathematical model simulating the pleural pressure of the respiratory system by using simple mechanical and electrical approaches.

5

Conclusion

In this paper, we discussed the role of the pleural fluid and pleural space in lung mechanics, especially in lung pathologies. We introduced a novel model of the respiratory system with emphasis on pleural dynamics using a combination of mechanical and fluidic plumped elements. An aquivalent electrical circuits could be applied for a simple estimation of the pleural pressure. The simulation result corresponds to our knowledges in physiology, which shows the model’s high potential in current research fields of lung mechanics.

6

S810

[8] J. Mead, Mechanical properties of lungs, The American Physiological Society 41 (1961), no. 2, 281–330. [9] A. Otis, C.B McKerrow, J. Mead, M.B McIlroy, N.J Selverstone, and E.P Radford, Mechanial factors in distribution of pulmonary ventilation, J Appl Physiol 8 (1956), 427–443. [10] A. Pomprapa, D. Schwaiberger, B. Lachmann, and S. Leonhardt, Amathematical model for carbon dioxide elimination: an insight for tuning mechanical ventilation, European Journal of Applied Physiology 114 (2014), no. 1, 165–175. [11] V.C Rideout (ed.), Mathematical and computer modeling of physiological systems, Prentice Hall advanced reference series. Physical and life sciences, Prentice Hall, Englewood Cliffs and N.J, 1991. [12] E. Starling, On the absorption of fluids from the connective tissue spaces, J. Physiol. 19 (1896), no. 4, 312–326. [13] J. Vogel and U. Smidt, Impuls-oszillometrie: Analyse der atemmechanik in ambulanz und klinik, epidemiologie und experimenteller forschung, Pmi-Verl.-Gruppe, Frankfurt am Main and Moskau and Sennwald and Wien, 1994. [14] E.R Weibel, Morphometry of the human lung, Academic Press, New York, 1963. [15] John B. West, Respiratory physiology: The essentials, 6 ed., Lippincott Williams & Wilkins, Philadelphia, 2000.

References

[1] J.H.T Bates, Lung mechanics: An inverse modeling approach: An inverse modeling approach, Cambridge university Press, Leiden, 2009. [2] A. DuBois, A. Brody, D. Lewis, and B. Burgess, Oscillation mechanics of lungs and chest in man, Journal of Applied Physiology 8 (1956), 587–594. [3] Guyton, Arthur C. Guyton, and John E. Hall, Textbook of medical physiology, 11 ed., Elsevier Saunders, Philadelphia, 2006. [4] Harris R.S, Pressure-volume curves of the respiratory system, RESPIRATORY CARE 50 (2005), no. 1, 78–99. [5] S.J Lai-Fook and J.R Rodarte, Pleural pressure distribution and its relationship to lung volume and interstitial pressure, Journal of Applied Physiology 70 (1991), no. 3, 967–978. [6] F. Lang and P. Lang, Basiswissen physiologie, 2 ed., Springer Lehrbuch, Springer Medizin Verlag Heidelberg, Berlin and Heidelberg, 2007. [7] S. Leonhardt, S. Böhm, and B. Lachmann, Optimierung der Beatmung beim akuten Lungenversagen durch Identifikation physiologischer Kenngrößen, atAutomatisierungstechnik 46 (1998), no. 9, 532–539.



S811

Indicating artificial maternal and fetal pulse waves by capacitive sensing in a phantom D. Laqua, J. Fischer, S. Pollnow, S. Ley, P. Husar, Biosignal Processing Group, Technische Universität Ilmenau, Ilmenau, Germany, [email protected]

Introduction Transabdominal reflective pulse oximetry can be one solution to determine non-invasivly the oxygen saturation of the fetal arterial blood with minimized risk for mother and fetus. For evaluating algorithms to seperate the fetal and maternal pulse waves a new physical-like phantom with two independent artificial blood circuits and different sensors was developed.

Methods The phantom consists of a pre-pressure system with a pressure tank, charged by a roller pump up to 150 kPa, and two artificial vascular systems representing the maternal and fetal blood circuit. The pre-pressure system and both artificial blood circuits are connected through proportional valves, which generate a pulsatile flow. Reproducing pulse waves a chamber filled with air, an elastic silicone tube and a clamp are integrated in the circuit, which simulate the physiological blood circuit. A pressure transducer is located after the elastic tube in each circuit. Additionally, the dilation of the elastic tubes, with a diameter of 5 mm and a wall thickness of 1 mm, are measured simultaneously with a coplanar capacitive sensor and transmission photoplethysmography. The conducting electrode of the coplanar capacitive sensor is connected to a microcontroller. The established oscillating frequency on this electrode changes due to the tube dilation and the subsequent change of the capacity.

Results The capacitive signal and the measured light intensity change significantly due to the generated pulse wave, which produces a tube extension about 15 %. The systole and diastole phase of the pulse wave can be detected in this capacitive signal by a sensor area of 40 x 15 mm.

Conclusion A simple coplanar capacitive sensor can detect the pulsation of an elastic silicon tube. The tube dilation is similar to the optical signal of an artificial pulsating vessel. The capacitive measurement method could be used as alternative for photoplethysmography to detect the pulse wave in test arrangements.



S812

Using an injection signal to reduce the effect of DC voltages and capacitance changes in capacitive biopotential measurements A. Serteyn1, R. Vullings1, M. Meftah, J.W.M. Bergmans1, 1Department of Electrical Engineering, Eindhoven University of Technology, 2Patient Care Solutions group, Philips Research, Eindhoven, The Netherlands, [email protected]

Introduction Capacitive electrodes for biopotential monitoring are a promising alternative to the conventional adhesive electrodes since they provide a biopotential signal, e.g. an electrocardiogram, without the need of direct skin contact. Despite their recent commercialization, they are still highly sensitive to motion artifacts and therefore have a limited reliability. Some of the newest capacitive measurement systems feature an injection signal to detect loose electrodes or measure impedance. Here, we use the injection signal as a basis for artifact reduction. The targeted motion artifact is due to capacitance changes at the body-electrode interface in the presence of a DC voltage across the interface. This type of artifact is always present with motion of the subject [1].

Method First, the injection signal is used to track the variations of the coupling capacitance in real-time. Then, an identification scheme, which was designed based on a model of the system, estimates the artifact and subtracts it from the biopotential signal.

Results The method assumes that the capacitive component of the body-electrode interface dominates the resistive component and shows sufficient capacitance variations in the presence of motion. Within these assumptions, a strong artefact reduction can be achieved in simulation and lab environment. The method was also tested on real-life ECG data measured from the wrist of an adult and provides, in the presence of motion, an estimate for the DC votage across the coupling capacitor.

Conclusion The developed method is an important step towards a proper understanding and removal of motion artifacts at the bodyelectrode interface during online capacitive biopotential measurements. The DC voltage estimation can be used for further studies, including ECG reconstruction [2] or studies of triboelectricity at the body-electrode interface [3]. The method can also be directly used for artifact localization in real-life data and therefore allows for more robust capacitive biopotential measurements. [1] J. Webster, “Interference and motion artefact in biopotentials”, in Region Six Conference Record. IEEE, pp 53-64, 1977. [2] S. Heuer, D. Martinez, S. Fuhrhop, and J. Ottenbacher, “Motion artefact correction for capacitive ECG measurements”, in Biomedical Circuits and Systems Conference. IEEE, pp113-116, 2009. [3] T. Wartzek, T. Lammersen, B. Eilebrecht, M.Walter, and S. Leonhardt, “Triboelectricity in capacitive biopotential measurements”, Biomedical Engineering, IEEE Transactions on, vol. PP, no. 99, p.1, 2010.



S813

Towards camera based extraction of physiological signals for automotive applications Timon Blöcher, Stephan Heuer, Johannes Schneider and Wilhelm Stork FZI Forschungszentrum Informatik, Karlsruhe, Germany, {timonmarius.bloecher, stephan.heuer}@fzi.de Karlsruher Institut für Technologie, Karlsruhe, Germany, {johannes.schneider3, wilhelm.stork}@kit.edu

Introduction In recent years, a growing number of publications deal with Photoplethysmographic Imaging (PPGI) as a means for contactless vital sign measurement, opening a wide field of monitoring and biofeedback applications. For automotive applications both the desire to assess the driver’s state and the growing number of in-vehicle cameras motivate an investigation of PPGI. In this work we have investigated a system for online pulse and breathing rate extraction from RGB videos of the human face as a step towards robust appraisal of psycho-physiological driver states.

Methods Using an off the shelf camera under ambient light conditions we have implemented a system constantly tracking a subject’s face and extracting the raw signal from a region of interest on the forehead area. Several image and signal processing methods are performed, including spatial filtering and Independent Component Analysis. For estimation of the desired parameters we have developed a peak detection algorithm using the signal’s Hilbert transform, which allows an online beat-to-beat heart rate estimation. In order to reduce the influence of motion and light artefacts, signal quality is continuously assessed using three different artefact indicators.

Results In measurements with 8 different subjects we achieved a sensitivity average value of 93,77% for pulse peak detection with our system under resting and constant lighting conditions. First test drives showed good results for pulse rate estimation while driving on highways outside city traffic (𝑅𝑀𝑆𝐸 ≤ 2 𝑏𝑝𝑚 ⁄ ). The results also indicate the need for improvement, especially under inner-city or alternating lighting conditions.

Conclusion The results show that PPGI can be a suitable solution for providing vital sign information in cars. Future work will include heart rate variability estimations based on our beat-to-beat algorithm. However methods to overcome the impact of long and heavy artefact periods need strong focus. For this purpose also fusions with other biomedical sensors are conceivable.



S814

Capacitive ECG in the clinical context M. Oehler, Capical GmbH, Braunschweig, Germany, [email protected] Focus Session: Contactless measurement of biosignals: opportunities and limitations

Introduction Capacitive electrodes can be used for different application areas especially for ECG and heart rate measurements. For ECG measurements in the clinical context, the electrodes and the measurement system has to fulfil all requirements of ECG devices regarding the normative standards.

Methods A 29 channel capacitive ECG system is used to measure the ECG in an array configuration. A flexible surface allows body adaption to the patient’s chest, no preparation is needed. For pre-market evaluation in the field of detection of myocardial infarction, a prospective, observer-blinded and non-inferior trial with 250 ACS patient was performed in comparison to the gold standard 12-channel ECG. Sensitivity, specifity, measurement duration and other technical parameters were investigated.

Results Comparing all available 29 channels of the cECG against the 12-channel ECG for diagnosing STEMI showed a sensitivity of 97% (95% CI, 89%-99%) and a specificity of 60% (95% CI, 51%-68%). The low specificity is mainly caused by the fact, that some patients had ST elevation in the capacitive ECG and not in the 12 channel ECG. A large amount of these patients with available data from catheter intervention had a stenosis of 90% or higher. The measurement time of the 29 channel ECG was significantly reduced compared to the 12 channel ECG (median 40 sec vs. 78 sec).

Conclusion Using 29 capacitive electrodes in an array configuration enables a fast access to the ECG and, compared to the 12 channel ECG, has the opportunity to use the additional channels to improve the diagnosis of myocardial infarction. Further studies will be performed to optimize cut off values for the ST elevation for different localizations of the infarction.



S815

Detection of Movement, Gait & Bio-signals for e-Health & Home Care RJ Prance, M Fatoorechi, H Prance Department of Engineering and Design, University of Sussex, Brighton, UK, [email protected]

Abstract A novel generic capacitive sensing technology, the Electric Potential (EP) Sensor, has been developed at Sussex and applied to activity monitoring and to the recording of life signs and bio-signals of individuals. A small number of static EP sensors, typically four, may be used in a domestic size room for movement, position sensing and basic gait analysis. Additionally, the same technology has been demonstrated, as a remote (seat back sensor), for vital sign monitoring of cardiac function and breathing. The same system may be used to monitor subjects during sleep. When the sensors are placed in physical, but electrically insulated, contact with the body clinical quality bio-signals may be obtained, including ECG, EMG, EOG and EEG.

1

Introduction

The use of a generic capacitive sensor technology, the electric potential (EP) sensor, to e-health and home care applications are described. The electric potential sensor is an extremely high impedance dry electrode technology which and may be used to sense all electrophysiological signals. The sensor may be placed either in close proximity to the surface of the skin or, remotely at a distance from the body under many circumstances. These sensors are capable of measuring spatial potential or charge without resistive electrical contact with the source. The signal is acquired through weak capacitive coupling, which may either take the form of an insulting layer or an air gap. The starting point for EP sensors is to build a high impedance amplifier using either an appropriate operational amplifier or a discrete FET circuit. By applying well known positive feedback techniques such as guarding, neutralisation and bootstrap, which are to be found in laboratory electrometer instruments, the effective input impedance of the amplifier may be increased by many orders of magnitude. Since the requirement here is for a stand-alone sensor without manual adjustment or setup this is not a straightforward procedure. In addition to the above, we require a mechanism which can supply a steady DC input bias current to the sensor in order to create a stable DC operating point. This bias arrangement must be implemented in such a way that it does not compromise the input impedance of the sensor. It is this particular combination of DC bias provision and the careful use of positive feedback techniques which give the electric potential sensor a unique level of performance and operational ease [1]. Data is presented from a four sensor system of static EP sensors in a domestic size space for movement, position sensing and basic gait analysis [2]. The same technology may also be used as a remote (seat back sensor), for vital sign monitoring of cardiac function and breathing [3], applicable to both drowziness and sleep monitoring. When the same sensors are placed in physical, but electrically insulated, contact with the body clinical quality bio-signals may be obtained [4], including ECG,

EMG, EOG and EEG. Examples of two of these are presented here, ECG and EEG event related potential. The relative ease of use and quality of raw data obtained using these sensors makes them ideal for use in the domestic environment either as static sensors or built into furniture or as wearable sensors in clothing. A review of the perceived advantages and disadvantages of dry electrode technologies together with a comparison of these with EP sensors has been conducted previously by one of the authors [5].

2

Method

In order to demonstrate the generic measurement capability of EP sensors four different expeiments were conducted. For the movement and gait data presented four sensors were positioned in the corners of a domestic room sized space with dimensions of 3.5 m x 3.5 m. The sensors were held at a fixed height of 1 m above the floor and the ambient level of the electric field due to the 50 Hz supply monitored. Differential voltage signals, derived from opposing sensors, were signal processed with a correction algorithm [2] to allow for the non-linear response with distance and to produce a target position. It is worth noting that this technique of monitoring the ambient a.c. field enables a static target to be located and tracked and is not depenedent on target motion. However, only a single target within an area may be unambiguously tracked with the current system, implying that it would be best suited to lone occupancy monitoring applications. The remote (off-body) detection of life signs allows both respiration and cardiac function to be observed simultaneously and the data was acquired in an unscreened electromagnetically noisy laboratory [3]. Hardware filters were used, consisting of a low pass (30 Hz) and a notch filter at the supply frequency (50 Hz). However, the main criterion for good signal acquisition in this mode of operation is the elimination of relative motion between the subject and the sensor, since motion artefacts occupy the same region of the frequency domain as these signals. It is therefore envisaged that this system would find use as a seatback sensor for drowsiness monitoring, or as a sleep monitor. De-



convolution of the respiration and cardiac signals is possible in the frequency domain with a simple digital filter. For sensors in physical contact with the body, either on skin or through thin clothing, clinical level electrophysiological results may be achieved [4]. EP sensors with electrically insulated (anodized) electrodes are used in a simple two sensor differential arrangement with no third reference connection required. In cardiology terms the I-lead may be derived from the left arm minus the right arm signal directly, using two wrist mounted sensors held in place using a pair of expanding bracelets of the type in common use as antistatic wrist bands, but could equally well have been built into clothing. The final experimental set-up demonstrates the ability of EP sensors to acquire some of the the smallestelectrophysiological signals usually recorded. Event related potentials resulting from an external stimulus are small (~1 V) EEG signals and are normally measured from the scalp with the use of reference electrodes (either ear lobe or mastoid) and employ signal averaging techniques. For the results presented here two EP sensors were used, one positioned at Pz through hair and the other as a mastoid reference. An oddball paradigm experiment was set up where an “X” and an “O” were displayed on a screen with different occurances (“X” 80% and “O” 20% of the time). The results were averaged for a total of 67 stimuli [6].

3

Results

Figure 1 shows EP sensor movement data, acquired using a four sensor set-up, with basic gait information apparent in the zig-zag nature of the target tracks. The subject in this case walked from position (0,0) to (0,-48) and then returned to (0,0).

S816

An example of an off-body measurement of a combined respiration and cardiac signal is shown in figure 2. This data was acquired at a distance of ~10 cm from the back of a seated subject in an unscreened environment. It is clear from this data that the fast cardiac signal is superimposed on the relatively slow respiration signal and that separation of these in the frequency domain should be possible. It should also be noted that the origin of these signals is likely to be mainly ballistic in nature resulting from chest wall movement.

Figure 2 An example of a combined respiration and cardiac signal acquired at a distance of ~10 cm from the surface of the body in an unscreened environment.

Figure 3 shows a typical example of a contact ECG recorded using two electric potential sensors positioned on the wrists. This is raw data obtained without the requirement for any skin preparation or cleaning. An interesting feature which may be observed is the slow variation in the amplitude of the QRS peak. This correlates with the breathing pattern of the subject. These wrist mounted sensors are suitable for extended ambulatory monitoring applications [4].

Figure 3 An example of a contact ECG recorded using two electric potential sensors positioned on the wrists.

Figure 1 Electric potential sensor movement data showing basic gait information. The subject walked form (0,0) to (0,-48) and then returned to (0,0), a total of 5 steps. A total of 5 steps were taken in each direction, as confirmed by simultaneous video data capture from a ceiling mounted camera. This also confirms that we are detecting the shift in the centre of gravity as seen in the side to side motion associated with each step.

The results of an oddball paradigm experiment are shown in figure 4. Here the signal has been averaged over a total of 67 events to produce the “O” and “X” event related potentials. It is worth noting that free running or spontaneous EEG signals are typically of the order of 10-20 V in amplitude, necessitating the use of signal averaging techniques to render the much smaller event related potentials visible. The duration of the letters on the screen was 100 ms with a 1.5 s gap between successive stimuli. The task the subject is asked to complete is to press the corresponding button on a keyboard when each letter appears.



For the occurances used, 80% “X” and 20% “O”, that equates to 52 averages for the “X” events (trace b) and 15 averages for the “O” events (trace a). The significant result here, apart from the signal to noise being comparable with conventional wet gel electrodes, is that the delay between these two signals is as expected with the latency for the more common event being shorter than that for the less common one.

S817

a wet electrode system and the next stage of this work will be to complete a direct simultaneous comparison with a commercial EEG system. Electric potential sensors are now available as a low power, low cost IC, from our commercial partners [7] in several variants suitable for either movement sensing or biosignals. The bio-compatable electrodes make them suitable for continuous electrophysiological monitoring applications.

Acknowledgements The event related potential work presented here forms part of the DeNeCoR project which is funded by the European ENIAC Joint Undertaking (JU) and the Technology Strategy Board (TSB).

References [1] Prance R.J, Debray A, Clark T.D, Prance H, Nock M, Harland C.J, Clippingdale A.J: J. Meas. Sci. and Technol., 11, 291-297 (2000) [2] Prance H, Watson P, Prance R.J, Beardsmore-Rust S.T: Meas. Sci. Technol., 23 (2012) 115101 [3] Prance R.J, Beardsmore-Rust S.T, Watson P, Harland Figure 4 An example of an averaged EEG (event related C.J, Prance H: Applied Physics Letters, 93 (2008) potential) resulting from an oddball paradigm experiment doi:10.1063/1.2964185 where (a) shows the response to the “O” and (b) to the “X” [4] Harland C.J, Clark T.D, Prance R.J: Meas Sci Tech, events. 14, 923-928 (2003), doi:10.1088/0957-0233/14/7/305 [5] Prance H, “Sensor Developments for Electrophysiological Monitoring in Healthcare,” Applied Biomedi4 Conclusions cal Engineering, Dr. Gaetano Gargiulo (Ed.), InTech, Data has been presented for electric potential sensors to 2011, ISBN: 978-953-307-256-2, Available from: demonstrate the versatility of these devices in the e-health http://www.intechopen.com/books/applied-biomedicaland home care sectors. A four sensor system operating in engineering/sensor-developments-fora 3.5 m x 3.5 m space yielded target position and tracking electrophysiological-monitoring-in-healthcare data as well as basic gait information. The sensors pro[6] Fatoorechi M, Prance H, Prance R.J, 7th UK and RI duce low data rates (typical operating bandwidth ~10 Hz Postgraduate Conference in Biomedical Engineering in this application) compared with video based movement and Medical Physics, Guildford, Surrey, UK, 9th-11th systems and so incur a commensurately small data proJuly 2013 cessing overhead, making them ideal for unattended lone [7] Plessey Semiconductors Ltd: Plymouth, UK, occupancy monitoring systems. http://www.plesseysemiconductors.com/epic-plesseyAt smaller subject to sensor spacing (~10-40 cm) life sign semicon ductors.php information may be obtained, with the proviso that the relative motion between the subject and sensor is carefully controlled. Both respiration and cardiac signals were demonstrated at a 10 cm spacing and this should prove to be a viable system for seatback drowsiness or sleep monitoring applications where movement is intrinsically restricted. Contact electrophysiological data has been demonstrated to produce comparable signal to noise to conventional wet gel electrode systems and offers advantages over other dry electrode technologies. ECG and preliminary event related potential (ERP) data were presented. The ERP data is of particular significance since these are among the smallest electrophysiological signals regularly recorded. The results obtained are in line with those expected from



S818

Active humidification for capacitive-resistive ECG-systems Lennart Leicht, Benjamin Eilebrecht, Sören Weyer, Tobias Wartzek, Steffen Leonhardt Philips Chair for Medical Information Technology, RWTH Aachen, Aachen, Germany [email protected]

Abstract Capacitive ECG measurement systems have been moving in the research spotlight in the past years. Due to their ability to measure ECG signals without direct contact to the body surface, they are well-suited for monitoring applications in personal healthcare scenarios. However, signal quality is insufficient for diagnostic purposes. Main causes are motion artefacts and triboelectricity. In this work, we propose active humidification using water vapour as a measure to improve signal quality. After a brief theoretical analysis and a proof of concept, we present a prototype of a capacitive ECG seat implementing the concept. Using this prototype, we obtained experimental proof that active humidification can in fact improve cECG signal quality.

1

Introduction

In 1967, Richardson first described the derivation of an electrocardiogram (ECG) using insulated, that is capacitive electrodes [1]. Signal acquisition with this electrode type does not require a conductive (galvanic) connection to the patient under examination, since ECG recording is accomplished by means of an electrical field: The patient’s body surface and the electrode form a structure similar to a parallel-plate capacitor, thus by electrostatic induction, potentials present on the body surface can also be measured at the electrode. Using a field based, non-contact measurement principle, no electrodes have to be attached to the skin (which can be irritating) and the ECG can be recorded even through layers of clothing. Due to this benefits, during the past years this electrode type has come into research spotlight: Capacitive ECG measurement systems have been integrated in several items of daily living, for example a toilet seat [2], a bathtub [3], a car seat [4] and a tablet-PC based system that can record body surface potential maps [5]. Also, a clinical proof of practicability of a cECG device has recently been shown [6]. However, cECG measurement systems require carefully designed ultra-high input impedance amplifier circuits with additional bias current compensation [7]. Recently, a cECG system has been proposed which does not require a Driven Right Leg Electrode [8]. Still, signal quality of cECG systems is inferior to conventional, conductive ECG systems; in particular in environments with a high level of motion between electrode and body surface, such as while driving on a bumpy road [9]. The causes are for one a bad and time variable capacitive coupling between electrode and body surface causing so called motion artefacts [10]. Due to their randomness, motion artefacts are difficult to separate from the cECG signal using conventional techniques such as filtering, but stochastic approaches such as Blind Source Separation (BSS) have shown promising results [11]. In addition to motion artefacts, electrostatic charge gen-

eration can occur on all interfaces between electrode, clothing and skin, for example between a wool sweater and the electrode. This triboelectricity can easily reach magnitudes in the kilovolt range [12], masking the weak ECG signal, which is in the millivolt range. Providing a discharge path that can remove triboelectric charges using an actively driven grid (as analysed in [12]) can only remove charges on the surface of the clothing next to the electrode. Charges generated inside the fabric or another clothing layer flow only slowly to the grid as dry fabric has a very high impedance (usually Gigaohm range), so the effect of the grid is limited. Here, we propose a different approach that can improve the signal quality in cECG systems by addressing both the poor capacitive coupling between electrode and body surface as well as triboelectric charge generation at the interfaces.

2

Methods

2.1

Analysis

Figure 1 shows a simplified equivalent circuit diagram of a cECG measurement setup, including the ECG signal on the body surface (represented by a voltage source), the capacitive interface (represented by a RC network to account for the capacitive coupling (Cc ) with a high impedance residual resistive component (Rc )) and an operational amplifier as the first stage of an ECG signal processing chain [13]. Cable capacitance and parallel input capacitance of the amplifier are accounted by a parasitic capacitor Cin . Rin represents the amplifier’s input resistance and the parallel bias resistor commonly used for bias current compensation [7]. The complex transfer function of the setup can be calculated according to equation 1 as H(j!) =

Rin . in Cin Rin + Rc · 1+j!R 1+j!Rc Cc


(1)

Cc +

Uecg

Cin

ECGout

Rin

Figure 1 Simplified equivalent circuit, adapted from [13].

Evaluating the equation shows that small coupling resistances in combination with very high input amplifier resistances are beneficial as in this case the transfer function becomes real valued and independent of frequency. The measurement setup then turns into a common galvanic ECG measurement in which coupling capacities are neglectable due to the small coupling resistance. Therefore, due to the improved transfer function, a better ECG signal quality can be expected. In prior experiments using an ECG system integrated into a car seat we observed an improving signal quality after the test person had been sitting on the seat for several minutes. The effect seemed to be better the more the test person was sweating. Based on this empirical finding, we formulated the thesis that humidity provided by the test person while sweating might condensate inside the clothing and lead to a reduced fabric impedance. This will reduce the coupling resistance. The cECG then changes into a capacitive-resistive ECG measurement system. We also expect that a reduced fabric impedance facilitates the discharge of local triboelectric charges using current paths inside the fabric or via the reduced coupling resistance to the electrode. To benefit from the effect of condensing humidity in measurement situations with little or no transpiration, we propose an implementation of a cECG measurement system that can actively dispense small quantities of water vapour to its electrodes and the clothing of a test person. The practicability and effectiveness of such a system was evaluated in this paper.

2.2

Proof of concept

According to our thesis, condensing humidity can reduce fabric impedances. For verification, we exhibited a small piece (length: 152 mm, width: 40 mm) of cotton fabric to water vapour from a bowl of hot water and measured the fabric impedance on a length of 28 mm for 14 minutes using a Agilent E4980A Precision LCR Meter (Agilent Technologies, Santa Clara, USA). Every minute, a measurement was taken. The results are shown in figures 2 and 3. During the experiment, the measured impedance significantly diminished from 760 M⌦ \ -89 to 23 M⌦ \ 0 . Hence, the capacitive coupling was transformed to a low impedance resistive coupling, which proves the thesis. After the experiment, the fabric felt moistly, but not wet.

S819

800 600 400 200 0

0

5

10

15

Time (min) Figure 2 Resistance development during active humidification. 0 Phase (degrees)

Rc

Magnitude of resistance (M⌦)


20 40 60 80 0

5

10

15

Time (min) Figure 3 Phase development during active humidification.

2.3

Experimental setup

To investigate the effect of an electrode with active humidification on ECG signal quality, we constructed a prototype. In our experimental setup, we used an ordinary car seat in which two electrodes made of a flexible, conductive fabric were integrated as shown in figure 4. Behind the electrodes a device that can release humidity as water vapour was attached. The electrode material is water permeable, so humidity can pass through the electrode into the clothing of a user sitting on the car seat as shown in figure 5. For ECG derivation, we used a common analog ECG interface. The signals from the two textile electrodes are directly fed into the inputs of an INA116 instrumentation amplifier (Texas Instruments Incorporated, Dallas, USA) [7], processed in a signal processing chain (lowpass-, notch-, highpass filter and amplifier) and are afterwards digitised. In order not to diminish the input impedance Ri for best transfer function performance, no bias resistor for bias current compensation was used as proposed by [7]. To reject common mode disturbances, a Driven Right Leg Circuit was used. The digitised signals were digitally filtered (lowpass, notch, highpass) using Matlab (The Mathworks Inc., Natick, USA).



S820

Figure 4 Car seat with two large electrodes made of conductive fabric in the dorsal region.

Reference ECG [V]

ECG signal can be observed. In contrast, in the sequence recorded with humidified electrodes, signal levels are smaller, but a typical ECG signal can be observed which is well aligned to the reference measurement. All relevant parts of an ECG cycle (P-wave, QRS-complex, T-wave [14]) are visible. Even so, one larger artefact is present in the signal, which was recorded while the test subject took a deep breath.

1 0.5 0 0

2

4

6

8

10

8

10

Time (sec) cECG [V]

1 0 1 0 Figure 5 Principle of water vapour dispensing at the electrodes.

4

6

Time (sec) Figure 6 Signal sequence recorded with dry electrodes.

In the experiment two healthy male test subjects were enrolled. To create a realistic measurement scenario, both were wearing two layers of own clothing, a cotton T-Shirt and a cotton sweater. For each test subject, we first recorded 5 min of ECG data using the measurement setup with deactivated humidification device to acquire a dry reference measurement. Then, the device was activated and after 5 min given for humidification, we recorded another 5 min sequence. After each test subject, the whole system was dried to remove residual humidity, so a dry reference measurement was possible for the next test subject. During all measurements, we recorded a conventional 3-lead ECG using a Philips MP30 patient monitor and adhesive Ag/AgCl electrodes as reference ECG.

Reference ECG [V]

Experimental trial 1 0.5 0 0

2

4

6

8

10

8

10

Time (sec) cECG [V]

2.4

2

0.1 0 0.1

3

Results

Figure 6 shows a 10 s signal sequence recorded using dry electrodes, figure 7 a 10 s signal sequence using humidified electrodes, respectively. In both figures, the upper graph shows the reference measurement, the lower graph the capacitive measurement. Qualitative data analysis reveals that during the dry measurement, the derived cECG signal is very erratic. The signal shows quick changes with high amplitudes that are not aligned with the R-peaks in the reference ECG. Overall, signal quality is very poor and no clear

0

2

4

6

Time (sec) Figure 7 Signal sequence recorded with humidified electrodes. For better comparison, the amplitude of the strongest signal part of an ECG, the R-peak was measured and the amplitude of the residual noise floor for both measurement scenarios was estimated. To derive a quantitative measurement for signal quality, the signal to noise ratio (SNR) between



R-peak amplitude and noise floor was calculated according to equation 2: SN R =

UR peak UN oisef loor

(2)

The measured amplitudes and calculated SNR are shown in table 1. R-Peak [V] Noise Floor [V] Derived Q-factor

Dry electrode 1.2 1.2 1

Wet electrode 0.14 0.04 3.5

Table 1 Derived quality markers As can be seen, the ratio between R-peak and amplitude and residual noise floor could be improved by a factor of 3.5 using active humidification. Therefore, a qualitative and quantitative improvement was observed.

4

Conclusion and Outlook

S821

[4] Marian Walter, Benjamin Eilebrecht, Tobias Wartzek, and Steffen Leonhardt. The smart car seat: personalized monitoring of vital signs in automotive applications. Personal Ubiquitous Comput., 15:707–715, October 2011. [5] Martin Oehler. Kapazitive Elektroden zur Messung bioelektrischer Signale. PhD thesis, Technische Universität Carolo-Wilhelmina zu Braunschweig, 2009. [6] M. Czaplik, B. Eilebrecht, R. Walocha, P. Schauerte, and R. Rossaint. Clinical proof of practicability of a contactless ecg device. Euroanaesthesia 2010, Helsinki, Finland, June 12th - 15th, 2010. [7] R J Prance, A Debray, T D Clark, H Prance, M Nock, C J Harland, and A J Clippingdale. An ultralow-noise electrical-potential probe for human-body scanning. Measurement Science and Technology, 11(3):291, 2000. [8] T. Komensky, M. Jurcisin, K. Ruman, O. Kovac, D. Laqua, and P. Husar. Ultra-wearable capacitive coupled and common electrode-free ecg monitoring system. In Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE, pages 1594–1597, Aug 2012.

In this work, a new concept for cECG signal quality en- [9] T. Wartzek, B. Eilebrecht, J. Lem, H.J. Lindner, hancement using active humidification was proposed. To S. Leonhardt, and M. Walter. Ecg on the road: Roevaluate the concept, a seat integrated cECG measurement bust and unobtrusive estimation of heart rate. Biomedsystem was constructed. For two test subjects, the effectiveical Engineering, IEEE Transactions on, 58(11):3112 ness of the concept was successfully demonstrated. –3120, nov. 2011. In future work, cECG data from a larger number of test [10] Jörg Ottenbacher and Stephan Heuer. Motion artesubjects wearing different types of clothing should be colfacts in capacitively coupled ecg electrodes. In Olaf lected to gather statistically relevant data. Dössel, Wolfgang C. Schlegel, and Ratko Magjarevic, editors, World Congress on Medical Physics and Biomedical Engineering, September 7 - 12, 2009, Munich, Germany, volume 25/4 of IFMBE Proceedings, pages 1059–1062. Springer Berlin Heidelberg, 2010. 10.1007/978-3-642-03882-2_282.

5

References

[1] P. Richardson. The insulated electrode. In Proceedings of the 20th Annual Conference on Engineeringin Medicine and Biology, Boston, MA, USA, volume 157, 1967.

[11] Aline Serteyn, Xintan Lin, and Oliver Amft. Reducing motion artifacts for robust qrs detection in capacitive sensor arrays. In ISABEL 2011: Proceedings of the 4th International Symposium on Applied Sciences in Biomedical and Communication Technologies. ACM, ACM, 2011.

[12] T. Wartzek, T. Lammersen, B. Eilebrecht, M. Walter, and S. Leonhardt. Triboelectricity in capacitive biopo[2] Hyun Jae Baek, Jung Soo Kim, Ko Keun Kim, and tential measurements. Biomedical Engineering, IEEE Kwang Suk Park. System for unconstrained ecg meaTransactions on, 58(5):1268 –1277, May 2011. surement on a toilet seat using capacitive coupled [13] Enrique Spinelli and Marcelo Haberman. Insulating electrodes: The efficacy and practicality. 2009. electrodes: a review on biopotential front ends for dielectric skin electrode interfaces. Physiological Mea[3] Yong Kyu Lim, Ko Keun Kim, and Kwang Suk Park. surement, 31(10):S183, 2010. The ecg measurement in the bathtub using the insulated electrodes. In Engineering in Medicine and Bi- [14] Stefan Silbernagl and Agamemnon Despopoulos, ediology Society, 2004. IEMBS ’04. 26th Annual Internators. Taschenbuch der Physiologie. Thieme Verlag, 6. tional Conference of the IEEE, volume 1, pages 2383 korrigierte auflage edition, 2003. –2385, sept. 2004.



S822

BIOHYBRID nerve transplants - development of chitosan-based nerve grafts Kirsten Haastert-Talini1,2, Stefano Geuna3, Abraham Shahar 4, Thomas Freier 4, Claudia Grothe1,2 and the BIOHYBRID Consortium 1 Hannover Medical School, Institute of Neuroanatomy, Hannover, Germany 2 Center for Systems Neurosciences (ZSN) Hannover, Hannover, Germany 3 Department of Clinical and Biological Sciences, and Cavalieri Ottolenghi Neuroscience Institute, University of Turin, Turin, Italy 4 N.V.R. Research Ltd, Nes-Ziona, Israel 5 Medovent GmbH, Mainz, Germany

Introduction: T h e E u r o p e a n r e s e a r c h consortium BIOHYBRID is working on the development of chitosan-based artificial nerve grafts. Hollow tubes were manufactured out of medical grade chitosan by a proprietary extrusion process followed by adjustment of different degrees of acetylation (DA): DAI (~2%), DAII (~5%), DAIII (~20%). In a first comprehensive and multidisciplinary attempt i n v i v o evaluations revealed that early regeneration processes and subsequently functional and structural regeneration was most similar to autologous nerve graft reconstruction when DAII chitosan tubes were used (Haastert-Talini et al., Biomaterials 2013, Dec;34(38):9886-904). Fine-tuned chitosan tubes with a DA of ~5% (ReaxonTM C h i t o s a n N e r v e G u i d e ) a r e currently processed for the development of more complex artificial nerve devices with the aim to make reconstruction of longer nerve defects addressable. Genetically modified Schwann cells over-expressing neurotrophic factors and seeded into a regenerative hydrogel, NVR-Gel, sufficiently support neurite outgrowth from dorsal root ganglion neurons in vitro. Methods: Reconstruction of 15mm rat sciatic nerve gaps was performed with (A) autologous nerve transplants (ANT, n= 7) or chitosan nerve guides containing NVR-Gel seeded with (B) non-transfected SCs (n= 7), (C) empty vector control transfceted SCs (n=7), (D) GDNF over-expressingSCs (n=7) or (E) FGF-218kDa over18kDa

functional recovery was monitored by regular assessment of the von-Freypain threshold-test, regular evaluation of the static sciatic index and repeated electrophysiological assessment. Results: Preliminary results demonstrate that the ANT is still anticipating recovery of nerve functions over artificial nerve grafting. However, after one month delay events of functional recovery with values comparable to ANT group animals were also detectable in animals of the experimental groups. 4 months after surgery the group that received FGF-218kDa-SC-grafts was the best recovered group among the enriched chitosan nerve guide reconstructed animals. Histomorphometrical evaluation of the regenerated nerve tissue is currently still in progress.

This work was supported by the European Seventh Framework Community's Programme (FP7-HEALTH-2011) under grant agreement n°278612. The certified medical grade chitosan used for all tests was supplied by Altakitin S.A. (Lisboa, Portugal).



S823

Stem cells for personalized medicine In vitro and in silica models Jari Hyttinen, Department of Electronics and Communication Engineering, Tampere University of Technology, Tampere, Finland and BioMEdiTech (www.biomeditech.fi), Tampere, Finland In this paper the application of stem cell derived human cells especially for personalized medicine are discussed. Moreover the technological development for in vitro and in silica needed for the applications are introduced. Human embryonic (hESC) and recently human induced pluripotent stem cells (hIPSC) provide means to produce most human cell types. hiPSC, that can be derived from patient own cells such as skin cells, provide means to get patient own cells for in vitro disease models and for drug testing thus they offer an important tool for future personalized medicine. In addition to novel stem cell based human patient cells a number of engineering technologies need to developed or optimized that will either enable or augment the use of the stem cell based cells in personalized medicine. The list of technologies needed include the biomaterials to grow the cells, automated culturing and microfluidic environment, imaging and measurement system to analyse the cell function and computational tools to model the functions of the derived cells. For example in vitro analysed patient specific cardiomyocytes (CM) can provide novel insight of various gene based cardiac diseases such as various forms of long QT syndrome. More importantly they offer drug model where the candidate drugs can be tested on patients own CMs. We have shown that the hIPSC derived CMs can be monitored in vitro by various methods such as electrophysiological as well as image based methods. Further, we have developed the first patient specific in silica models of the hIPSC CMs providing us ion channel level models of these CMs electrophysiology. These models are developed also for hIPSC CMs with specific ion channel mutations in long QT syndrome showing that the pathological condition observed in cardiac scale is seen also in vitro and can be modelled in silica. These kinds of platforms are in development in addition to CMs also for neuronal and retinal cells. The future use of stem cell technology for personalized medicine will need a number of various engineering expertise. The computational modelling has its role on understanding the causes of behaviour initiated by the gene mutations as well as for pre screening of compounds for drug optimization.



S824

High voltage alginate encapsulation of monkey stem cells for application in regenerative medicine Sasha Gryshkov, Birgit Glasmacher Institute for Multiphase Processes, Leibniz Universitaet Hannover, Hannover, Germany [email protected] Hypothesis. High voltage alginate encapsulation is advantageous over the commonly used air flow encapsulation method. High voltage does not affect the proliferation and viability of mesenchymal stem cells encapsulated and cryopreserved in alginate beads. Here we present the results on the effect of high voltage on the proliferation and viability of mesenchymal stem cells (MSCs) derived from a Common Marmoset, encapsulated in alginate beads under high voltage and cryopreserved under slow cooling protocol. The results on optimization of process parameters revealed that high voltage alginate encapsulation is advantageous as compared to traditional air-flow encapsulation. The proliferation and viability of encapsulated MSCs was not altered by high voltage. The cryopreserved encapsulated MSCs possessed high proliferation rate after thawing. Taken together, high voltage alginate encapsulation is perspective for further application in cellbased therapies, also in non-human primate model which is necessary for further translation of this approach to clinical trials.



S825

Medical Device Regulations – Potential impact on patient care and biomedical research M. Imhoff, Abteilung für Medizinische Informatik, Biometrie und Epidemiologie, Ruhr-Universität Bochum, D44780 Bochum, Germany, [email protected]

Introduction An often posed argument in the discussion of regulatory approval of devices is that tighter regulations may limit patient access to modern medical device technology and respective advances in diagnostics and therapy. Moreover, tight regulations may stifle biomedical research due to limited opportunities and increased cost of commercialization. A comparison between the perceived two poles of the regulatory spectrum, the EU MDD and the FDA, may help to better understand these challenges.

Technologies not available for patient care As one example, target controlled infusion (TCI) pumps are not available in the US but considered a standard of anesthesia care in Europe where they were approved first. While it is unlikely that these pumps will be approved in the US anytime soon, there is also not conclusive evidence that they improve patient outcomes.

Technologies with delayed availability Many medical devices get approved in the EU first, but later will also receive FDA approval. This would imply that the more time-consuming and more expensive FDA pathway does not result in better patient safety or better effectiveness of the respective devices, but only in much later access to such technologies for US patients.

Impact on biomedical research The current MDD does not consider “investigational devices” as a separate entity, while the FDA offers the “investigational device exemption” (IDE) as a specific pathway for the use of non-approved, newly developed medical devices in clinical study. In that respect the latter may be considered advantageous for biomedical research. On the other hand, funding for biomedical research is directly and indirectly dependent on the commercialization of research results which may be negatively impacted by longer and more expensive regulatory pathways. This may be evidenced by the fact that an increasing number of medical devices developed in the US will not first be submitted for approval there but in the EU. Moreover, investors increasingly take potential regulatory hurdles into consideration when deciding on investments into biomedical technology.

Conclusions Tighter, more time-consuming and expensive regulatory pathways will without doubt impact patient care and the availability of new medical device technologies for the individual patient. On the other hand, an impact on patient outcomes may be difficult or impossible to show. While tighter regulatory approval for market access may not affect clinical research with medical devices, it will most likely have negative long-term impact on funding for biomedical research. The impact on patient care and biomedical research from regulatory directives is complex and requires careful consideration.



S826

Comparison between different regulatory systems K. Neuder, VDE (DKE), Frankfurt/M., Germany, [email protected]

Introduction Europe and the United States have different philosophies for the regulation of Medical Devices. In the United States of Amerca (US) the FDA is responsible and this organisation is authorized by the government. In Europe the basis for application CE-mark based on the discussion between manufacturer and Notified Bodies. This presentation seeks to provide an overview of the different regulatory systems.

Results The two different regulatory systems for Medical Devices in Europe and the United States of America have different levels with regard to the “authority”. The more flexible regulatory system in Europe (between manufacturer and Notfied Bodies) may be an advantage for patient to get innovative medical earlier than the system in US. On the other hand the guidance for manufacturer in the regulatory system of US is better supported than the regulatory system in Europe. However, the next final step will be the free-trade-zone between US and Europe and the outcome which regulatory system will be supported (both or only one?) in the future is open. This may lead to more harmonisations between the two regulatory systems.

Conclusion In summary, the different regulatory systems may lead to different “time-to-market” results for the manufacturer of Medical Devices. The more “governmental-guided” regulatory system in US are less in criticism than the system in Europe.



S827

Proposed changes of the MDD and related regulations T. Prinz, German Society for Biomedical Engineering within VDE (DGBMT), Frankfurt/M., Germany, [email protected]

Introduction Following the scandal of defective breast implants (PIP), the European Parliament asked the Commission “to develop an adequate legal framework to guarantee the safety of breast implants and medical technology in general”. This presentation seeks to provide an overview of the Commission proposals for a regulation of the European Parliament and of the Council on medical devices and in-vitro medical devices.

Results The replacement of three directives by two regulations can be seen as an important step towards regulating the medical devices sector in a more simple way. As a matter of fact the proposed documents would apply as-is to all Member States. Overall the proposed changes aim to improve the medical device security in a life-cycle approach. Among others the proposals are focussing on tightening the regulation scopes, streamlining the conformity assessment procedures, improving the tracebility, introducing a qualified person, monitoring the Notified Bodies (NB) and at the same time changing the role of NB from a business partner to an executive arm of the Competent Authorities, e.g. by conducting unannounced inspections. Finally, the installation of a regulatory body called the Medical Device Coordination Group (MDCG) is attempring to support the cooperation between Member States and to strengthen the acting power of the Commission.

Conclusion In summary, the Commission proposals contain more details and guidance documents such as MEDEVs are incooperated to a larger extend.



S828

Safe reprocessing of medical devices with a view of the entire process chain - the guideline VDI 5700 Prof. Dr. med. Clemens Bulitta Leiter Studiengang Medizintechnik Lehrgebiet Diagnostische Systeme und Medizintechnik Management Ostbayerische Technische Hochschule (OTH)

Introduction The reprocessing of low-germ or sterile used medical devices is even potentially risky, although the aim of reprocessing is the avoiding of hygienic or technically functional risks. The methodological principles of risk management for medical devices are described in the standard DIN EN ISO 14971. The KRINKO-/BfArM-recommendation "Hygiene requirements for the reprocessing of medical devices" clarify numerous reprocessing-specific risks and are structured with reference to the different steps of reprocessing. A practical combination of the normative risk management methodology with the process-oriented KRINKO-/BfArM-recommendation was the aim, which has provided an interdisciplinary group of experts moderated by the Association of German Engineers (VDI). The main contents of the guideline VDI 5700 "Hazards associated with the reprocessing - Risk Management in the reprocessing of medical devices - Measures for risk control" and the process of the development of this guideline is described. Keywords: medical devices, reprocessing, hazards, risks, risk reduction



S829

Methods and first results for assessment of microbiologic burden of non sterile used medical devices S. Buhl1, B. Rußwurm1 and C. Bulitta1 Abstract The risks of nosocomial infections have a significant impact on postoperative recovery and hospital hygiene in general. Publications from the Robert Koch Institute (RKI) show an amount of approx. 225.000 patients in Germany that contracted a surgical site infection in 2008 (Epidemiologisches Bulletin Nr.36; 13.09.2010). Because of the increasing quantity of multiresistant germs (MRG) follow up care becomes more challenging. Therefore the evaluation of infection prevention moves more and more into focus since the last years. Important areas of this research are ventilation technologies, antimicrobial surfaces and new developments on routine testing for proper hygienic cleaning and disinfection. Especially in the case of screenings on clinical hygiene the “Kommission für Krankenhaushygiene und Infektionsprävention” (KRINKO) of the RKI gives only case specific recommendations (Bundesgesundheitsbl Gesundheitsforsch Gesundheitsschutz 5 2001). One first approach to establish easy and cost effective general testing for proper hygiene in daily clinical routine are the adenosintriphosphate (ATP) based handheld devices. ATP is the molecular base for many energetic processes and can be found in every cellular organism. Hence these devices are able to reliably detect organic residues in critical areas in the timespan of few minutes. This work gives a first overview of the bacterial burden in a medical facility on the basis of an ATP based testing method. In this way general hygienic issues are uncovered by swabbing of surfaces and devices pre and post cleaning and disinfection. Potential solutions are suggested and discussed. 1

Studiengang Medizintechnik Ostbayerische Technische Hochschule (OTH) Amberg Weiden Hetzenrichter Weg 15 92637 Weiden i. d. OPf.



S830

Abstract The Hybrid OR: Example for challenges of “Medical Engineering meets Hygiene”

S. Buhl1, B. Rußwurm1 and C. Bulitta1 New treatment options using intraoperative imaging, mainly in Heart and Vascular Surgery drive the concept of hybrid OR. These ORs are on the rise integrating MRI and CT technology as well as in traoperative angiography, by far the most frequent intraoperatively used technology. The latter has almost become a standard especially in endovascular surgery. Nevertheless, the implementation of these devices still mandates a smooth clinical workflow accommodating the hygiene standards of surgical care. Therefore a variety of requirements for clinical procedures, hygiene, medical equip ment and building services must be identified and observed. One challenge is proper and validated cleaning and disinfection of technical equipment. For non sterile used medical devices there are few regulations regarding cleaning and disinfection procedures. In contrast normative standards for in traoperative cleaning and disinfection procedures are well defined for sterile medical devices. The existing recommendations of the “Kommission für Krankenhaushygiene und Infektionsprävention” (KRINKO) are therefore difficult to follow for non sterile used devices. Subject of this paper is to illus trate and discuss these challenges with examples from hybrid OR installations as well as to present suggestions for measures to prevent surgical site infections.

1

Studiengang Medizintechnik Ostbayerische Technische Hochschule (OTH) Amberg Weiden Hetzenrichter Weg 15 92637 Weiden i. d. OPf.

Mail: c.bulitta@oth aw.de



S831

Abstract BMT 2014: Hygiene meets medical engineering: The industry´s perspective Ernst W. Schubert1

Hygienic reprocessing of medical devices provides multiple challenges for medical device manufacturers who are selling products globally. The disinfectant markets are dominated by locally acting chemical companies and national approvals. Globally accepted standards e.g. DIN EN ISO 17664 may be a help for device manufacturers to reduce complexity. Medical workplaces in ICUs and ORs are individually arranged for each hospital unit and need to be flexibly adaptable to the acuity level of patients. Setups with devices from different suppliers and missing standardization lead to workplaces that are far from optimal for infection control.

1

Dräger Medical GmbH

Moislinger Allee 53 55 23558 Lübeck, Germany Email: [email protected]



S832

The oxygen content of the cerebral efferent vessels – First steps to a sensor design K. Rackebrandt1,2,3, H. Gehring2 1 Institute of Medical Engineering, Universität zu Lübeck, Luebeck, Germany, [email protected] 2 Departement of Anesthesiology and Intensive Care, University Medical Center, Luebeck, Germany 3 Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Germany

Abstract The oxygen content of the venous blood in the cerebral efferent vessels is a physiological parameter for the assessment of the perfusion, metabolism and the oxygenation of the brain. There is no near infrared spectroscopy (NIRS) system available to measure this parameter non-invasively. We present a phantom based on a three layer model to simulate the anatomical target region and made a first attempt to measure concentration changes inside a measurement cell, representing a vessel. Our sensor consisted of one LED with the wavelength of 850 nm in combination with two photo diodes to detect the reflective signals. The inflow and outflow of Intralipid (IL) and Indocyanine green (ICG) in different concentrations could be detected inside of the cell. The results illustrate the capability to measure the target values in a specific vessel, represented by the measurement cell. Based on these first measurments a next generation sensor was designed which is geometrically aligned with the three layer phantom model and provides the opportunity to investigate several wavelengths and photodiode distances.

1

Introduction

The tissue oxygenation defined by the oxygen extraction ratio ( O2 ER ) (1) depends on the rate of delivered oxygen ( DO ) and the rate of consumed ( VO ) oxygen. 2

2

O 2 ER

VO2

(1)

D O2

The oxygen delivery (2) is the product of the cardiac output (CO, flow in [ml / min]) and the oxygen content of the arterial blood ( CaO2 in [ml / min]). DO2

CO Ca O2

(2)

The oxygen content of the arterial blood is defined as, CaO2

k1 Hb SaO2

k2 PaO2

(3)

where k1 is the Hüfner´s constant, tHb is the hemoglobin concentration, SaO is the arterial oxygen saturation, k2 is 2

blood inside of the sigmoid sinus (Layer 2) and the brain tissue (Layer 3) in the depth. Through the biological variability the thickness of the scalp varies from 4 - 8 mm and the skull from 4 – 10 mm [3, 4]. NIRS is a non-invasive method to determine blood properties. There is a wide field of application for this technique already, e.g. the cerebral oximetry. Scattering and absorption processes are basically the reason for the way of the light propagation through biological tissue. The optical properties of the tissue correspond with the amount of reflected and transmitted light. For a reflective NIRS system the irradiated light can be expected to follow a curved path through the head [5]. The penetration depth of the light corresponds with the wavelength as well as the source and detector separation. The optical properties of biological tissue are mainly characterized by the absorption coefficient µa [1/mm], the reduced scattering coefficient µs´ [1/mm] and the anisotropy g (g = ). There is a diagnostic window from 700 nm – 1000 nm where light can penetrate the scalp and the skull [6].

the solubility coefficient of oxygen and PaO is the arteri2 al partial pressure of oxygen. The oxygen consumption (4) describes the amount of consumed oxygen per minute by the tissue, VO2

CO (CaO2

CvO2 )

(4)

where CvO is the venous oxygen content [1]. 2 As the O2 ER is a variable to assess the perfusion and metabolism of the brain, it is the intention to measure the oxygen inside a cerebral efferent vessels as a first step towards this objective. The deep veins in the target region are visualized in Image 1. The anatomical target region consists of three layers, the scalp & skull (Layer 1), the

Image 1 Overview of the deep veins in the target region [2]



2.3

2

Methods

2.1

Phantom Model

A liquid phantom, representing the combined scalp and skull layer was designed in the lab. The diffuse reflectance, the total transmission and the collimated transmission were measured with a double integrating sphere setup [11] and the optical properties were calculated using the Inverse-Adding-Doubling method (IAD) [8] and the Kubelka-Munk-Method (KMM) [9]. The phantom consists of distilled water and about 3% Intralipid 20 % (Baxter Deutschland GmbH). The properties of the phantom can be manufactured reproducible in the lab setup (Table I). The calculated properties are within the variation extracted from the literature. Table 1 Calculated optical properties of the liquid scalp & skull phantom 850 nm Literature[7] 0.019 µa [1/mm] 0.86 µs´ [1/mm] 0.89 g

2.2

KMM 0.03 0.851 0.87

IAD 0.029 0.87 0.86

S833

Hardware and software setup

The hardware for the reflective sensor can be divided into a sending and a receiving unit. The former was equipped with an LED with a wavelength of 850 nm and a power of 3 mW driven in continuous mode. The receiving unit consists of 2 photo diodes (PD) with a peak sensitivity at 850 nm. The sampling rate was 1 kHz and the signals were acquired with a data acquisition card (USB 6259, National Instruments, TX, USA). A low pass filter (5th order Sallen-Key) with a cut-off frequency of 30 Hz was implemented to avoid any disturbance coming from the power supply. Other filters were not attached to the electronic circuit to avoid losing any part of the signal from the target region. The board was operated with 5 V DC. The software was fully implemented in LabView (National Instruments, TX, USA). The light source could be controlled and the data from the PDs were recorded with a given timestamp. A second low-pass filter with a cut-off frequency of 5 Hz was implemented (5th order Butterworth) in the software, because the venous flow signal has a frequency of 1-3 Hz. The light source was operated with 5 V.

Measurement Setup

The setup is based on a similar model presented by [10]. Two glass beakers made from DURAN (DURAN Group GmbH), with a transmission of 99 % in the interesting wavelength region, were arranged non-concentrically to cover the whole biological variability of the combined scalp & skull thickness (Image 2, (a)). The inner beaker is placed on a special shaped frame (Image 2, (b)) to ensure that the liquid phantom in the gap between the beakers can be stirred constantly during the experiments. This setup was simulated for different velocities of the stirrer with COMSOL (Comsol Inc.) to verify the fluid movement of the phantom.

Image 3 Measurement setup

2.5

Measurements

The measurements were divided into two pairs of source detector separation (Image 4). A thickness of 13 mm for the gap between the beakers was chosen (Layer 1). During the experiments the designed liquid scalp & skull phantom was filled in this gap. The second layer (Layer 2) in the schematic represents the measurement cell and the third layer (inner beaker) was filled with distilled water (Layer 3). Image 2 Arrangement of the glass beaker model A measurement cell (Layer 2) was attached to its inner side, simulating the target vessel (Image 2, (c)). The cell was connected to a peristaltic pump (MS-Reglo, Ismatec SA) with a flow rate of 43 ml/min to change the liquid inside during the experiments. The whole setup (Image 3) was covered with a black box to avoid the interference of the ambient light.

Image 4 Sensor arrangement; layered structure of the target region



S834

The measurement cell was flushed with distilled water to get a baseline and the changes of the signal intensities at the PDs, caused by the in- and outflow of pure Intralipid 20 % (Baxter Deutschland GmbH) in the first case and Indocyanine green (ICG) (Pulsion Medical SE, max. absorption at 805 nm) in the second case, were detected with the attached sensor. Intralipid (IL) has a higher level reflectance und less absorption compared with distilled water. This was the reason for the concentration changes in the measurements M1-M3. For M4-M6 this effect was generated by the absorbance of ICG.

3.

Results

To identify the needed distances of the photodiodes and the light source to measure exactly in the target region (layer 2), the results for both PDs were analyzed separately. The measurement configurations of M1-M3 were not suitable to extract signals out of the target region because PD1 detected signal information from Layer 1 and 2 (Target for PD2) and PD2 measured Layer 3 (beneath target region). By adapting the PD distances the penetration depths could be shifted into the target region. The results from the measurements M4-M6 revealed that for a source detector separation of 5 mm (PD1) and 15 mm (PD2) the first PD only measured the phantom (Layer 1) and the second PD additionally measured the signal change inside the cell (Layer 2). Due to the optical properties of ICG the signal amplitude decreases (inflow) and increases (outflow) during the experiments (Images 5-7).

Image 7 Signal intensity of PD2 (ICG outflow) The results from the series of measurements are visualized in the table below. Table 2 Overview of the measurements, results and conclusions Measurement M1 (IL inflow) M2 (IL outflow) M3 (IL inflow) M4 (ICG inflow) M5 (ICG outflow) M6 (ICG inflow)

4.

Image 5 Signal intensity of PD1 (ICG inflow)

Image 6 Signal intensity of PD2 (ICG inflow)

Result and conclusion Less absorption, more reflection PD1 detected signal changes in layer 2 Target region of PD2 The effect of M1 is reversible Signal of PD2 not influenced PD2 measured photons from the deeperlying layer 3 – out of the target region No influence on PD1 Only detection of layer 1 Hit target PD1 More absorption, less reflection PD2 measured signal changes in Layer 2 Hit target PD2 The effect of M5 is reversible

Discussion

The results of the experiments proved the capability to measure signal changes in the anatomical target region (Layer 2). For a source detector separation of 5 mm (PD1) and 15 mm (PD2) (M4 – M6) the first PD only measured the liquid scalp & skull phantom (Layer 1) and the second PD additionally measured the signal change inside the cell (Layer 2). In the clinical application the weighted difference of these signals and of at least two wavelengths should provide the information from the anatomical target region. The previous experiments revealed that the optimal distances of the PDs und suitable wavelengths have to be definied. Concerning these requirements and the geometry of the phantom model, a sensor was specified. This design should feature the following characteristics: Measurements in two dimensions coeval – in this case horizontal and vertical. Suitable for up to 6 PDs in each dimension with variable distances in-between. Suitable for 2 wavelengths per dimension.



6.

Image 8 Sensor schematics Image 9 visualizes the developed sensor concerning the assumptions mentioned above. The spherical shape corresponds with the radius of the outer glass beaker to ensure an exact fitting. The placement area for the horizontal PDs is curve-shaped as well to provide a constant spacing between to glass beaker and the sensor.

Image 9 Sensor design model The design specifications require an adaption of the electronics in order to control the new setup and the software has to be modified to record the data from the PDs. To avoid any reflection from the material of the sensor, it will be covered with black neoprene. This also provides a smooth attachement to the glass beaker.

5.

Conclusions

In this first test of the setup the optical properties should be clearly defined to verify the sensor configurations. The design of a bloodlike phantom to simulate the optical properties of the target region sufficiently is in progress. Furthermore a liquid phantom representing the brain tissue is part of the next model to realize a realistic setup. To manufacture the design concept of the sensor a 3D printer is utilized. During the production process the electronics are adjusted.

S835

Acknowledgements

The authors thank Ankit Malhotra for bringing in his electrical engineering skills and the ARSMD group (http://www.arsmd.uni-luebeck.de/) for technical assistance. This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B). LUMEN is a joint research project of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck.

7.

References

[1] S. A. McLellan and T. S. Walsh, “Oxygen delivery and haemoglobin”, Continuing Education in Anaesthesia 4 (4), 2004 [2] http://what-when-how.com/neuroscience/bloodsupply-of-the-central-nervous-system-grossanatomy-of-the-brain-part-2/, (27.01.2014) [3] Moreira-Gonzalez, F. E. Papay and J. E. Zins, “Cavarial Thickness and Its Relation to Cranial Bone Harvest”, American Scoiety of Plastic Surgeons, Vol 117 (6), 2005 [4] F. G. Persons, “The thickness of the living scalp”, University of London, Anatomy LXIII, pp. 427-429, 1929 [5] E. Okada and D. T. Delpy, “Near-infrared light propagation in an adult head model. II. Effect of superficial tissue thickness on the sensitivity of the nearinfrared spectroscopy signal”, Applied Optics, Vol. 42, No. 16, 2003 [6] W. Cormick, M. Stewart, M. G. Goetting, M. Dujovny, G. Lewis and J. I. Ausman, “Regional cerebrovascular oxygen saturation measured by optical spectroscopy in humans”, Stroke, pp. 596-602, 1991 [7] A. Barnett, J. Culver, A. Sorensen, A. Dale, and D. Boas, “Robust inference of baseline optical properties of the human head with 3D segmentation from magnetic resonance imaging”, Applied Optics 42 (16), 2003 [8] S. Prahl, “Optical Property Measurements using the Inverse Adding-Doubling Program”, 1999. [9] S. K. Loyalka and C. A. Riggs, “Inverse problem in diffuse reflectance spectroscopy: Accuracy of the Kubelka-Munk equations”, applied spectroscopy, Vol. 49, Issue 8, pp. 1107 – 1110, 1995 [10] R. K. Almajidy and U. G. Hofmann, “On the design of a multi-channel NIRS system to monitor functional brain activity”, 16th International Conference on Near Infrared Spectroscopy, La Grande-Motte, 2013 [11] A. Opp, “A Noninvasive Method to Determine Hemoglobin Concentration in Tubes”, Dissertation, Universität zu Lübeck, 2012



S836

Experimental Evaluation & Optimization of a UWB Localization System for Medical Applications C. Bollmeyer1, H. Hellbrück1, H. Gehring2 1 Center of Excellence CoSA, Lübeck Univsersity of Applied Sciences, Germany, {christian.bollmeyer,hellbrueck}@fh-luebeck.de 2 Klinik für Anästhesiologie und Intensivmedizin, Universitätsklinikum Schleswig-Holstein, Lübeck, Germany,

Abstract In the field of logistics, industrial applications and especially medical systems indoor localization plays an important role. In this paper we investigate the usage of a general purpose ultra wide band (UWB) localization system for an extended monitoring of invasive blood pressure (IBP) measurement of a patient. For precise IBP measurements the height of the transducer has to be on the same level as the reference point (e.g. the right atrium) of the patient. Changes in body posture of a patient during surgery introduce an offset error to IBP measurement. Manual adjustment of transducer to the altitude of the reference point is difficult and error-prone in daily routine. Consequently, the important localization parameter of the UWB system is the altitude of this reference point on a lying patient. So far no evaluation of UWB for this medical application or detailed measurements in multi-path environments is available. The accuracy in this medical environment is unknown. In this paper we perform an experimental evaluation of the accuracy of the UWB system. For a further improvement of the performance we design and implement fingerprinting and Kalman filtering as an extension of the original system especially targeted for this medical application. Finally, we evaluate these extensions via measurements in comparison with the original system.

1

Introduction

In the last years advances of wireless systems and embedded devices enabled new technologies in satellite based localization and indoor localization. Satellite based localization systems, like Global Positioning System (GPS), do not work properly in buildings due to signal attenuation and multipath propagation of satellite signals [1]. In intensive care vital data of a patient is essential. Besides oxygen saturation of blood and actual heart rate, blood pressure is an important parameter. Invasive blood pressure (IBP) measurement allows continuous and accurate blood pressure monitoring in intensive care [2]. IBP measurements strongly depend on body posture of the patient and are usually referenced to the right atrium of the heart [3] estimated by the midaxillary line. Figure 1 shows the setup of an invasive blood pressure measurement. Main components of IBP monitoring are a measuring apparatus (cannula and tubing), a transducer and a monitor. To avoid errors in IBP measurement, the transducer has to be on the same level as the reference point (e.g. the right atrium). A misalignment in the altitude of 1 cm results in an offset error in the IBP measurement of ~0.74 mmHg [10]. Therefore, releveling of the transducer after changes in the altitude of the reference point is necessary. In our medical application, we aim for an altitude measurement of the reference point with an accuracy of 1cm. Indoor localization systems locate so called tags. A tag fixed at the thorax of the patient can be used to estimate the altitude of the reference point. In the following we define terms for the rest of this paper. Accuracy is the mean absolute error between real and measured position [6]. Precision describes the variation of measurements – estimated altitude – caused by a random error [6]. In this

z

y x

Figure 1 Setup of an invasive blood pressure measurement paper we perform a Shapiro-Wilk and assume the random error to be Gaussian distributed. Therefore, we define precision as standard deviation of the measurements. Location lag is the time a localization system needs to update the measured position when a tag was moved to a new position. Ultra Wide Band (UWB) indoor localization systems use radio signals with a bandwidth over 500 MHz to estimate the position (also called: location) of a tag. An advantage of UWB systems is that they can provide accurate localization in multipath environments [4]. Chosen system for our work is the Ubisense IP 7000 UWB indoor localization system. In this paper we evaluate accuracy and precision of this system in a multipath environment. Later we present improvements of the accuracy of the measured altitude (z-coordinate). The contributions of our work are as follows:



We are the first to suggest an UWB localization system to measure the position of a reference point for IBP measurements We present detailed evaluation of the accuracy in a multipath environment We propose a fingerprinting algorithm combined with a Kalman filter to enhance the accuracy of the measured altitude The rest of the paper is structured as follows: Section 2 provides related work about the accuracy of wireless indoor localization systems. Section 3 describes the test setup and shows measurement results of the current state UWB indoor localization system. Section 4 proposes an fingerprinting algorithm to enhance accuracy of the measured altitude. Section 5 evaluates the optimized system according to accuracy and precision. Section 6 concludes our work and gives an overview about our future work.

2

Related Work

Farid et. al. provide an overview about recent advances of wireless indoor localization techniques in [5]. Lateration and angulation are commonly used by radio frequency based indoor localization systems to locate a tag by multiple base stations with known positions, called anchors. Angle of Arrival (AoA) measurements estimate the angle of the radiated signal by a mobile tag e.g. by directional antennas. Time of Arrival (ToA) and Time Difference of Arrival (TDoA) based methods estimate the tags position by measuring the propagation delay and thereby the distance from a tag and each of the three necessary anchors. Some indoor localization systems measure the Received Signal Strength (RSS) of a radio signal to estimate the distance and calculate the position of a tag by trilateration. RSS based indoor localization systems use theoretical and empirical models to translate the measured RSS difference in an range estimation [6]. These models are prone to radio effects like multipath propagation and provide accuracies of up to approx. 2m [6]. Therefore, RSS based solutions are not suitable for the described medical application. Ultrasound based systems can measure distances with an accuracy of centimeters. However, they are sensitive to interference caused by metal objects [5] and therefore inappropriate for the intended medical application. Dead reckoning systems are based on inertial sensors like accelerometers and prone to cumulative errors [5]. Such indoor localization systems use the knowledge of past position and additional sensor data to provide the current position. Cumulative errors lead to degrading accuracy, which can be corrected by Wi-Fi signals. Wi-Fi supported dead reckoning systems have an accuracy of up to 1.53 m [7] which makes these systems inadequate for the outlined medical application. The available UWB indoor localization system measures both TDoA and AoA to estimate the position of a tag. Those systems are called hybrid localization systems. Gezici et. al. investigated the theoretical accuracy of UWB base localization systems in [8]. An accuracy of less

S837

than 3 cm using an Signal to Noise Ratio (SNR) of 0 dB and 1.5 GHz bandwidth can be achieved under ideal conditions. However, a practical evaluation is not provided. Knox et. al. carried out a practical evaluation [9] but focused on tracking movements in x- and y- direction without optimization. In this paper we focus on evaluation and optimization of the z-coordinate (altitude) that is essential for our medical application.

3

Evaluation of Current State

We chose a test setup in our laboratory for an initial performance evaluation of the original UWB indoor localization system. Four anchors were placed at similar height and at different x- and y-coordinates spanning a quadrangle, as it is shown in Figure 2. Each anchor was mounted at an altitude of z=2.60 m to decrease damping of the signal caused by objects between anchor and tag. The area depicted in gray is called location cell. A dedicated measurement area was chosen for evaluation. Objects in the location cell, like laboratory desks and electronic devices, ensure realistic conditions e.g. multipath propagation.

Figure 2 Anchor positions and definition of measurement area The propagation delays for the requested accuracy are in the range of ns. Hence, time synchronization of anchors plays an important role for TDoA measurements. As a consequence our UWB indoor localization system is calibrated at P(320 cm,160 cm) to avoid timing offsets caused by timing cable connections. Calibration reference points are indicated by a cross in Figure 3 and Figure 4. For the initial evaluation of accuracy and precision, a tag is placed at various positions P(x,y) within a grid of ∆x=10 cm and ∆y=10 cm inside the measurement area. Our UWB indoor localization system operates at a localization rate of fr≈ 10 Hz. For statistical evaluation N=500 measurements are recorded for each potion. To calculate accuracy of the measured altitude at each position of the tag, following equation is applied:

Ez

1 N

N

( zi

z)

(1)

i 1

Ez stands for the mean position error – accuracy – of N=500 measurements. zi represents a single measurement of the UWB indoor localization system. z=75±0.2 cm is the true altitude of the tag, measured by an optical Bosch PLR 25 rangefinder. By applying (1) to each position, a 2-



D accuracy map is generated. Figure 3 shows the accuracy of the UWB indoor localization system, placing a tag at a reference level of z=75 cm within the defined measurement area.

Figure 3 Accuracy [cm] of UWB localization of a tag at z=75 cm within the measurement area Accuracy of the measured altitude at P(260 cm,180 cm) is 20 cm. Table 1 shows the mean values for the defined measurement area. We proved the hypothesis of normal distributed measurements by a Shapiro-Wilk with a significance level of α=0.05. Therefore, the standard deviation is used as performance metric of the precision and is listed in Table 1. Coordinate x-coordinate y-coordinate z-coordinate Accuracy 3.55 cm 3.93 cm 13.07 cm Precision 2.79 cm 4.32 cm 2.28 cm Table 1 Mean Accuracy and precision within dedicated measurement area Accuracy of the measured altitude is 13.07 cm. The xcoordinate has an accuracy of 3.55 cm. Our measurement results show, that the absolute accuracy of the z-coordinate is lower compared to the x- and ycoordinate. Furthermore, the systematic error of the measured altitude depends on the position P(x,y) of the tag as shown in Figure 3. These measurement results are reproducible. The initial evaluation shows that our intended accuracy of 1 cm is not achieved by the original UWB indoor localization system. Precision of the z-coordinate is 2.28 cm according to Table 1. In our medical application we aim for a precision below 1.35 cm to ensure that the majority of measurements are below 1 mmHg.

4

Optimization

The previous section showed, that the location error of the measured altitude follows a normal distribution and is systematic. The systematic error depends on the position of the tag P(x,y). To optimize the UWB indoor localization system regarding accuracy and precision of the measured altitude, we propose a fingerprinting algorithm and Kalman filter in this section. Fingerprinting algorithms consist of two phases [5]. The first phase is called training phase. The second phase is called localization phase. A characteristic feature that depends on the location of the tag is called fingerprint. During the training phase fingerprints are measured, e.g. RSS indicators from multiple access points, at each position of the tag. Fingerprints and their location are stored in a file or in a database to be used in the localization phase. Dur-

S838

ing the localization phase, the position of a tag is unknown. For localization, fingerprints are measured and compared with the data base or file to estimate the position of the tag. Our work uses a UWB indoor localization system which already provides the position of a tag. However, fingerprinting can be applied to reduce the systematic error of the measured altitude. In the training phase we measure the systematic altitude error, together with the current location P(x,y) of the tag. During the training phase, we chose a spacing of ∆x=20 cm and ∆y=20 cm between different positions of the tag. Figure 4 shows an example of a tag at an unknown position. The tag is surrounded by four points where fingerprints were collected during the training phase. In the localization phase our algorithm browses a file, created in the training phase, for the four surrounding fingerprints. First, the altitude error at positions I1 and I2 is approximated by bilinear interpolation. Finally a bilinear interpolation of the altitude error at the tag position x- and y-coordinate of the tag is calculated. This calculated offset value is subtracted from the measured altitude of the UWB system to reduce the systematic error. y P3(x2;y2) P4(x1;y2) I1

*

*

*

Δy

Tag(x;y) P1(x* 1;y1)

*I2

P2(x*2;y1) Δx

x

Figure 4 Interpolation between fingerprints to reduce altitude error of a tag In the next step we implemented a Kalman Filter [12] to increase precision of the UWB indoor localization system. To implement a Kalman filter, a mathematical system description is necessary [11]. Body posture of the patient is not continuously changed and static most of the time. Therefore, we neglect movements in the system description within the Kalman Filter which leads to simplification of filter equations. Based on the initial evaluation we estimate σ²=4 cm² as variance of the measurement error R. We heuristically estimated a value for the process noise variance of Q=0.4 cm². In our application we apply a Kalman filter to the position P(x,y,z) of a tag.

5 Evaluation of the Optimized System In this section, accuracy and precision of the optimized system are compared with the initial evaluation to demonstrate and quantify the improvement. Finally, the location lag of the optimized UWB indoor localization is compared with the initial state. Measurements of the optimized UWB indoor localization system were conducted in the same test setup as described in Section 3. For this experiment, fingerprinting and Kal-



man filter are applied to increase accuracy and precision of the measured altitude of the UWB indoor localization system. 500 measurements are recorded for each position of the tag.

Figure 5 Accuracy [cm] of optimized UWB localization of a tag at z=75 cm within the measurement area To test interpolation between measured fingerprints the spacing between different positions of the tag is ∆x=10 cm and ∆y=10 cm. Figure 5 shows a 2-D contour map of accuracy of a tag at an altitude of z=75 cm. Compared with the previous results shown in Figure 3 the accuracy is increased. The mean error is reduced to 4.14 cm. Precision of the measured altitude increased from 2.28 cm to 0.63 cm. Table 2 shows accuracy and precision of the optimized system. Fingerprinting and Kalman filtering improve the location system substantially. x-coordinate y-coordinate z-coordinate Accuracy 3.56 cm 3.91 cm 4.14 cm Precision 0.60 cm 0.92 cm 0.63 cm Table 2 Accuracy and precision of optimized UWB indoor localization system within dedicated measurement A small location lag of the UWB indoor localization system is needed for the medical application. The system needs to react within 5.0 s to a change in the position. To measure the location lag a tag was moved fast from a reference altitude to a higher altitude at a time t. At t=0 s a tag was moved from a reference altitude of 0.94 m to an altitude of 1.15 m. The system without filtering reacts within ms to the movement of the tag. The optimized system with Kalman filter needs ~2 s to react to the updated altitude. Therefore, the Kalman filter results to a slightly higher location lag which is below the required 5 s.

6

Conclusion and Future Work

In this paper, we present an initial evaluation of an UWB indoor localization system for medical applications. We suggest a tag placed at the reference point of the patient to correct IBP measurements. Our initial evaluation shows stable and accurate measurement results compared with other radio based localization techniques. UWB localization for usage in IBP measurement requires an improved accuracy better than 1cm. This paper shows, that fingerprinting together with Kalman filtering substantially improves accuracy and precision of a measured altitude. In the future, we will investigate the systematic errors in more detail. Additionally, we plan to add more sensors to the tag to increase the accuracy of measured altitude.

7

Acknowledgements

8

References

S839

This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B). LUMEN is a joint research project of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck. [1] R. Mautz: Overview of Current Indoor Positioning Systems. Geodesy and Cartography, vol. 35, no.1, pp. 18-22, 2009 [2] B. Gupta and Sir C. Gairdner: Invasive Blood Pressure Monitoring. Update in Anaesthesia, vol. 28, pp. 37-42, 2012 [3] R. Kramme: Medizintechnik: Verfahren-SystemeInformationsverarbeitung. Springer, 2011 [4] S. Gezici and H. V. Poor: Position Estimation via Ultra-Wide-Band Signals. Proceedings of the IEEE, vol. 97, no.2, pp.386-403, 2009 [5] Z. Farid, R. Nordin, and M. Ismail: Recent Advances in Wireless Indoor Localization Techniques and System. Journal of Computer Networks and Communications, vol. 2013, doi:10.1155/2013/185138 [6] H. Liu, H. Darabi, P. Banerjee, and J. Liu: Survey of wireless indoor positioning techniques and systems. Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions, vol. 37, no. 6 ,pp. 1067-1080, 2007 [7] F. Evennou and F. Marx: Advanced integration of wifi and inertial navigation systems for indoor mobile positioning. EURASIP Journal of Applied Signal Processing, vol. 2006, pp. 164-164, 2006. [8] S. Gezici, Z. Tian, G. B. Giannakis, H. Kobayashi, A. F. Molisch, H. V. Poor and Z. Sahinoglu: Localization via ultra-wideband radios: a look at positioning aspects for future sensor networks. IEEE Signal Processing Magazine, vol. 22, no. 4, pp. 70-84, 2005. [9] J. Knox, J. Condell and K. Curran: Investigating ultrawideband location positioning as a guidance system for mobile robotics. Advanced Engineering in Informatics Journal, vol. 36, no. 1, pp. 3504-3510, 2011. [10] K. K. Figg and E. C. Nemergut: Error in Central Venous Pressure Measurement. Anesthesia & Analgesia, vol. 108, no.4, pp.1209-1211, 2009. [11] Ra. Faragher: Understanding the Basis of the Kalman Filter Via a Simple and Intuitive Derivation [Lecture Notes]. IEEE Signal Processing Magazine, vol. 29, no.5, pp.128-132, 2012. [12] R. E. Kalman et al: A new approach to linear filtering and prediction problems. Journal of basic Engineering, vol. 82, no.1, pp.35-45, 1960.



S840



S841



S842



S843



S844

Speckle-based holographic detection for non-contact Photoacoustic Tomography C. Buj1,3, J. Horstmann2, M. Münter1 and R. Brinkmann1,2 1 Institute of Biomedical Optics, Universität zu Lübeck, Lübeck, Germany 2 Medical Laser Center Lübeck GmbH, Lübeck, Germany 3 Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Lübeck, Germany

Abstract The main problems of most state-of-the-art Photoacoustic Imaging approaches are long acquisition times and the requirement of acoustic contact. We introduce a very fast innovative holographic-optical full field non-contact detection method to overcome these problems. In order to increase the acquisition speed significantly, the surface displacements of the object, caused by the photoacoustic pressure waves, are interferometrically measured by a high speed camera in two dimensions. Phase alterations in the observed speckle field determined by Electronic Speckle Pattern Interferometry (ESPI) are used to identify changes in the object’s topography. The total acquisition time can be reduced to 100 ms. This approach was validated with a silicone phantom with an embedded spherical absorber. After recording the topography of the object over time, a tomographic reconstruction leads to the three dimensional location of the different absorbers. A reliable reconstruction proves the ability of the method.

1

Introduction

To overcome the depth resolution limitations of many optical imaging techniques like Optical Coherence Tomography OCT, Photoacoustic Imaging has become very popular in recent years. Photoacoustic Imaging is based on the emission of thermoelastic pressure waves, generated by absorbers subsequent to short-pulsed optical excitation (Fig. 1). The high optical contrast is caused by the chromophore-dependent absorption coefficient of the tissue. Thus, the choice of the correct wavelength, where the target structure has a high absorption coefficient and the surrounding tissue a lower, is crucial. After the pulsed light has reached the tissue surface, it is distributed depending on scattering and absorption. Due to absorption by desired absorbing structure, its temperature 1. Pulsed irradiation 2. Light distribution inside the tissue 3. Absorption 4. Thermoelastic expansion 5. Propagation of Pressure transients 6. Signal detection Fig. 1 Abstract schematic illustration of the photoacoustic effect, which is been used in different imaging techniques.

increases followed by a rapidly rising pressure. Thus, the absorber expands thermoelastically and pressure transients are emitted. Because of their low refraction and scattering, pressure waves can propagate large distances in tissue. In comparison to optical imaging techniques, this is a major advantage. There are various methods to detect these pressure transients at the tissue surface. In many experimental setups, single element sensors are used. However, their use requires acoustic contact, and due to the need for mechanical scanning over the object a long acquisition time up to several minutes [1]. In particular for in vivo measurement, this is not acceptable because e.g. motion artifacts cannot be compensated. For this reason, a novel holographic detection method that reduces the data acquisition time to 100 ms was developed. The main difference compared to classical piezoelectric detection is that surface displacements are measured in order to reconstruct the absorbing structures. For tomographic imaging, the position of the absorbers can be reconstructed in three dimensions by post-processing the detected signals, using e.g. time reversal approaches like kwave [2] or triangulation based algorithms like delay and sum [3].

2

Material and Methods

Fig.2 illustrates the principle of a novel holographic detection method. The detection laser light is scattered mainly in the superficial tissue. A Mach-Zehnder interferometer is the main component to measure the displacement of the tissue surface. The light of a reference wave interferes with the multiple backscattered object light which leads to a holographic interferogram that is detected by the high-speed camera. The individual components of the setup are controlled in a triple-pulse mode (Fig. 3).



S845

Fig. 2 Principle of the holographic-optical full-field measurement technique to determine the surface displacement. Main components of the system are a short pulsed excitation laser and a Mach-Zehnder interferometer that is composed of a short-pulse detection laser, a high-speed camera and optical components. After the excitation by the excitation laser, pressure waves are emitted by absorbers. With a defined delay thereon follows the illumination of the surface by the detection laser. The light of a reference wave interferes with the multiple backscattered object light which leads to a holographic interferogram that is detected by the high-speed camera.

In the first step, the phase distribution is measured. This measurement is used as a reference and makes the method insensitive to thermal expansions of the entire tissue due to the previous pulses, artefacts by viscoelastic attenuation of the free boundary or other motion artefacts. In the second step, the tissue is laminary irradiated with the excitation laser, resulting in generation of thermoelastic pressure waves by the absorber. In the third step, the altered phase distribution is determined after a time delay ∆t relative to the excitation pulse. By subtracting the two phase distributions, the geometrical object surface displacement can be evaluated. In a repetitive measurement, the time delay is increased continuously by choosing appropriate trigger frequencies for the excitation and the detection laser. The computed phase difference images contain the actual topographical displacements of the considered period, which are used for the reconstruction. Based on the acquired surface displacements over time, the positions of the absorbers within the tissue can be tomographic reconstructed.

2.1

Proof of concept

To demonstrate the applicability of the detection approach, an experimental setup was created (Fig. 4). In this setup, a transparent silicone cube (Wacker Silicones RT604 A/B) of 10 mm edge length serves as tissue model. A highly absorbing black silicone spherical absorber of about 2 mm diameter is located inside the silicone cube. One side of the cube is irradiated by a Nd:YAG laser (Quantel YG571C, wavelength 1064 nm, pulse duration 6 ns, pulse energy 25 mJ, spot size 5 mm, repetition rate 10 Hz) . For the used excitation wavelength, the absorption coefficient of the black silicone was measured to be 29.6 cm-1. The acoustic properties of the phantom material were measured to be 0.93 mm/µs for the speed of sound and 0.62 kN/mm² for the elasticity modulus. The density is 0.79 g/cm³. The thermal expansion coefficient is 2 x10-4 K-1. For ease of operation, the detection was performed on an orthogonal side of the phantom. To increase backscattering

Fig. 3 Principle of the repetitive triple pulse detection mode.



S846

wave with the free boundary is visible. Assuming a sound velocity of 0.93 mm/µs, we expect a depth position of the surface of the absorber of 4.46 mm. In the further course, the circular displacement spreads omnidirectional and is reflected at the boundary surfaces of the phantom. Due to existing hardware limitations the measurement is not equidistant in time.

3.2

of the detection laser light (CryLas FTSS 355-50, wavelength 532 nm, pulse duration 1 ns), the illuminated surface was coated with a 150 µm thick layer of white silicone. A CCD camera (Basler Pilot piA1600-35g, resolution 1600x1200 pixels, pixel size 7.4 µm) with a frame rate of 20 Hz was used to record the images. The temporal resolution of the measurement technique is about 100 ns. The phase variations can be evaluated by Electronic Speckle Pattern Interferometry (ESPI) [4]. ESPI in Spatial Phase Shifting (SPS) Mode [5] is implemented, in order to get the phase information in a single image acquisition. In this approach, the lateral resolution of the phase measurement is influenced by the mean speckle size (4 pixels), the SPS phase period on the camera chip (4 pixels), the pixel pitch (7.4 µm) and the imaging scale (1). In order to reduce speckle artefacts, a Gaussian blur filter with 12 pixels in size was applied to the complex phase maps. Validation measurements have shown that these parameters lead to a lateral resolution of 100 µm. The interferometric stability is restricting the axial resolution. For the presented setup it is about 4 nm. The first tomographic reconstructions were performed in two dimensional space and have been carried out using the delay and sum algorithm as introduced by Carp and Venugopalan [6], which is similar to the Synthetic Aperture Focusing Technique (SAFT) [7]. As a reference measurement in order to determine the depth of the spherical absorber, the phantom was imaged with an SLR (Canon EOS 500D - 15.1 Megapixel CMOSSensor, Canon EF-S Objective 18-55mm). In the background of the phantom a graph paper was placed as scale.

3

Results and Discussion

3.1

Holographic measured surface displacements over time

In the presented measurement, 228 surface displacements were recorded, which lead after pre-processing to 114 temporally phase difference images. A small representative selection of these can be seen in Fig. 6. After 4.8 µs, a spherical displacement as an interaction of the spherical pressure

Fig. 5 shows the result of the photographic determination of the distance between the phantom surface and the absorber. With the knowledge of the pixel pitch, the distance between the phantom surface and the surface of the absorber can be determined to be 3.95 mm. Because the absorber has a measured diameter of 1.95 mm, its center is located at a depth of approximately 4.9 mm. Due to many possible aberrations of this measurement method, these results serve only as approximate reference values. Nevertheless, this depth value differs only slightly from the depth value, which has been estimated based on the displacement measurements. In theory, a surface displacement should become visible after 4.2µs. Because of the current measurement sensitivity, this small displacements are not detectable. In prospective phantom measurements, the depth of the absorber should be determined by OCT, which will increase the accuracy of the reference method.

Depth

Fig. 4 Schematic illustration of the measurement of a spherical point absorber which has been positioned in a transparent silicone volume of 10 x 10 x 10 mm. The highly absorbing black silicone absorber has a diameter of 2 mm and was excited by a short pulsed laser.

Photographic specified position of the spherical absorber

Fig. 5 Results of the photographic localization: Distance between the spherical absorber and the surface of the phantom is 3.95 mm. The center of the absorber is located at a depth of 4.9 mm.

3.3

Reconstructed position of the spherical absorber

For the two-dimensional reconstruction, a line was extracted from the surface displacement data set, which crosses the epicenter of deformation. The result after performing a threshold filter is shown in Fig. 7. Accordingly, the center of the absorber is at a depth of 4.9 mm. Crucial for the validity of the edge structures is the choice of the threshold value, which was executed for artefact and noise suppression. This has the disadvantage that the information of the contours of the absorber are also extinguished. In further studies, the reconstruction algorithm should be validated in terms of its accuracy with respect to the absorber geometry. Moreover, the number of the absorbers in the phantom, and a varying configuration should be analyzed.



S847

Fig. 6 Surfaces displacements at different times. a) 1.0µs: Flat surface - pressure waves are still on their way b) 4.8µs: A centered circular displacement becomes visible. c) 5.8µs: The circular displacement increases omnidirectional. d) 6.9µs: An annular wave was formed. e) 7.8µs: Annular wave has spread. f) 8.4µs: Annular wave has spread and was reflected at the boundary surfaces of the phantom.

In addition, the algorithm will be expanded to reconstruct the contours.

of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck.

6

References

[1] Xu M., L. Wang V.:Photoacoustic imaging in biomedicine. Review of Scientific Instruments, vol. 77, no. 4, (2006). [2] Treeby, B., Cox, B.: k-Wave: MATLAB toolbox for the simulation and reconstruction of photoacoustic wave fields. Journal of Biomedical Optics, 15(2), 021314 (2010). Fig. 7 Results of the reconstruction algorithm after performing a threshold filter. Depth of the center: 4.9 mm

[3] Hoelen, C., de Mul, F.: Image reconstruction for photoacoustic scanning of tissue structures. Applied Optics, 39, 5872-5883 (2000).

4

[4] Pedrini, G., Pfister, B., Tiziani, H.: Double pulse-electronic speckle pattern Interferometry. Journal of Modern Optics, 40 (1), 89-96 (1993).

Conclusions and outlook

With the results presented in this paper it could be shown that non-contact displacement measurement is a suitable approach for photoacoustic detection. In addition, the reconstructed positions agree well with the values determined by a reference measurement method. In the following steps, the sensitivity will be increased. For comparison with other detection methods the noise equivalent pressure (NEP) value should be determined.

5

Acknowledgements

This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B). LUMEN is a joint research project

[5] Helmers H., Burke J.: Performance of spatial vs. temporal phase shifting in ESPI. Proc. SPIE 3744, 188-199 (1999). [6] Carp S., Venugopalan V.: Optoacoustic imaging based on the interferometric measurement of surface displacement. Journal of Biomedical Optics, 12(6), 064001 (2007). [7] Feng, D., Xu, Y., Ku, G., Wang, L.V.: Microwave-induced thermoacoustic tomography: Reconstruction by synthetic aperture. Medical physics 28, 2427–2431, (2001). [8] Horstmann, J., Brinkmann, R.: Optical full-field holographic detection system for non-contact photoacoustic tomography. Proc. SPIE 8943-55 (2014).



S848

Insight in Scanner Construction for a Dynamical Field Free Line for Magnetic Particle Imaging M. Weber1,4 , K. Bente2,4 and T. M. Buzug3 , 1 Institute of Medical Engineering, Universität zu Lübeck, Lübeck, Germany, [email protected] 2 Institute of Medical Engineering, Universität zu Lübeck, Lübeck, Germany 3 Institute of Medical Engineering, Universität zu Lübeck, Lübeck, Germany 4 Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Lübeck, Germany

Abstract Magnetic Particle Imaging (MPI) is a new imaging modality capable of visualizing the distribution of super-paramagnetic nanoparticles while achieving high spatial and temporal resolution. It promises medical application for high-contrast angiography and cancer imaging. MPI was first published in 2005 and shows a fast development in hardware design, spatial encoding schemes and reconstruction techniques. Commonly, a magnetic field configuration with a field free point (FFP) is used to enable spatial encoding. At lower particle concentrations, the signal-to-noise ratio (SNR) decreases to a critical range. An enhanced spatial encoding scheme uses a field free line (FFL) and promises to improve the sensitivity by one order of magnitude. The presented work describes the manufacturing of an optimized scanner topology that realizes a FFL with excellent field quality. Based on curved rectangular coils the power loss is efficiently minimized and high field quality accomplished. Furthermore, an overview of the structure and setup of the signal chain is demonstrated, which is a critical element in MPI imaging. Both, field generating part and signal chain give important insights on future FFL imaging.

1

Introduction

The sensitivity of the new imaging modality MPI [1] can be optimized by using the spatial encoding scheme of an FFL [2]. Simulations on optimizing the FFL scanner design have shown an improved electrical power consumption to values comparable to those for FFP scanners of equal size and gradient performance [3]. A proof of concept on magnetic field generation for dynamic FFL imaging with a gradient strength of 0.25 T m 1 µ0 1 was implemented in 2011 [4]. The presented work describes the assembly of an enhanced FFL prototype that is optimized with respect to electrical

1

SF

O

O

SF

4

Outer selection field

power consumption and field quality [5]. Here, a novel coil design of curved rectangular coils serves as a template featuring excellent field quality. These very homogenous gradient fields have the characteristic that reconstruction processes can be performed with x-space theory [6]. Here, the particle signal can be efficiently reconstructed with Radon-based approaches which are familiar from computed tomography [7, 8]. This is a critical advantage concerning the established reconstruction with a system matrix since measurement and inversion are very time consuming and place high demands on computational resources. Next to the field generating part in MPI, the signal chain plays a significant role. Thus, components of the implemented setup are explained in detail which complete the hardware side of the scanner.

SFI 1

Outer drive field

2 2.1

⇠

SFI 2

⌘ x

DFO 1

DFO 2

y SFI 4

Methods

DFI 1

FFL Generation

The central goal behind the presented coil array is to find a compromise between optimal current distribution and practiInner drive field cal realization. The field generating parts can be separated into four different coil types. Two coil types belong to the y FFL generating and rotating array respectively two types z build the coil parts that shift the FFL. The FFL generation is x done with the inner and outer selection field coils as one can see in Fig. 1. According to the current direction it is possible Figure 1: The picture shows the FFL generator consisting to either generate a horizontal or vertical FFL with the inner of 24 coils. These coils build the inner (SFI, double bundle) selection field (SFI) coils. The outer selection field (SFO) and outer selection (SFO, triple bundle) field array and inner coils feature a rotation of 45 and thus, the corresponding (DFI) and outer (DFO) drive field array. FFL is also rotated by 45 . The combination of both arrays L

FF

O

SFI 3

3

DFI 2

O

SF

2

SF

Inner selection field



enables the generation of an FFL with arbitrary angle. Additionally, two NdFeB permanent magnets with opposing south poles undertake the static part of the magnetic field to minimize the currents in the coils. They are aligned on the bore axis and are shielded with copper. Litz wire is used for the winding process and epoxy resin guarantees strength of shape of the coils. The final shape is obtained by a pressing and heating process. The assembly consisting of 24 single parts is designed in such a way that appropriate air cooling through small slits (⇡ 1.5 mm) between the coils establishes good heat dissipation (Fig. 3). To evaluate the cooling strategy a temperature profile is measured. Furthermore, a scanner case is developed that embeds the coils in a stable environment. An important requirement is an air cooling device for appropriate heat dissipation. Here, the fine mechanical workshop (Universität zu Lübeck) and a 3D printing system (FORMIGA P 110, EOS, Krailling, Germany) is used.

S849

from the measured data to obtain the correct absolute values of the magnetic field.

2.2

Signal Chain

The signal chain in MPI is a very failure-prone factor in the imaging process. Thus, it is decided to fully copper shield cable routing and passive filters. The design is done with CAD software (Solidworks, Dassault Systèmes Solidworks Corp, MA, USA). Additionally, it could be felt back on a shielding room. An autarkic battery supply shall be realized. The filters have to be adapted for 25 kHz as excitation frequency, which is applied on the drive field coils. Here, a bandpass filter will be realized to ensure clean signals. Before the generated particle signal is digitized the coupled excitation signal has to be eliminated. Hence, a bandstop filter rejects 25 kHz. Furthermore, filters are implemented to reject unwanted disturbing signals before entering the shielding room concerning the selection field channels and also to protect DC sources from coupled AC signals.

Actual FOV FOV

3

5.0

7.5

3.1 2.08 mm y-shift

y [mm]

7.5

5.0

2.5 0.0

2.5

5.0

2.5 0.0

FFL Generation

Fig. 3 shows the constructed coil array. The visual impression in comparison to the design in Fig. 1 is promising. Geometric aspects as well as coil distances are kept and build a stable construction. Spacers between the coils support stiffness and optimize the air cooling that is lead through the slits.

2.08 mm x-shift

7.5

Results

2.5

5.0

7.5

x [mm]

Figure 2: Offset of the measured magnetic field components. The x- and y-sensors of the Hall probe are perpendicular aligned, but shifted about 2.08 mm compared to the geometric center of the probe. Thus, the actual magnetic field has to be interpolated and the actual FOV is smaller (highlighted area). The background shows the bore that limits the FOV. To proof assembly and functionality the magnetic field is measured. These measurements are realized with a threedimensional Hall probe (MMZ-2502-UH, Lake Shore Cryotronics, OH, USA). The field of view (FOV) is chosen to be 10 mm ⇥ 10 mm. The measurement setup is calibrated by simulating a three-dimensional volume of the magnetic fields, wherein the displacement and rotation of the actual two-dimensional measured magnetic field is determined. Hence, the position of the Hall probe can be adjusted with the new data. A drawback of the three-dimensional hall probe is a certain offset of the sensors. The scheme in Fig. 2 gives an impression. The x- and y-sensors are each shifted about 2.08 mm in their corresponding direction referred to the probe center. Thus, an algorithm is implemented to overcome this problem. The actual FOV is linearly interpolated

Figure 3: Assembled coil configuration. 24 curved coils form selection and drive field coil arrays. The bore has a diameter of 36 mm. Magnetic field measurements confirm high field quality. A gradient of 0.4 T m 1 µ0 1 was realized. The normalized root mean square deviation (NRMSD) amounts to 2.39 % compared to the simulated FFL. These minimal differences can be a result of inaccuracies of the hall probe, the adjustment and imprecise setting of the currents respectively. Nevertheless, these minimal errors are comparable to similar works [4, 9] and build the basis for future imaging processes



that resort to Radon-based imaging techniques which require gradient fields with highly homogeneous gradients. Analogous results are measured for an additional shift of the FFL. The average NRMSD amounts to 2.78 %. Furthermore, the measured electrical power consumption is determined which amounts to 37.8 W. It is slightly higher than the simulated power loss with 32.94 W. The higher power loss might be a result of broken litz wire as a consequence of the pressing process in the outer selection field coils. These values correspond to a gradient of 0.4 T m 1 µ0 1 . The customized scanner case is shown in Fig. 5. The construction is based on four plates (Polyoxymethylene) that are connected via threaded rods. The coil array is mounted on a centered tube and enclosed by the cooling system. The cooling system consists of six circularly aligned fans building the inlet on the left side. First tests show that a gradient of 1.08 T m 1 µ0 1 can be achieved without an overheating of the coils. Fig. 4 demonstrates this by showing a temperature profile over 300 s during the application of the maximum currents and the following cooling process with 300 s. The measurement is done for 300 s since later imaging is planned to be realized in less than 60 s. The peak temperature is about 68 C and hence below a critical limit of approximately 100 C at which the epoxy resin melts. Although the saturation point is not reached, this temperature profile is sufficient for later imaging in shorter time ranges. The temperature is measured with a thermal camera (Testo 875-1i, Testo AG, Lenzkirch, Germany).

S850

Figure 5: Embedded coil array in the scanner case. The circular aligned fans provide the necessary air cooling. On the right side one can see a copper shielded permanent magnet. is boost to the desired amplitude by two amplifiers (7796 POWER AMPLIFIER, AE Techron, Elkhart, IN, USA), i.e. one amplifier per channel. Afterwards, the signal line is fully copper shielded up to the drive field coils. Before the signal enters the shielding room a third order Butterworth filter rejects higher harmonics. Our filters feature a damping of 55.9 dB and 56.0 dB at 75 kHz. Finally, impedance matching minimizes reflections from the load.

70

T [ C]

60

50

40 0

100

200

300 t [s]

400

500

600

Figure 4: Temperature profile of the coil array at a gradient Figure 6: Assembled coil configuration. 24 curved coils of 1.08 T m 1 µ0 1 . After 300 s the currents are switched form selection and drive field coil arrays. The bore has a diameter of 36 mm. off. However, for higher gradients, stronger NdFeB permanent magnets would be necessary. Nevertheless, for future MPI FFL imaging and the proof of principle, this gradient is sufficient.

3.2

Signal Chain

An overview of the whole MPI FFL system in our shielding room is given in Fig. 6. The center of the signal chain is a standard PC with Intel Core i5-3330 with 4 cores (2.4 GHz), 8 GB RAM and 64-Bit Windows 7 Professional (SP1). Here, signal generation and signal acquisition is done with corresponding control cards (X3-A4D4 and X3-10M, Innovative Integration, CA, USA). The 25 kHz excitation signal

Furthermore, the PC controls the DC sources (SM 7.5-80 and SM 18-50, Delta Elektronika, Zierikzee, Netherlands) which set the appropriate currents for the FFL rotation. Behind the DC sources two filter systems are installed. The first one is a DC filter that rejects higher frequencies (Bajog electronic, Pilsting, Germany) before the signal line enters the shielding room. The second one protects the DC sources of coupled 25 kHz signals. They damp signals at 25 kHz by approximately 30 dB. The receive channel uses a customized receive coil array to acquire the particle signal (see Fig. 7). A 3D printed form serves as a template in which litz wire is wound. This method avoids breaking of the litz wire and establishes very precise realization. The array consists of two Helmholtz



S851

[2] J. Weizenecker, B. Gleich, and J. Borgert, “Magnetic particle imaging using a field free line,” Journal of Physics D: Applied Physics, vol. 41, p. 105009, May 2008. [3] T. Knopp, M. Erbe, S. Biederer, T. F. Sattel, and T. M. Buzug, “Efficient generation of a magnetic field-free line,” Med Phys, vol. 37, pp. 3538–40, Jul 2010.

Figure 7: Customized coil form for the receive coil array. It is 3D printed and consists of an inner coil that records in [4] M. Erbe, T. Knopp, T. F. Sattel, S. Biederer, and T. M. Buzug, “Experimental generation of an arbitrarily roy-direction and an outer coil that records in x-direction. tated field-free line for the use in magnetic particle imaging,” Med Phys, vol. 38, pp. 5200–7, Sep 2011. coil pairs that build one unit. The inner coil is designed to record signals in the y-direction and the outer coil in the x- [5] M. Erbe, T. F. Sattel, T. Knopp, and T. M. Buzug, “Enhancing the efficiency of a field free line scanning device direction. Respectively, the received signal is than processed for magnetic particle imaging,” in IEEE Medical Imagin a fourth order Butterworth filter to reject the coupled ing Conference, Anaheim, USA, 2012. 25 kHz excitation frequency. Subsequently, the signal is lownoise amplified and differentially led outside the shielding [6] P. Goodwill and S. Conolly, “Multidimensional x-space room where it is digitized with the control cards. magnetic particle imaging,” Medical Imaging, IEEE Transactions on, vol. 30, pp. 1581–1590, Sept 2011.

4

Conclusion

[7] T. Knopp, M. Erbe, T. F. Sattel, S. Biederer, and T. M. Buzug, “A fourier slice theorem for magnetic particle This paper demonstrates an FFL generating setup which imaging using a field-free line,” Inverse Probl, vol. 27, proofs that power loss can be minimized by optimal current no. 9, p. 095004, 2011. distributions around the bore without any drawback concerning FFL field quality. Therefore, the FFL was measured for [8] T. M. Buzug, Computed Tomography: From Photon Statistics to Modern Cone-Beam CT. Springer, Berlin, different angles and compared to the simulation. The average 1. auflage ed., 2008. NRMSD is less than 3 %, which promises great results in future imaging processes, since the basis for fast and effective Radon-based reconstruction algorithms is laid. Furthermore, [9] P. Goodwill, J. Konkle, B. Zheng, E. Saritas, and S. Conolly, “Projection x-space magnetic particle imagthe system is completed by designing and constructing the ing,” Medical Imaging, IEEE Transactions on, vol. 31, signal chain with the focus on shielded components to minpp. 1076–1085, May 2012. imize externally interfering signals. Future work has to assemble these single parts to a complete imaging system by adding appropriate signal processing and reconstruction and would complete the first dynamical FFL scanning device. Additionally, this offers great chances to investigate the sensitivity gain in FFL imaging compared to FFP imaging, since the scanner is capable of generating an FFP by just setting a different current profile.

Acknowledgment This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Bundesministerium für Bildung und Forschung (BMBF) (FKZ 13EZ1140A/B). LUMEN is a joint research project of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck.

5

References

[1] B. Gleich and J. Weizenecker, “Tomographic imaging using the nonlinear response of magnetic particles,” Nature, vol. 435, pp. 1214–7, Jun 2005.



S852

Investigation of particle dynamics near the endothelial glycocalyx by multi focus FCS L. Kreutzburg1,2, V. Dolezal1, C. G. Hübner1 1 Institute of Physics, Universität zu Lübeck, Lübeck, Germany, [email protected] 2 Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Lübeck, Germany

Abstract Much effort has been spent doing research of biophysical and rheological properties of the endothelial surface layer (ESL) in recent years. The ESL is put into context with different physiological processes like blood regulation or oxygen transport as well as pathological issues like diabetes and arteriosclerosis [1]-[3]. Various publications consider the structure and composition of the ESL, but very little is known about the dynamics of small particles near the ESL or the dynamic equilibrium between constituents of the ESL and the plasma. We propose a multi focus fluorescence correlation setup which enables for the measurement of diffusional and translational motion for the investigation of particle dynamics near the ESL. First dynamics results as well as our approach for modelling blood capillaries are presented.

1

Introduction

The endothelial glycocalyx is a layer of proteins, glycolipids, glycoproteins and proteoglycans that line the wall of blood capillaries. Together with proteins of the blood plasma and glycosaminoglycans, the glycocalyx constitutes the endothelial surface layer (ESL). The ESL features a dynamic equilibrium regarding its composition and thickness, as membrane-bound components may be loosened by enzymatic degradation or shear stress and an exchange exists between proteins of the ESL and the flowing blood [2], [3]. Although the endothelial glycocalyx was described more than 70 years ago, its relevance for the physiology of the capillary system became evident just in the last recent years [1]. It functions as an additional barrier to colloids of the blood plasma and thus has additionally been appended to the Starlin model that describes the hydrostatic and colloid osmotic pressures involved [4]. The ESL is further associated with influences on the flow resistance and regulation of blood flow, oxygen transport, coagulation, angiogenesis and shear stress sensing by mechanotransduction [3], [5]. Ischemia and reperfusion lead to a degeneration of the glycocalyx and increase the permeability of the capillary wall. Tissue edema formation with an subsequent organ dysfunction may be the consequence. Besides, the role of the ESL at diseases such as diabetes, sepsis, arteriosclerosis and tumor metastasis are suspected [1]. Despite many publications about the structure of the ESL, its influence on blood rheology and the permeability of the capillary wall, little is known about the dynamics of single particles at the glycocalyx. To the best of our knowledge, there is a single paper by Stevens, Hlady and Dull [6] which concerns the investigation of albumin dynamics near by and in the ESL by fluorescence correlation spectroscopy (FCS). They measured diffusion rates and concentration profiles at lung endothelial cells and found that the diffusion constant is reduced by 30% and the concentration being increased

fivefold by the ESL. By enzymatic digestion of the glycocalcyx with pronase, heparanase and hyaluronidase before the measurement, they determined hyaluronan as an ESL component that is mostly involved in albumin-ESL interacions. Certainly they did not perform these measurements under flow conditions, as conventional FCS is not capable of detecting translational motion. In this report we demonstrate the multi focus fluorescence correlation spectroscopy (mfFCS) as a FCS variant which enables for both, detection of undirected diffusion and translational motion. We present our approach to model the biological blood capillaries by a microfluidic chip in combination with a custom-built syringe pump for flow generation.

2

Methods

2.1

Theory

In standard FCS experiments, fluctuations of the fluorescence signal 𝐼(𝑡) from a single confocal volume are analyzed in order to obtain the diffusion coefficient and concentration of the fluorophores. This is done by the calculation of the autocorrelation (AC) of the signal, given by 𝐺(𝜏) =

〈𝐼(𝑡) ∙ 𝐼(𝑡 + 𝜏)〉 〈𝐼(𝑡)〉

(1)

In general, the correlation of the signal measures the probability of detecting a photon at time 𝑡 + 𝜏, if there was a photon detected at time 𝑡. This probability decreases as the dye leaves the focus, which leads to characteristic decay of correlation curve. As the confocal volume is rotation-symmetric, any translational motion of the dyes may be measured in terms of absolute values. The direction, however remains undetermined. To overcome this limitation, we propose a multi-focus setup with up to four spatial shifted foci. This is accomplished by



shifting the detection foci away from the central excitation focus as shown in Image 1.

These expressions are valid if the investigated system is ergodic and the flow is uniform. As both Eq. (3) and Eq. (4) share most parameters, global curve fitting can be performed on the measurement data. In this way, six cross-correlations and three autocorrelations can be used to fit the free parameters. Table I gives an overview of the common parameters for the auto- and cross-correlations. One exception is the number of molecules in the volume N, which appears in both correlations but differs because of the different volumes. Table I

Image 1 Configuration of excitation and detection foci. Left: The shifted detection foci are placed around the excitation focus. Right: The resulting volumina form a triangle. For the sake of clarity, the volumes are not depicted entirely, but only their position. The translational motion of the dyes is measured along the connecting lines of the volumes. The cross-correlation of the fluorescence signals 𝐼m and 𝐼n of two different volumes 〈𝐼m (𝑡) ∙ 𝐼n (𝑡 + 𝜏)〉 𝐺m,n (𝜏) = 〈𝐼m (𝑡)〉〈𝐼n (𝑡)〉

(2)

S853

parameter number of molecules in the volume specific diffusion time (ms) specific translation time (ms) structure factor structure factor orientation focus shift δ m (rad)

𝐺m,n N 𝜏d 𝜏f S R 𝛼

𝐺m,m N* 𝜏d 𝜏f S -

orientation focus shift δ n (rad)

𝛽 𝜎

-

flow direction (rad)

2.2

Experimental realization: The multi focus FCS setup

depends on the amount of translational motion of the dyes along the connecting line, i.e. the projection of the flow velocity vector on the connecting line can be measured. If the detection efficiency is approximated by a three dimensional Gaussian distribution with half axis 𝑊xy and 𝑊z , the following expression for the cross-correlation can be derived from Eq. (2): 𝜏R 𝜏 (1 − cos(𝛼 )) +1 2𝜏d 𝜏d 𝜏R 𝜏 exp − (cos(𝛽 − 𝜎) − cos(𝛼 − 𝜎)) +1 𝜏f 𝜏d 𝜏 𝜏 exp − +1 × 𝜏f 𝜏d 𝐺m,n (𝜏) = exp

N

d

+1

d

S +1

× × (3)

+ 1.

Where 𝜏d = 𝑊xy /4D is specific diffusion time, 𝜏f = 𝑊xy /𝑉 is the specific translation time and 𝑅 = 𝛿/𝑊xy and 𝑆 = 𝑊xy /𝑊z are structure factors. The shift of the detection foci is described by the path length 𝛿 and the angles 𝛼 or 𝛽, respectively. The variable 𝛼 denotes the difference 𝛼 − 𝛽. The autocorrelation simplifies to 𝐺m,m (𝜏) = exp − N

d

+1

d

𝜏 𝜏 +1 𝜏f 𝜏d

S +1

× (4) + 1.

Image 2 Experimental setup. 1: mono mode glass fiber, 2: galvano scanner, 3: telecentric lense system, 4: sample stage / objective, 5 & 6: beam splitter and mirrors, 7 pinhole The sample is excited with a 488 nm continuous wave (cw) laser. The beam is directed through a mono mode glass fiber (1 in Image 2) and subsequently passes the galvano scanner (2 in Image 2), that enables for shifting the beam perpendicular to the optical axis in the x-y plane. The scanner is followed by the telecentric lens system (3 in Image 2). A 1.2 NA water immersion objective (Nikon Plan Apo, Düsseldorf, Germany) generates the excitation focus in the sample and is fixed to a piezo stage with a driving range along the optical axis (z-direction, 4 Image 2) of 100 µm. The sample is localized on a cover glass or in a microfluidic chip, respectively. The fluorescent light emitted by the sample is collected by the objective, follows the optical path of the laser light back and passes the beam splitter (5 in Image 2). Beam splitter



and mirrors split the fluorescence in equal parts on the avalanche photo diodes (APD). In front of each APD there are sets of two lenses and one confocal pinhole (which implement a confocal focus placement). The APDs are mounted on stages, that can be positioned mechanically by high-precision threads. In combination with the telecentric lens system, these stages enable for a shift of the detection focus corresponding to the single APD. Flow measurements were performed with an ATTO 488 solution of 45 nmol\L at an adjusted flow rate of 1200 nl\ min.

2.3

Experimental realization: The measuring cell

3

Results

3.1

Syringe pump test

S854

The results are shown in Image 4. The syringe pump is driven in the time interval from 675 seconds to 1910 seconds. It takes some 425 seconds until the flow rate remains stable at a value slightly lower than 400 nL/min. To the end of the driving time interval, a peak appears, followed by a strong slope of the curve. After the syringe pump is switched off after 1910 seconds, the flow rate decreases as expected.

Image 4 Flow, Q generated by the custom made syringe pump. Image 3 The measuring cell. The syringe pump composed of a high-resolution actuator and a 50 μL syringe generates a flow that can be detected in the micro channel of the PMMA chip, placed on the sample stage. The amplification of the channel shows the planned blood capillary model, where endothelial cells with a glycocalyx mimic the capillary wall of biological vessels. The measuring cell consists of a custom-built syringe pump and a PMMA microfluidic chip (Klaus Ziemer, Langerwehe, Germany) with a channel of 100 x 200 µm crosssectional area. The pump is driven by a high-resolution DCMike Actuator M-227 (Physik Instrumente, Karlsruhe/Palmbach, Germany) with 10 mm travel range and 0.05 µm minimum incremental motion. The syringe (Hamilton, Reno, Nevada, USA) has a volume of 50 µL. The syringe is connected via Luer-Lock connector and a PFA tube with an inner diameter of 500 µm to the channel of the PMMA chip (see Image 3). In order to evaluate the performance of the syringe pump, the generated flow was measured. For this, the PMMA chip was replaced by a LG16-0025D flow Sensor (Sensirion, Staefa ZH, Switzerland). The flow rate 𝑄 was adjusted to 400 nL/min and the measurement was performed with pure water. The endothelial cells in the microchannel shown in Image 3 are planned for future experiments and are not used for the measurements presented here.

This flow measurement shows, that the syringe pump enables for the generation of a stable flow over a reasonable period of time, which is important as the FCS measurement requires a system that does not change significantly over time. The fact that the adjusted flow rate of 400 nL/min is not perfectly achieved may be related to the peak at the end of the measurement. This may be explained by an impurity in the microfluidic system causing a congestion that was finally cleared away. The time it takes to establish the desired flow rate takes more than one third of the available driving time, what remains unsatisfying.

3.2

Flow measurements

The results of the FCS measurements under flow are shown in Image 5. Since we have not established a sufficient characterization procedure for the MDF function so far, the discussion has to be limited to qualitative aspects. The autocorrelations AC1 and AC2 of the volumes one and two are almost identical (Image 5a). This indicates the similarity of the first and the second volume, resulting from a favorable adjustment of the optical elements involved in the corresponding detection pathway as all three volumes should be ideally identical. AC3 on the contrary has a much lower amplitude than AC1 and AC2 which indicates a respectively poor adjustment. Dittrich and Schwille showed that in the presence of a flow both, autocorrelation and cross-correlation exhibit a characteristic curve progression [7]. They implemented a specialized spatial two-photon fluorescence cross-correlation setup with two detection volumes



that, in contrast to our volume scheme, do not overlap. They found the autocorrelation of those volumes being shifted to shorter times as the flow was increased. The cross-correlation by contrast features a sharp peack at a characteristic 𝜏f , indicating the transition of a dye from one volume to the other. What can be seen in Image 5b – c is a mixture of both phenomena: as two detection volumes partially overlap, the correlation of the signal from the shared region appears like an autocorrelation, whereas the correlation of the signal from regions which do not overlap leads to a pure cross-correlation-like contribution.

(a)

(b)

as a proof of concept; further technical work steps regarding the data analysis and characterization of the confocal volumes are required in order to establish a suitable measurement procedure. In the future we will add a fourth focus to the three focus scheme in order to measure the translational motion in three spatial directions. The measuring cell is planned to be expanded by endothelial cells to imitate the conditions near the wall of biological blood vessels.

5

(d)

Image 5 Results from mfFCS measurements in the flow cell. (a) Autocorrelations AC1, AC2 and AC3 of the detection volumes. Compared to the correlation curves of FCS experiment without flow, these ACs are shifted to shorter , because the residence time of the dyes is reduced by the flow. (b) – (c) The cross-correlations of the fluorescence signals. The curves show differences in forth and back correlation which originate from the translational motion of the dyes. Differences in the back and forth cross-correlations indicate the direction of translational motion of the dyes. The largest difference can be found at the cross-correlation between volume one and three (Image 5d): the forward correlation has a higher amplitude than the backward correlation, which indicates a translational motion from volume one to three. The other correlation pairs have smaller differences, indicating smaller fractions of the flow from volume two to three and from one to two.

4

Conclusion

We presented a multi focus FCS scheme which enables for the measurement of both, diffusion and translational motion of dyes or labeled particles, respectively. First flow measurements show that an artificially generated, stable flow in the measuring cell can be detected. This results may be seen

Acknowledgement

We want to thank Jörg Schroeter from the Medical Sensors and Devices Laboratory of the Lübeck University of Applied Sciences for supporting us with the implementation of the microfluidic hardware components and for borrowing the flow sensor. This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B). LUMEN is a joint research project of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck.

4 (c)

S855

References

[1] D. Chappell, M. Jacob, B. F. Becker, K. HofmannKiefer, P. Conzen and M. Rehm: Expedition glycocalyx. A newly discovered “Great Barrier Reef”, Anaesthesist, vol. 57, no. 10, pp. 959-969, Oct. 2008 [2] S. Reitsma, D. W. Slaaf, H. Vink, Marc A. M. J. van Zandvoort and Mirjam G. A. oude Egbrink: The endothelial glycocalyx: composition, functions, and visualization, Pflugers Arch., vol. 454. no. 3, pp. 345-359, Jun. 2007 [3] A. R. Pries, T. W. Secomb and P. Gaehtgens: The endothelial surface layer, Pflugers Arch., vol. 440, no. 5, pp. 653-666, Sep. 2000 [4] M. Rehm, S. Zahler, M. Lotsch, U. Welsch, P. Conzen, M. Jacob and B. F. Becker: Endothelial glycocalyx as an additional barrier determining extravasation of 6% hydroxyethyl starch or 5% albumin solutions in the coronary vascular bed, Anaesthesist, vol. 100, no. 5, pp. 1211-1223, May 2004 [5] S. R. Collins, R. S. Blank, L. S. Deatherage and R. O. Dull: The endothelial glycocalyx: emerging concepts in pulmonary edema and acute lung injury, Anesth. Analg., vol. 117, no. 3, pp. 664-674, Sep. 2013 [6] A. P. Stevens, V. Hlady and R. O. Dull: Fluorescence correlation spectroscopy can probe albumin dynamics inside lung endothelial glycocalyx, Am. J. Physiol.Lung Cell. Mol. Physiol., vol. 293, no. 2, pp. L328L335, Aug. 2007. [7] P. S. Dittrich and P. Schwille: Spatial two-photon fluorescence crosscorrelation spectroscopy for controlling molecular transport in microfluidic structures, Anal. Chem., vol 74, no. 17, Sep. 2002



S856

Path to construct synthetic synapses for enhanced electrode-nerve interaction in neuroprostheses S. Ebrahimpoor1, C. Zeilinger2, T. Doll1,3, T. Lenarz1,3, P. Aliuos1,3 1. Clinic for Othorhinolaryngology, head and neck surgery, Hannover Medical School, Hannover Germany 2. Department of Biophysics, Leibniz University Hannover, Hannover, Germany 3. Cluster of Excellence Hearing for All (H4All), Hannover, Germany

Neuroprostheses delivering adhesion sites for direct connections with neural tissue (i.e. neurons and their extensions) are desired in biomaterial research for enhancing the function of the implants. Lower impedances, higher number of electrode contacts as well as decreased neural activation thresholds all leading to higher selectivity of electrical stimulation can only be realized in the presence of direct neuron-implant contacts. Neural cell adhesion molecules of immunglobuline super family (Ig-NCAMs) play crucial role in dendritic adhesion, axon guidance and synaptogenesis. The long term goal of the present study is to investigate the possibility of implant surface functionalization using Ig-NCAMs to improve the electrodenerve interface in cochlear implants as the best established neuroprosthesis. Thus, two main questions are to be answered. First, which Ig-NCAMs are present in spiral ganglion neurons (SGN; first auditory neurons)? Second, can SGN adhesion be enhanced by surface immobilized Ig-NCAMs? Therefore we first aimed to investigate the presence of several IgNCAMs in the spiral ganglion neurons of neonatal rats (SGN). The presence of NCAM and SynCAM in SGNs was confirmed by RT-PCR and the corresponding genes were cloned into pETSUMO expression vector and the proteins synthesized in bacteria, respectively. The purified fusion his-tagged proteins were immobilized onto nitrocellulose surfaces as a model to evaluate the effect of functionalized surfaces on cell and neurite adhesion of SGN. Effects of coated surfaces on SGNs will be investigated soon by employing immunocytochemistry.



S857

Carbon Nanotubes as Coating-Material for Cochlear Electrodes N. Burblies1*, K. Kranz2*, A. Warnecke2*, H.-C. Schwarz1*, P. Behrens1* 1

Institute of Inorganic Chemistry, Leibniz Universität Hannover, Germany Department for Otolaryngology, Hannover Medical School, Germany * Cluster of Excellence Hearing4all [email protected] 2

Introduction Ensuring the long-term stability and function of cochlear implants is a major research focus in biomedical technology. A chemical modification of the electrode surface should improve the contact of the cochlear electrodes to the nerve fibers and minimize the impedance of the electrical contact. Carbon Nanotubes (CNTs) seem to be a promising material for a biocompatible and conductive coating of these electrodes.[1,2]

Methods As-received CNTs (Fraunhofer IWS, Dresden, Germany) were acid-treated to obtain aqueous CNT-dispersions. These dispersions were used to coat gold- and platinum-coated silicon samples with different coating methods (e.g. doctor blading method, drop or spray coating). The CNT films were investigated with SEM, confocal microscopy, Raman and IR spectroscopy. An electrochemical investigation was carried out by impedance spectroscopy. Cell culture experiments were performed with NIH3T3 fibroblasts.

Results Purification and functionalization of the CNTs via acid-treatment was performed to obtain long-term stable aqueous dispersions. The varying coating methods lead to different film thicknesses of the CNT films. CNT films with a thickness below 100 nm have been produced by applying spray coating, whereas doctor blading and drop coating resulted in films of about half a micron thickness. SEM images show a homogeneous covering of the substrate with CNTs (Fig.1). Electrochemical characterization via impedance spectroscopy shows that CNT coatings reduce the impedance for frequencies below 102 Hz compared to a non-coated noble metal electrode (Fig.2). Cell culture experiments of the CNT films indicate a good biocompatibility of the coatings (Fig.3).

Fig.1: SEM micrograph of CNT-coating on Au-substrate.

Fig.2: Impedance spectroscopy measurements of CNT-coated and uncoated Ausubstrates.

Fig.3: fluorescence microscopic image of NIH3T3 fibroblasts on CNT-coated glass substrates

Conclusion Acid-treatment is an appropriate method to obtain long-time stable aqueous CNT dispersions. These dispersions can be used to prepare CNT films on noble metal substrates. The CNT coatings are a promising approach to reduce the impedance of the electrical contact of the cochlear electrode. Cell culture experiments indicate a good biocompatibility of the the CNT coatings.

References [1] Stöver T., Lenarz T., GMS Curr Top Otorhinolaryngol Head Neck Surg 8:1-22. 2009. Unauthenticated Download Date | 4/21/17 2:48 PM [2] Kotov N. A. et al., Adv. Mater. 21:3970-4004, 2009


S858

Nanoporous Platinum Coatings for the Advance of Electrode Surfaces Kim D. Kreisköther*, Nina Ehlert*, Natalja Wendt*, Hans-Christoph Schwarz*, Institute of Inorganic Chemistry, Leibniz Universität Hannover, Hanover, Germany, Athanasia Warnecke*, Katharina Kranz*, Departement of Otolaryngology, Hanover Medical School, Hanover, Germany, Peter Behrens*, Institute of Inorganic Chemistry, Leibniz Universität Hannover, Hanover, Germany *

Cluster of Excellence Hearing4all [email protected]

Introduction For enhancement of the long-term biointegration and the function of the cochlear implants in the inner ear, the contact between cochlear electrode and nerve fibers has to be improved and the impedance has to be minimized. This can be realized by chemical modification of the surface of the cochlear electrode or by integrating a porous drug delivery system in the platinum sections.[1] Due to the porosity of the platinum coatings the impedance can also be reduced by the increased surface.[2]

Methods Porous platinum coatings were chemical deposited on gold- and platinum coated silicon wafers by dip coating from surfactant-containing Pt4+ solution. Removing of the surfactants and reduction of the platinum were implemented by calcination at 350 C. For electrochemical platinum deposition from an aqueous surfactant-containing Pt4+ solution, a three electrode setup was used. Working electrodes were gold- or platinum-coated silicon wafers, as counter electrode a platinum wire was applied. The reference electrode was an Ag/AgCl system in saturated KCl-AgCl solution. Two different methods, direct and pulse potential electrodeposition was investigated at a constant potential of -0.5 V. The coatings were characterized by SEM, EDX, confocal microscopy, impedance measurements and cell culture tests with fibroblasts.

Results The chemically deposited films consisted of a network of platinum nanoparticles which forms pores with diameters between 50 and 100 nm (Fig. 1). Cell culture experiments with NIH3T3 fibroblasts of the chemically deposited platinum coatings exhibit a good cellcompatibility (Fig. 2). Impedance measurements of the chemically and electrochemically deposited platinum coatings showed a decrease of the impedance in comparison to the uncoated substrates (Fig. 3).

Fig. 1: SEM image of a nanoporous platinum coating by chemical deposition.

Fig. 2: NIH3T3 fibroblasts cell culture test of nanoporous platinum coating by chemical deposition.

Fig. 3: Impedance measurements of uncoated and platinum-coated gold wafers.

Conclusion We were able to deposit porous platinum coatings in two different ways. The chemical deposited porous platinum coatings by dip coating and subsequently calcinations showed good cellcompatibility in cell culture experiments. Both, chemically and electrochemically deposited porous platinum coatings exhibit decreased impedances.

References [1]. Ehlert, N. et al., Chem. Soc. Rev. 42:3847-3861, 2013 [2]. Schlie-Wolter, S. et al., ACS Appl. Mater. Interfaces 5:1070-1077, 2013



S859

Evaluation of an ambisonics system for psychoacoustical measurements in non-anechoic conditions J Heeren1, G Grimm2, V Hohmann3 1 Universität Oldenburg and Cluster of Excellence “Hearing4all”, Germany, [email protected] 2 Universität Oldenburg and Cluster of Excellence “Hearing4all”, Germany, [email protected] 3 Universität Oldenburg and Cluster of Excellence “Hearing4all”, Germany, [email protected] Structure:

1. Introduction / 2. Methods / 3. Results / 4. Discussion / 5. Conclusion

Abstract Spatial hearing experiments require audio reproduction methods with a high spatial resolution. This study has been conducted to investigate the perceptual spatial resolution of an 11 th order ambisonics system with 24 loudspeakers in a sound treated room. Minimum audible angles (MAA) for a speech shaped noise stimulus were measured for three azimuths: 0°, 30°, 60°. The measurements were also performed via headphones by convolution of the loudspeaker signals with appropriate 24 head related transfer functions (HRTF) of an artificial head. The data are largely in line with results from the literature. Median MAA of 2.7° (loudspeaker) and 2.8° (headphones) in the front and 4.3°/7.3° at 60° were observed.

1

Introduction

Virtual acoustics based on Higher Order Ambisonics (HOA) is an interesting method for the reproduction of sounds in psychoacoustical and audiological measurements regarding spatial hearing. For applications with a single listener in the center of a horizontal loudspeaker array spatial properties of sound can be reproduced using a relatively low number of loudspeakers. Correct phase responses for frequencies up to 3000 Hz are possible with 5th order ambisonics, which is sufficient due to the duplex theory [10]. The virtual sound sources can be presented from any azimuth without depending on loudspeaker positions. This is especially useful to simulate moving audio sources. Ambisonic signals can also be rendered for headphone listening by convolution of the loudspeaker signals with head related transfer functions (HRTF). Therefore HOA provides a high comparability between loudspeaker and headphone measurement conditions. In the case of moving audio sources no additional HRTF interpolation is needed. A measure for the perceptual spatial resolution is the minimum audible angle (MAA) according to Mills [8]. An overview is given by Grantham [3]. For white noise this resolution is about 1° MAA in the front (0° reference azimuth) and it increases towards the sides to about 2.5° in the direction of 50°. The MAA is also depending on room conditions. Hartmann [6] observed an increase of 0.9° in a reverberant room compared to a sound treated room. The study was set up as an identification task, which was performed with different stimuli and in different room conditions. The identification task method led to a lower resolution than the MAA paradigm (2.3° for white noise at 0° azimuth). An inversely proportional relation between localization accuracy and the spectral density of the stimuli was found. Except from white noise another two broadband stimuli were tested. A rectangular signal with a frequency of 200 Hz and a duty factor of 10% led to the 2nd

best performance with localisation errors of about 4°. The third example was a sparse tone complex consisting of a 200 Hz tone and 10 selected harmonics, which have at least 1.5 bark spectral distance. It led to localisation errors of about 7°. Localization experiments in reverberant rooms are affected by the precedence effect [12]. Click stimuli from different directions, which are presented with delays of a few milliseconds, get localized at the position of the first click. A relativation of this theory is given by Perrott et al. [9], who conducted MAA measurements for a lag sound source with and without the presence of a lead source for pink noise stimuli. The result was that the presence of the lead source did affect the performance of only some of the participants. So some of them were able to resolve the direction of a simulated reflection. Freyman [2] showed that the detection of a delayed lag sound source varies in the range of 4 – 10 ms over the participants. Echo suppression as suggested by the precedence effect theory increases if stimuli (lead and lag) are presented for a conditioning period before the test token for a localization task is presented. To evaluate a 24 channel ambisonics setup in a sound treated room for speech perception tests a MAA measurement was performed with a speech shaped noise stimulus. The azimuths 0°, 30° and 60° were tested using both loudspeakers and headphones.

2

Methods

1.1

Stimulus

A speech shaped noise signal called “Olnoise” was used. It belongs to the Oldenburg Sentence Test (OLSA) and consists of a mixture of overlapping OLSA sentences [11]. Trials of 0.5 s duration were presented at 65 dB HL. Hanning ramps with a length of 25 ms were applied at the be-



ginning and end of the stimulus to avoid clicks at the onand offsets.

2.3

S860

Loudspeaker setup

0 -5 -10

3rd octave rms / dB

-15 -20 -25 -30 -35 -40 -45 -50

2

10

3

10 center frequency / Hz

4

10

Image 1 3rd octave spectrum of Olnoise (blue, dashed line), pink noise (pink, straight line), 10% rectangular 200 Hz tone (green, crossed line), sparse tone complex (red, dotted line) Image 1 shows the 3rd octave spectrum of Olnoise, pink noise and the other two broadband stimuli, which have been used in the Hartmann experiment [6]. These will be referred to in the discussion. The Olnoise stimulus was chosen because this study was part of a larger study on speech intelligibility with moving sources.

1.1

Procedure

The measurement was set up as 2-alternative-forcedchoice procedure with an 1-up-2-down rule. According to Levitt [7] the probability of positive response at convergence is 0.707. It was performed for the nominal azimuths φ = 0°, 30°, 60°. The azimuth was randomized by +- 5° around the nominal azimuth to avoid identification of the reference stimulus based on spectral cues. One test trial consists of two stimulus presentations: one from the direction φ (reference condition) and one from φ' = φ + α (target condition) with α > 0°. Between the presentations 0.3 s pause time elapsed. The sequential arrangement was randomized regarding the reference condition and the order of φ and φ'. After a trial the participant has to answer the question whether the first stimulus was more to the right than the second or the second was more to the right of the first. Answers were entered via a graphical user interface on a touchscreen. If the participant answered correctly the amount of α was reduced and another trial was presented. In case of a wrong answer α was raised. The stepsize for α was 8° until the first reversal and 4° until the second one. After two reversals the stepsize was fix at 1° and α was averaged until another two reversals were completed. The measurement was performed on both the loudspeaker system and the headphones/HRTF setup in randomized order. One session took about 10 minutes.

Image 2 Ambisonics system with 24 loudspeakers A horizontal 11th order ambisonics system with 24 loudspeakers (Genelec 8020) was set up in an acoustically treated room (Communication Acoustics Simulator, KAS, House of Hearing, Oldenburg, see Image 2). According to Behrens [1] the RT60 of the KAS is almost linear across all frequencies at 0.4 – 0.6 s. Loudspeakers were placed on a circle with a radius of two meters and were orientated towards the center. The spacing between the speakers was 15°. Speaker stands were set to a height of 1.20 m. Ambisonics panning was performed by the software toolbox „TASCAR“ [4,5]. RME ADI 8 DS were used for DA conversion. The frequency and phase responses of the direct path of the loudspeakers to a central measurement microphone were compensated in the range of 400 Hz – 20 kHz. The impulse responses (IRS) were recorded using a Neumann KM183 and a RME Micstasy. The direct-to-reverb ratios of the loudspeakers are in the range of 4.0 – 4.5 dB, thus all loudspeakers are within the critical distance. A directto-first-reflection ratio of 11 dB was observed. Participants were seated on a chair in the center of the loudspeaker array and told to sit in a comfortable position. They were instructed to visually focus on the loudspeaker in front of them (0° azimuth) and to avoid head movements.

2.3

Headphone setup

In the headphone setup loudspeaker signals were convolved with the appropriate HRTF. These were recorded for each loudspeaker using a KEMAR. The frequency response of the headphones (Sennheiser HDA 200) and of the outer ear of the KEMAR were compensated by the impulse response of the headphones on the KEMAR.

2.4

Participants

12 normal hearing subjects in the age of 21 – 42 years participated in the experiment (7 male, 5 female). The mean age was 26 years. All of them had hearing levels of 20 dB HL or better and got paid for participating.



12

3

Results

3.1

Loudspeaker session

MAA / °

10

MAA / °

6

2

10

0

8 6

0°

30°

60°

Image 4 MAA for an 11th order ambisonics via HRTF system (medians and quartile ranges)

4 2

Azimuth 0°

30°

60°

Image 3 MAA for an 11th order ambisonics system with 24 loudspeakers (medians and quartile ranges) Image 3 shows a boxplot of MAA for the three azimuth conditions measured using the loudspeaker setup. Additionally the mean values and medians are displayed in Table 1, each of them with standard deviation. In all conditions the inter-individual deviation is large. For example in the 0° condition the individual performance varies from 0.6° to 9° MAA, although the individual standard deviations are much smaller. The mean individual standard deviation is 0.8°, indicating that participants were able to perform the task. Azimuth

Mean Median (standard deviation) (quartile ranges) 0° 3.4° (± 2.2°) 2.7° (2.1°- 4.4°) 30° 3.5° (± 2.3°) 2.8° (1.8°- 5.0°) 60° 4.4° (± 2.1°) 4.3° (2.4°- 6.6°) Table 1 Average and median MAA across subjects for an 11th order ambisonics system with 24 loudspeakers

3.2

8

4

12

0

S861

Headphone session

In Image 4 a boxplot of the MAA of the headphone experiment is presented. Table 2 shows mean values and medians with standard deviations. In the 30° and 60° conditions the individual performance varies even stronger than in the loudspeaker session while individual standard deviations are in the order of 0.9°.

Mean Median (standard deviation) (quartile range) 0° 3.5° (± 2.2°) 2.8° (1.6°- 5.4°) 30° 6.7° (± 4.4°) 5.1° (3.7°- 8.3°) 60° 7.9° (± 4.7°) 7.3° (5.0°- 8.9°) Table 2 Average and median MAA across subjects for an 11th order ambisonics via HRTF system

4

Discussion

Median MAA values in the loudspeaker condition are in the order of 3 – 4°. This is slightly higher than the literature data [3], which is 1 – 2.5° for white noise in anechoic rooms. This deviation may be caused by three factors: the ambisonics reproduction method, a different stimulus (Olnoise) and not perfectly anechoic conditions. The influence of different stimuli as investigated by Hartmann [6] is depending on the spectral density. Olnoise led to a performance which is at least comparable to the second best example in the Hartmann study: a 10% rectangular 200 Hz tone, which showed a deviation in localization error of 2° from white noise. The sparse tone complex led to a deviation of 5°. As Olnoise has a steady state noise character across all frequency components we assume that it is actually better for localization than rectangular tones, which have a tonal character in low frequencies. Although the room has an almost flat frequency response and a short reverberation time it is likely that there is still a small influence on the localization performance. In the Perrott experiment [9] delays of 2.5 ms led to very different localization performances across participants regarding the influence of a lead sound presence using pink noise. At equal presentation levels they observed MAA deviations of about 5° in an anechoic chamber. The floor reflection in the KAS has a similar delay and with a ratio between the direct sound and the first reflection of 11 dB it may have some influence. In the 0° condition 92% of the values are in the range of 0.6 – 4.5° MAA, which is simlilar to the deviations by Perrott et al. [9]. Freyman [2] reported that the echo suppression proposed by the precedence effect theory is increasing on conditioning presentations for click



trains with several delays. It is yet unclear if this holds for continuous noise. Hartmann [6] measured a mean difference in localization errors of 0.9° between a reverberation room and a sound treated room for white noise. The RT60 of the sound treated room was in the order of 1s, though. The absorbing condition may already be influenced by reflections as well. For the side azimuths (30°, 60°) median MAA are higher than in the front (0°), which is normal due to the literature data. This effect is much stronger in the headphone condition (4.5°) than in the loudspeaker condition (1.6°). For each participant about the double value of the loudspeaker condition MAA was observed. At 0° the results for the two conditions are equal. The data of all participants show the same trend. Comparable results were measured by Yost et al. [13]. The same behavior was observed.

5

Conclusions

The perceptual spatial resolution of an 11th order ambisonics system was investigated by an adaptive measurement of the minimum audible angle (MAA). It was shown that the MAA is between literature values for anechoic presentation and in real rooms. With binaural reproduction of the same stimuli a slightly larger MAA at azimuths except 0° was achieved. The results indicate that the influence of the 11th order ambisonics system on the MAA is negligible in the given context. Thus the tested system seems to be suitable for applications in the hearing research with high spatial resolution requirements.

S862

[7] Levitt H.:Transformed up-down methods in psychoacoustics, J. Acous. Soc. Am., 49, 467-477, 1970 [8] Mills A.W.: On the minimum audible angle, J. Acous. Soc. Am., 30, 237-246, 1958 [9] Perrott D.R., Marlborough K., Merrill P.: Minimum audible angle thresholds obtained under conditions in which the precedence effect is assumed to operate, J. Acous. Soc. Am., 85, 282-288, 1988 [10] Rayleigh, Lord: On our perception of sound direction, Phil. Mag., 13, 214-232, 1907 [11] Wagener K., Brand T., Kollmeier B.: Entwicklung eines Satztests für die deutschen Sprache I: Design des Oldenburger Satztests, Z. Audiol., 38, 4-15, 1999 [12] Wallach H., Newman E.B. and Rosenzweig M.R.: The precedence effect in sound localization, J. Psychol, 12, 315-336, 1949 [13] Yost W.A., Dye R.H., Sheft S.: A simulated “cocktail party” with up to three sound sources, Perception & Psychophysics, 58, 1026-1036, 1996

Work funded by DFG FOR1732

6

References

[1] Behrens, T.: Der ‚Kommunikations-AkustikSimulator’ im Oldenburger ‚Haus des Hörens’, in Fortschritte der Akustik - DAGA 2005, DEGA e.V., Düsseldorf, 2004 [2] Freyman R.L., Clifton R.K., Litovsky R.Y.: Dynamic processes in the precedence effect, J. Acous. Soc. Am., 90, 874-884, 1991 [3] Grantham D.W.: Spatial hearing and related phenomena, Hearing, Academic Press inc, Chapter 9, 1995 [4] Grimm G., Coleman G., Hohmann V.: Realistic spatially complex acoustic scenes for space-aware hearing aids and computational acoustic scene analysis, in 16. Jahrestagung der Deutschen Gesellschaft für Audiologie, pp. CD-Rom, 4 pages, DGA, Rostock, 2013 [5] Grimm, G., Wendt, T., Hohmann, V., Ewert, S.: Implementation and perceptual evaluation of a simulation method for coupled rooms in higher order ambisonics. Proceedings of the EAA Joint Symposium on Auralization and Ambisonics., in press, 2014 [6] Hartmann W.M.: Localization of sound in rooms, J. Acous. Soc. Am., 74, 1380-1391, 1983



S863

SSVEP measurements for BCI applications with higher frequencies based on capacitive EEG Olze, K., Su, Y., Schilling, M. Institut für Elektrische Messtechnik und Grundlagen der Elektrotechnik, TU Braunschweig, Braunschweig, Germany [email protected]

Introduction For non invasive EEG capacitive electrodes are an alternative way to standard Ag/AgCl electrodes, not only for medical applications but also in the field of Brain Computer Interfaces (BCI). With our prototypes of capacitive helmet systems, the measurement of steady-state visually evoked potentials (SSVEP) is routine procedure since many years.

Methods For SSVEP stimulation flickering chessboard patterns (8-18Hz) are used. However, in this range many disturbances caused by e.g. the µ waves of the motor cortex, the electronics itself and unknown external sources are observed. Thus, the signal noise ratio (SNR) is affected and the brain signal triggered by the SSVEP can sometimes not be clearly identified. To meet this challenge the stimulation frequencies have been extended up to 40Hz. In the range of 19Hz to 40Hz the noise level is much lower, nearly constant and less disturbances are observed. The frequencies were chosen accordingly to the monitors refresh rate. Very promissing measurements of recognition rates in a spelling application (12s per decision, no averaging) are reported.

Results First results confirmed the assumption that brain signals evoked by higher frequencies can be identified more reliable due to the better SNR in this range. In some cases the recognition rate of the spelling machine could be raised up to 100%. However, not all subjects showed equal reactions to the higher frequencies.

Conclusion To increase the number of validated applications for BCI with capacitive EEG, it is useful to shift the stimulation frequencies in a range with low white noise and without external disturbances. However, the EEG of some subjects showed reduced response at these higher stimulation frequencies.



S864

Laser-generated, catalytic nanoceria as efficient ROS scavenger A. Barchanski, C. L. Sajti, B. N. Chichkov Nanotechnology Department, Laser Zentrum Hannover e.V., D-30419 Hannover, Germany Email: [email protected]

Introduction Cerium oxide (ceria) is frequently applied as material for catalytic converters due to it’s outstanding characteristic to switch regeneratively between the oxidation states +III and +IV. In nanoparticulate form (nanoceria) the catalytic function may be capitalized to scavange free radicals like reactive oxygen species (ROS). As the conventional nanoceria fabrication by chemical reduction methods often reduce the particles’ biocompatibility and catalytic properties, a laserbased fabrication is presented as attractive and “green” alternative, profiting from the adoption of cerium target, ultrapure water and laser light exclusively, while delivering colloids with maximal material purity.

Methods The technique of picosecond-pulsed laser ablation in liquids (ps-PLAL) is adopted for nanoceria fabricaton. The particle characterization is performed by photospectroscopy (UV-Vis), Fourier-transform infrared spectroscopy (FTIR), Xray diffraction (XRD) and electron microscopy (SEM), while their catalytic behaviour is demonstrated by hydrogenperoxide treatment and 2 ,7 -dichlorofluorescin assay for ROS quantification in vitro. Nanoceria functionalization was accomplised by the method of in situ biofunctionalization during ps-PLAL.

Results The ablation of cerium target by ps-PLAL delivered spherical nanoparticles with a mean size of 80 nm, which were clearly identified as pure ceria. The autoregenerative redox cycle between Ce+III and Ce+IV oxidation state was visualized by colloidal coloration upon hydrogenperoxide treatment followed by subsequent decoloration. Efficient ROS scavenging effects of nanoceria were proven by in vitro studies, showing significantly enhanced catalytic activity than obtained with commercial nanoceria. Finally, the surface functionalization with polymers was successfully achieved by in situ bioconjugation, enabling the flexible tuning of nanoceria for biomedical applications.

Conclusion Nanoceria were fabricated by pulsed laser ablation for the first time and featured excellent catalytic properties. They may provide an interesting ROS scavenging tool for the treatement of ROS-linked diseases like diabetes, neurodegenerative disorders or cancer, especially if linked to functional moieties.



S865

Are Smartphones Applicable for Posturography? A Classification Analysis Based on Computational Intelligence D. Sommer1, A. Schenka1, J. Krajewski2, M. Golz1 1 Faculty of Computer Science, University of Applied Sciences Schmalkalden, Germany, [email protected] 2 Work and Organizational Psychology, University of Wuppertal, Germany

Introduction Comprehensive devices of the IT mass market such as smartphones are characterized by multi-sensor equipment, high computational power and relatively low costs. Over the last 5 years they have found increasing applications in biomedical engineering. We examined the question if the internal 3-axes acceleration and the 3-axes angular velocity sensor signals can be processed by computational intelligence algorithms in order to have quantitative indications of standing stability of patients.

Methods A pilot study involved 12 volunteers. Postural sway was recorded with and without visual feedback in Romberg pose. Smartphones at 3 different locations, on the head, on the palm, and at the hip, recorded 6 different time series of both internal 3-axes sensors. Simultaneously, the center of foot pressure (COP) was recorded by a posturography platform. COP consists of 2 time series, the antero-posterior and the mediolateral component. Signals were pre-processed and spectral domain features were extracted. For COP also distinctive time domain features were extracted. Feature sets were processed by machine learning algorithms, namely the Learning Vector-Quantization and the Support Vector-Machine. Parameters of all algorithms were optimized empirically by searching for minimal test errors. They were estimated by cross-validation using repeated random subsampling.

Results Posturography resulted in mean test errors of 16.5 ± 7.6% (N = 40). Smartphone measurements on the palm led to 16.5 ± 9.9% (N = 40). At the two other locations they were 3.6% and 7.8% worse.

Conclusion The measurements of both device types are different from a biomechanical point of view, because the COP depends on forces. On the one hand the force of the center of gravity and on the other hand the regulating foot forces. In contrast, smartphone measurements represent acceleration and velocities of an inverse physical pendulum, which is more or less rigid. A precise inference of the center of mass location is hardly achievable by both methods. Signal processing using model-free methods of computational intelligence led to comparable classification accuracies. Higher standard deviations of smartphone results can be explained by the larger influence of disturbances. Presented results give rise to an extended study in order to confirm the equivalence of both measurement types.



S866

Tactile Textile Sensors for Collision Avoidance Verena Schmidt Siemens Healthcare, Kemnath, Germany, [email protected]

Abstract A safety system against collisions is needed when using medical devices with moving axis. One distinguishes between collision detection and collision avoidance. This paper shows the possibilities of tactile textile sensors for postcollision-detection. A special focus lies on safety under single fault conditions as well as the challenges when implementing a safety function either in hard- or software.

1

Why is a Safety System for Medical Devices needed?

Medical devices like devices for fluoroscopy, angiography or radiography have motor-driven axis which partly move automatically. This presents a safety risk for patients, operators, service people and even other technical devices. Image 1 shows a fluoroscopy device. An anti-collision system is used for increasing the safety of both, operators and patients.

anti-collision system

The task of anti-collision systems is to increase the safety of the operator and patient and at the time making the best usage of the available space. In the daily routine the anticollision systems are even used for adjusting the medical devices as near as possible to the patient (see Image 3).

Image 3 Device contact with patient for improved view (source: Siemens picture database)

2

Image 1 Fluroscopy Device (source: Siemens picture database) During complicated surgeries, e.g. catheter intervention at blood vessels or at the heart, the operation theatres are crowded by many people and different medical devices. (see Image 2).

Currently, medical devices use mainly tactile sensors for collision detection of surfaces. Most principles are based on data fusion of several single sensors which deliver a local signal or using mechanical constructions. As the areas which have to be observed are normally surfaces, the next step is to build up simple surface sensors within one technology. Alternatively, contact-free sensors can be used. However, their usage is not yet wide-spread as meeting the safety requirements is quite challenging and leads to a high complexity or costs.

3

Image 2 Crowded operation theatre (source: Siemens picture database)

State of the Art

Collision Avoidance vs. Collision Detection

There is a difference between collision detection and collision avoidance. When regarding the former, a collision occurs which is detected by tactile sensors (post-collisiondetection). In case of the latter, sensors are used which detect a collision before it actually happens (pre-collisiondetection). Consequently, different sensor principles are needed for both cases. Post-collision-detection can be realized with strain gauges, switches or load cells. Pre-



collision-detection uses contact-free sensors like ultrasonic, light curtain, cameras or capacitive sensors. When choosing the sensor principle, the safety function within the device, the reliability as well as the meeting of normative requirements are of utmost importance. The industry defines this with the safety integrity level (SIL) which is not 100% compatible with the requirements according to EN 60601-1 [1]. When comparing post- and pre-collision detection regarding safety levels it is vital that collision-detection systems have to detect the collision under all circumanstances and consequently stop the machine. In most cases this is comparable with SIL 2.

S867

In the next step possible sensors have to be evaluated regarding the safety function. For instance, ultrasonic sensors do not detect 100% of all objects in every situation. This is due to different reflection properties of the material or blind angles for instance. That is why ultrasonic sensors are only a good choice for pre-collision avoidance. Light curtains are an example for pre-collision avoidance systems which do not need an additional sensor. However, they are quite expensive and the constructional requirements cannot always be met. Siemens Healthcare uses different types of tactile sensors for collision-detection, like foam including fiber optics, industrial pressure sensitive mats, pressure sensitive pumpers or other mechanical constructions. The disadvantage of the current solution is the complexity and the costs. That is the reason for investigating textile tactile sensors.

4

Requirements for collision detection systems

Robustness Design and Colour

Image 4 Safety Concept using only post-collision When using post- and pre-collision-detection in one system, the requirements for the pre-collision-detection are not that high any more as the post-collision-system also serves as a backup system. The pre-collision system can improve the workflow e.g. by allowing a better exploitation of the limited space.

Desinfection of Surface Requirements for Collision Detection

COSTS

Technical Safety

Image 6 Requirements for Collision Detection Image 6 shows the different requirements for collision detection systems in medical devices. Most important is the safety followed by the costs. A special challenge is the ability to disinfect the surface. This leads to special requirements regarding cleaning agents as well as the surface texture and geometry. For instance small gaps are difficult to disinfect. Especially for textile sensors the edges have to be designed accordingly. Image 7 shows a bad example.

Image 5 Safety Concept using both pre- and postcollision



S868

low. In the former case, the sensor might not detect a collision in the latter case, the machines stops unintentional by the smallest contact.

Image 7 Edge which is difficult to disinfect. Additionally, the surface has to be robust towards scratches or deformation. Finally, the industrial design has a big influence regarding the acceptance.

5

Possible Physical Principles

There are different physical principles possible for building up a tactile sensor: • Resistive – Option 1: Change of thread resistance by $ distension → R = ρ % • Resistiv – Option 2: two conductive layers of textiles or threads are separated by a perforated isolator → Collision leads to a short cut of both textile layers • Capacitive: A collision changes the distance d and thus the capacity → & = '( ∗ '* ∗ •

•

6

+ ,

Inductive: Threads are enwinded by magnetic material and build up a coil. The sensor consists of two layers with coils which are separated by an isolator. By changing the distance of both coil-layers a current is induced. Pressure-Principle: In this case a plastic hose with an integrated knit fabric is used. The latter is used as crush zone. A collision changes the pressure within the hose which is detected with an according sensor.

Textile Sensors for Collision Detection

For monitoring bedsore patients, smart textiles are already used in combination with pressure sensors or directly as pressure sensors. It is easy to build up a surface sensor with the help of textiles. Therefore, tactile textiles are good solution for collision detection. Currently, several demonstrators based on different physical principles are evaluated by Siemens. In the following a demonstrator using a resistive principle is introduced and the challenges are described. The resistive sensor consists of two conductive tissues which are separated by a perforated foam (see Image 8). The material parameters of the foam define the switch force. The number of holes and theirs diameters have also an impact on the switch-force as well as the local resolution. The switch force must neither be too high nor too

Image 8 Tactile Textile Sensor

7

Safety under Single Fault Conditions

A failure of the collision detection systems can lead to severe or even lethal injuries. Therefore, post-collisiondetection has to be safe under single fault conditions. Single-Fault-Safety according to EN 60601-1 [1] means: • Medical devices must be developed in such a way, that no single fault leads to an unsafe device during its lifetime • Medical devices are are considered safe under single fault conditions, if o a single fault leads to the activation of the safeguard o a single fault is detected and a clear signal is given to the operator o a single fault is detected by regular service and then repaired Taking the resistive textile tactile sensors as an example: First, the different failure possibilities are discussed in an interdisciplinary team, e.g. during a FMEA (Failure Mode and Effect Analysis). In order to do this an exact knowledge of the working principle is need (see Image 9). The red lines indicate the conductive tissue while the grey



S869

boxes represent the perforated foam. If both tissues come in contact, a short cut between the tissues occurs, which can be measured.

Image 9 Active Principle of the sensor. The following faults have to be analyzed: Single fault which leads to a safe condition: ! An example for this case is a permanent short cut between to wires of both tissues which leads to the activation of the safeguard system. The system stops and thus is in a safe condition. Reduction of the risk: ! As many filaments are across a single hole, even with a broken wire, the system can detect a collision by the other filaments. The systems works as normal Regular Service ! A defective contact cannot be safely detected. Therefore, redundancy is required. Additionally, during regular service the contacts have to be checked in order to detect the error and change them. This inspection has to take place before a second contact might break and the system becomes unsafe.

8

Future Prospects

The usage of pre-collision-sensors regarding the safety is more complex. Therefore, the safety concepts will be most likely a combination between pre- and post-collision-systems, using data fusion to enhance safety. In other industrial sectors, like automation and robotics, the collision detection and avoidance is more advanced. However, the costs are considerably higher than acceptable for medical devices. As an example serves a robot which uses near-field sensors combined with tactile sensors in order to realize emergency stop. [2]

9

References

[1] EN 60601-1:2006, Medical electrical equipment - Part 1: General requirements for basic safety and essential performance [2] Product Description KR 5 SI, MRK-Systeme, http://www.mrksysteme.de/produkte_interaction.html, 14.11.2013



S870

Coating of stents with poly-L-lactide via fluidized-bed technology M. Wentzlaff1, A. Seidlitz1, C. Harder2, N. Grabow3, W. Weitschies1 1 Institute of Pharmacy, University of Greifswald, Greifswald, Germany 2 Biotronik AG, Bülach, Switzerland 3 Institute for Biomedical Engineering, University of Rostock, Germany [email protected]

Abstract Stents are net-like metal-tubes used for the treatment of arteriosclerotically narrowed blood vessels. Bare-metal stents often cause restenosis whereas Drug-eluting stents, which are typically coated with a drug/polymer mixture reduce the incidence of restenosis. However, commonly used single coating techniques are associated with a small output and high loss of coating liquid resulting in hight production costs. To overcome these limitations, simultaneous batch coating of large quantities of stents in a fluidized-bed apparatus is pursued. We were able to show previously, that stent coating was feasible using fluidized-bed technology with a waterdispersable polymer. Nevertheless, polymers that are commonly used for coating of stents are processed as a solution of the respective polymer in organic solvents. To verify the performance of the fluidized-bed technology for stent coating with organic polymer solutions poly-L-lactide was used as a model polymer. Despite the high process pressure which was needed to fluidize the stents it was possible to generate homogenous coatings with a high degree of integrity within a comparable short processing time and with a high deposition rate.

1

Introduction

Stents are meshed hollow cylinders used for the treatment of arteriosclerotically narrowed vessels. Bare-metal stents often cause re-narrowing (in-stent restenosis) of the treated vessels, which is commonly treated by re-stenting. The development of stents coated with antiproliverative drugs (Drug-eluting stents) lead to a reduction of the incidence of in-stent restenosis [1]. However, coating of stents is often associated with high costs, partially due to a high loss of coating liquid and small output of single coating technologies [2]. To overcome these limitations, simultaneous batch coating of large quantities of stents using the fluidized bed apparatus is pursued. The fluidized-bed technology is a well-established method used in pharmaceutical industry for coating of small dosage forms with a high coating homogeneity. Therefore, the coating goods are placed in a product container with a slotted bottom plate that allows for air distribution. After the coating goods become randomly fluidized in the airstream the coating liquid is injected via spraying until the desired coating thickness is achieved [3]. In previous studies the fluidized-bed technology has proven suitable for coating of stents using the waterdispersable Eudragit®RS30D as a model polymer [4]. Nevertheless, some difficulties are associated with coating of stents using this technology. To name an example, the cylindrical geometry of the stents can result in bad fluidizable behaviour [5] because the stents tend to agglomerate. Consequently, a high air flow is needed to create a stable fluidized bed of the coating good. Commonly used polymers are typically processed as a solution in organic solvents where the high air flow results in a high evaporation rate of the organic solvent, which in turn may cause incomplete

spreading of the coating droplets on the product or even spray drying. Therefore, the applicability of this coating technique maybe limited to waterdispersable polymers. Consequently, the aim of this study was to check whether the fluidized-bed coating technology is suitable for coating of stents with low defects and high uniformity using an organic solution of poly-L-lactide (PLLA). PLLA is a biodegradable polymer used for drug delivery implants [6,7]and has proven to be suitable for single-spray coating procedures of stents [8].

2

Methods

2.1

Coating process

The Mini-Glatt (Glatt GmbH, Germany) fluidized-bed apparatus equipped with the MikroKit product container, the Wurster bottom plate and a spray nozzle (Düsen-Schlick GmbH, Germany) with an orifice diameter of 0.3 mm was used to coat 50 Bare-metal stents (lenghth 15 mm, diameter ~1.6 mm) in bottom-spray fluidized-bed process. 250 steel springs of comparable measures were used as filling material (length 14.6 mm, diameter 2.1 mm, Gutekunst &Co.KG, Germany). At the end of the process the entire coating mass was determined via differential weighing of the entire product before and after the coating procedure. The coating liquid consisted of the poly-L-lactide Resomer®L210S (Evonik Industries AG, Germany) and the fluorescent model substance quinine (Sigma-Aldrich GmbH, Germany) which were solved in a mixture of dichloromethane and methanol. For this purpose Resomer®L210S was dissolved in pure dichloromethane and quinine was added. After dissolving the solids methanol



was added under continuous stirring. The composition of the coating liquid is shown in table 1. Tab. 1: Composition of coating lacquer. Substance type Substance name Quantity Polymer Resomer®L210S 0.7 g Model substance Quinine 0.3 g Solvent Dichloromethane 94 mL Solvent Methanol 47 mL The identified most suitable process parameters are described in section 3.1.

2.2

Evaluation of coated devices

To examine the homogeneity of the coating distribution 20 randomly chosen coated stents were weighed before and after the coating process. To verify the coating homogeneity 10 randomly selected stents were studied by fluorescence microscopy (BZ-8000, Keyence Corporation, Japan, λex 360 nm, λem 460 nm). Additionally, 3 randomly selected stents were inspected by scanning electron microscopy (SEM, Phenom, FEI Company, USA). To investigate the integrity of the coating after expansion 3 randomly selected stents were expanded in dry state to a total diameter of 3.5 mm using a balloon catheter and catheter pump (7 atm, 10 seconds). Afterwards, the stents were examined microscopically using fluorescence microscopy. Additionally, to investigate the impact of the expansion procedure on coating adhesion and integrity 3 randomly selected stents where expanded in a moist artificial vessel lumen as previously described [9]. Therefore, a thin calcium alginate film (500 µm) was enrolled on a stainless steel rod (diameter of 3 mm). Afterwards, the rod was removed and the catheter with the stent mounted on it was pushed into the lumen and inflated in the way described above. The calcium alginate film was then removed and the stents and the gel-compartment were examined via fluorescence microscopy for detached coating fragments. To determine the in vitro drug release 4 randomly selected stents were expanded in dry state and incubated in 10 mL of phosphate buffered saline pH 7.4 at 37 °C on an incubative shaking device (300 rpm; Inkubator1000 and Titramax1000, Heidolph Instruments, Germany). To maintain sink conditions the release medium was completely replaced during sampling. The quinine content of the incubation buffer was determined fluorometrically (Varioskan Flash, Thermo Scientific, USA; λex 326 nm, λem 382 nm).

3


3.1

Coating process

Preliminary tests showed that due to the Wurster-bottom a fountain-like product movement is induced which results in reduced impact of the stents on the wall of the apparatus and consequently in reduced defects of the stents. However, the stents and filling materials tend to form an air chan-

S871

nel when using this set-up, which may result in a collapse of the fluidized-bed. Consequently, a high process pressure of 0.3 bar (35 Nm³/h) was needed to achieve a stable fluidized-bed of the cylindrical stents. This resulted in a comparably high drying air capacity (0.12 Nm³/ (min•g)). A spraying pressure of 1.1 bar was needed to generate droplets which were fast enough to realize a good spreading. However, the high spraying pressure may results in small droplets, which in turn could cause spray drying. Therefore, a spraying rate of 2.1 g/min/4.8 g product was needed to generate bigger droplets. Furthermore the product temperature should be kept as low as possible to avoid rapid evaporation of the solvent. Consequently, an inlet temperature of 30 °C was selected. Based on these observations the final process was performed at a process pressure of 0.3 bar, a spraying pressure of 1.1 bar, an inlet air temperature of 30 °C and a spraying rate of 2.1 g/min per 4.8 g product.

Figure 1: Process parameters during the coating process, arrows indicating the beginning and end of coating procedure. In Figure 1 the coating process flow is given, whereby the first arrow indicates the end of the heatup phase and the beginning of the coating phase and the second arrow the end of the coating phase. The rather short phases of heatup and final drying should protect the stents against additional impact- related defects. As the coating procedure is a random process the coating time should not be chosen too short to realize a homogeneous coating distribution. Therefore, a low polymer content of the coating liquid is needed to ensure that the coating time is long enough. This becomes particularly important for smaller batch sizes. Also, the low polymer content of the coating liquid provides a good spreading of the coating droplets even though the drying air capacity is rather high. Even though the coating process was performed with a rather small batch size a deposition rate of 33 % was achieved.



3.2

Coated stents

The average mass of the uncoated goods was 16.945 ± 0.738 mg and the average mass of the coated goods was 17.959 ± 0.737 mg. In figure 2 the average masses of the uncoated and coated stents are depicted. The dotted lines represent the maximum and minimum. The similar standard deviations demonstrate that the variations of the average mass after the coating procedure were approximately the same compared with the variations of the average mass of the uncoated goods.

S872

Nevertheless, the coating was spread over the entire surface of the stent. In addition the stent surface featured a fine-grained structure (fig. 4).

Furthermore, all individual masses of the coated stents were within ± 10 % of the average mass. These observations demonstrate a high homogeneity of the deposited masses.

Figure 4: Representative SEM images of a stent coated with Resomer®L210S.

Figure 2: Average mass of 20 uncoated and 20 coated stents, dotted line indicating the maximum and minimum.

After expansion in dry state (fig. 5, top) some defects of the rather stiff coating layer where observed especially at the strut bends. These bends are the points of largest mechanical deformation during expansion and the polymer has to be very flexible to avoid cracking in this area. We supposed that after expansion in a moist gelcompartment detached coating fragments or loosely sticking spray dried particles will attach to the gel and can be observed microscopically. As visualized in figure 5c just a very small part of the fluorescent model substance can be detected on the gel layer even though the polymer film showed cracks at the strut bends. This indicates the coating presents a high degree of adhesion despite the expansion defects. Furthermore, the absence of loose sticking particles is confirmed.

Based on the given surface area of the stents an average coating mass per area of 12.7 µg/mm² was calculated. Due to the rather thin coating layer it is conceivable that most of the drug is located near the surface which may lead to a high initial release rate of the poorly soluble drug.

Figure 3: Representative fluorescent microscopic image of a stent coated with Resomer®L210S and quinine (left), focus on small bridges at a strut bend (red frame). As shown in Figure 3 the coating showed some defects such as small bridges at some strut bends and a few cracks at the ends of the stents. Moreover, the coating was thicker at the ends of the stent. The use of dry inlet air and waterless coating liquid leads to electrostatic charging of the coating goods and coating droplets during the coating process. A higher electrostatic field strength can be observed at the ends of the stents. Consequently, a higher deposition rate of the coating droplets can be detected at the endings.

Figure 5: Fluorescence microscopic images of stents inflated a) in air, b) inside an artificial-vessel and c) image of the artificial-vessel-compartment; red framed pictures focus on cracks at some strut bends.



As suspected most of the drug was released within a short time frame (figure 6). In particular nearly 80 % of the total amount was released within the first 24 hours which indicates that the majority of drug particles is located near the surface. Excluding the initial burst on day one the release was linear when plotted versus the square root of time (insert of figure 6) which is consistent with a diffusion controlled drug release. As also shown in figure 6 one of the tested stents featured a slightly higher total drug release due to higher drug content.

Figure 6: Cumulative drug-release profile of 4 stents in phosphate buffered saline pH 7.4. The insert shows the average cumulative drug-release from day 1 to 14 plotted versus the square root of time.

4

Conclusion

This case study indicates that coating of cardiovascular stents with polymers solved in organic solvents via fluidzed-bed technology is feasible and can offer an alternative to single-coating techniques. It was possible to adapt the coating parameters to the special requirements of stents on the one hand and to the requirements of the coating laquer on the other hand. The coating featured a high degree of integrity and homogeneity. At this point some slight irregularities in the coating layer can be oberserved and further improvements would be favourable in this regard. In this case study a mixture of 250 steel springs and 50 Bare-metal stents was coated within 80 minutes which is equivalent to a coating time of less than 20 seconds per piece.

5

S873

References

[1] Trevor Simard, Benjamin Hibbert, F. Daniel Ramirez, Michael Froeschl, Yong-Xiang Chen, Edward R. O'Brien, The Evolution of Coronary Stents: A Brief Review.Canadian Journal of Cardiology, 30 (1), 3545, 2014. [2] N. Grabow, M. Wentzlaff, V. Senz, A. Seidlitz, C. Harder, K. Sternberg, W. Weitschies, K.P. Schmitz: Feasibility of polymer/drug coating on absorbable and permanent stent platforms - technological challenges. Biomed Tech (Berl), 2013. [3] D. Jones: Air suspension coating for multiparticulates; Drug development and industrial pharmacy, 20 (20), 3175-3206, 1994 [4] M. Wentzlaff, A. Seidlitz, V. Senz, N. Grabow, C. Harder, K. Sternberg, W. Weitschies: Investigating the Applicability of Fluidized-Bed Technology for HighThroughput Coating of Stents Biomed Tech (Berl), 2013. [5] D. Geldart: Types of Gas Fluidization; Powder Technology, 7, 285-292, 1973. [6] Tamai, K. Igaki, E. Kyo, K. Kosuga,. Kawashima, S. Matsui, Hidenori Komori, T. Tsuji, Seiichiro Motohara, H. Uehata:Clinical Investigation and Reports: Initial and 6-Month Results of Biodegradable Poly-lLactic Acid Coronary Stents in Humans; Circulation. 102:399-404, 2000. [7] Verheye S, Ormiston JA, Stewart J, et al.: A NextGeneration Bioresorbable Coronary Scaffold System: From Bench to First Clinical Evaluation: 6- and 12Month Clinical and Multimodality Imaging Results; J Am Coll Cardiol Intv ;7(1):89-99.2014. [8] K.Sternberg, S. Kramer, C. Nischan, N. Grabow, Th. Langer, G. Hennighausen, K.-P. Schmitz: In vitro study of drug-eluting stent coatings based on poly(llactide) incorporating cyclosporine A- drug release, polymer degradation and mechanical integrity; J Mater Sci: Mater Med 18: 1423-1432,2007. [9] Seidlitz A, Kotzan N, Nagel S, Reske T, Grabow N, Harder C, Petersen S, Sternberg K und Weitschies W: In vitro determination of drug transfer from drugcoated balloons. PLOS ONE; 8(12): e83992, 2013.

Acknowledgement This project was funded by the Federal Ministry of Education and Research (BMBF) within REMEDIS.



S874

Development and Evaluation of a Drug Multiplexing Infusion System Saif Abdul-Karim1, Jörg Schroeter1, Bodo Nestler1 1 Medical Sensors and Devices Laboratory, Luebeck University of Applied Sciences, Luebeck, Germany, [email protected]

Abstract The objective here is to develop an intravenous (IV) infusion system to significantly reduce medical errors related to drug administration. In so doing, consideration is given to fact that the drugs being applied to the patient should not mix in the delivery catheter to avoid unwanted reactions. The concept employed here (Drug Multiplexing) uses gaseous media like CO2 to separate the flowing drugs inside a single delivery catheter, which then goes to the patient. Before entering the patient’s circulatory system, the gaseous medium is removed from the end of the catheter. A new infusion system modality was developed. The aim to separate the drugs with gas bubbles was not only attained but the cumulative data from testing endorsed the viability of such a concept.

1

Introduction

1.1

Problem Definition

Medical errors are a leading cause of death in hospitals. About 210.000 are killed by preventable hospital errors each year in the U.S [1]. A body of research in medical human factors has begun to emerge investigating how medical equipment can be designed to minimize errors and increase efficiency. However, relatively few studies have addressed common usability issues and design solutions of infusion pumps as complex drug delivery systems. IV infusion pumps are medical devices that deliver medication and fluids more accurately, efficiently, and safely than adjusted gravity systems [2]. The improper use of IV infusion pumps can contribute to patients suffering complications. Errors during the infusion process are a significant source of harm to the patient. Between years 2005 and 2009, more than 56.000 infusion pump incidents were reported, including 710 deaths [3].This problem is further exacerbated when multiple infusion pumps are used at a time, as the administration of the wrong drug to the wrong patient or in the wrong way may occur, giving multiple drugs that interact negatively. This will result in what called Adverse Effect (AE). AE may cause a reversible or irreversible change, including an increase or decrease in the susceptibility of the individual to other chemicals, such as drug interactions. Chemical incompatibility means that the drug may be chemically degraded, due to oxidation, reduction, hydrolysis, or decomposition. Chemical reactions can manifest themselves through turbidity, precipitation and color changes. As a consequence, the amount of the active agent decreases and/ or form toxic by-products [4]. Multi-infusion setups for medication administration, e.g. in Intensive Care Units (ICUs), seem uncontrolled due to flow rates and pressure differences between syringe pumps [5]. Furthermore, the more catheters are used, (e.g. when moving patient be-

tween operation rooms and ICU), higher the chance for hygiene problems to arise. This particular problem is more associated to patient transportation. Essentially, because surviving bacteria in air can be transported along with infusion lines and mix with drugs, especially when dis/reconnecting of catheters, and then reach patient’s blood stream. The potential risk factor, clearly, increases as higher numbers of infusion lines are used, and vice versa [6].

1.2

Motivation

The idea of supplying patients with different medications through a single catheter is not new. Since the year 1977, there have been various patents on this topic [7 - 19]. The aim is to design and test a miniaturized IV delivery station. By inserting only a single catheter and applying a proper separation method between flowing drugs, the total volume of the delivery station will be miniaturized. Conceptually, the IV infusion related risk could be minimized. Using a gas bubble of some microliters volume as separator is far from values that might cause harm. From case reports of accidental intravascular delivery of air, the adult lethal volume has been described as between 200 and 300 ml [20].

2

Methods

2.1

Design Concept

The concept of Drug Multiplexing aims to provide a high miniaturized medications (and fluidics) delivery station, from which dosages with high precision (tolerance of ± 2 % for 1 μl dosing) could be delivered through a single thin catheter conveyed and inserted into patients, see Fig. 1.



S875

measurements showed that, for example with a Polyvinyl Chloride (PVC) catheter of 1 mm inner diameter, the set volumes of gas bubbles remains constant but with mean error of about 4% [21]. Table 1 shows the air volumes reproducibility using a flow source for air bubbles generation.

Fig. 1. The concept of Drug Multiplexing employs gaseous media as separator between flowing drugs. The setup consists basically of four systems. 1) A 4channel dosing unit was used for delivering the mimicked drugs. 2) A selection valve was used for orderly flow of drugs. 3) A matching electronic system to synchronize work of the syringe pumps with the selection valve was designed and implanted. 4) A software console to steer and control the overall work flow was considered to be the fourth part of the test bed. The produced volumes from the dosing units were measured at the end of the catheter using a flow sensor. The measurements showed a slight difference (±2%) between the dosed and received volumes. It was sufficient to demonstrate the concept and feasibility of a test bed.

3

Results

3.1

System realisation

The design of Drug Multiplexing infusion system was realised and developed. The first design is a didactic laboratory platform, which task is to enable investigations and gather knowledge of construction and the way of working of a new setup. It is then further developed for investigate Drug Multiplexing principle extensively. The system connected multiple infusion pumps to one catheter via a rotary valve which then administered the drugs successively. This was done by employing a 4-channel syringe pump (Cetoni, Germany) to generate flow. A 6-way selector valve (Biochem Fluidics, U.S) was used to select which drugs flowed through the catheter at a given time and in which order. Gas bubbles of some microliter volume were applied, both before and after any drug was sent through the catheter, to act as a cleaning medium and a separator between each drug. This ensured no mixture between the drugs [21]. The Drug Multiplexing setup is shown in Fig. 2.

3.2

Test of setup

3.2.1 Air bubbles generation System ability to reproduce the desired volumes of air was tested. Using one of the dosing units as flow source for generation of equal volumes of air bubbles was not possible. In addition, air bubbles were fragmented into two smaller bubbles through the catheter. Quantitative

Fig. 2. Test Bed for Drug Multiplexing connected to an arm model. Flow of drugs is generated by dosing units, while the order of flow is controlled by the selection valve.

3.3

Measurements

3.3.1 Air volumes A pressure source was added to the setup for better reproducibility of constant air bubbles volumes. A pressure regulator (17 bar in and 0.01 – 1.0 bar output range) (AirCom Pneumatic, Germany) provided small ranges (about 20 mbar) of pressure with no pressure drop when connected to the selection valve. It also allowed to produce much smaller volume of air bubbles. Table 2 shows the generated volume of air bubbles using pressure source.

3.3.2 Separation of Dyes For the investigation of the separation quality an optical setup was used. A defined volume of colored solution (that mimic drugs) was sent to the catheter, followed by portions of distilled water of the same volume. These solutions were separated from each other using gas bubbles in microliter volume range. The absorbance of the dye (and the successive flowing water volumes) was measured inside the tube using the optical setup. The setup consisted of a Tungsten Halogen light source and spectrophotometer (Avantes, Netherlands) with which the absorbed spectra measurements were done inside a measuring chamber. This was designed and fabricated at Lübeck University of Applied Sciences especially for the tested catheters. Allura Red dye was used (with distilled water) for colored solu-



tion preparation. Tests were applied on 3 tubes, arranged to examine 2 materials and 2 different inner diameters. All tubes have standard length of 1.5 m and their characteristics are shown in table 3. Table 1. Dosed and measured air bubble volumes produced by the flow source in Drug Multiplexing system.

S876

to separate two incompatible drugs using only one volume of cleaning agent or compatible drug separated by gas bubbles. The results are summarised in table 4. Table 4. Separation quality using air bubble volumes produced by the pressure source in Drug Multiplexing system. Concentration of first Dye

Concentration of Water [%]

Error [%]

Material of Catheter

Dosed Volume [ml]

Mean Measured Volume [ml]

Standard Error [%]

Tube

PUR *

0.41

0.48

+ 17

PVC*

100

0.44 – 1.64

0

0.059

0.16

0.17

+ 6.25

PVC**

100

4.98 – 15.84

0

0.031

0.41

0.40

- 2.43

PE

100

0 – 3.83

0

0.135

0.16

0.17

+ 6.25

* 1 mm inner diameter ** 0.6 mm inner diameter

PVC **

* Polyurethane (Inner diameter: 2.5 mm) * Polyvinyl chloride (Inner diameter: 3 mm)

Table 2. Air bubble volumes produced by the pressure source in Drug Multiplexing system. Material of Catheter

Applied Pressure [m bar]

Mean Volume of Produced Air Bubbles [ml]

Standard Error [%]

PVC *

44.48

0.029

0.411

PE **

24.54

0.012

0.239

* Polyvinyl chloride (Inner diameter: 1 mm, Outer diameter 2 mm) ** Polyethylene (Inner diameter: 1 mm, Outer diameter 1.5 mm)

First Volume

[%]

4

Second Volume

Conclusion

A new infusion system modality was developed. The aim to separate the drugs with gas bubbles was not only attained but the cumulative data from testing endorsed the viability of such a concept. Fig. 3 shows experimental separation of two fluids by 6.2 µl of air bubble. The used gas volumes of some µl are far below any dangerous value [20]. Nevertheless, air filters could be used at the end of catheter to remove any remaining gas bubble.

Table 3. Characteristcs of tubes used for examination of the dye separation quality. Material of Tube

Diameter [mm] Inner

Outer

1. Polyvinyl chloride (PVC)

1

2

2. Polyvinyl chloride (PVC)

0.6

1.4

1

1.5

3. Polyethylene (PE)

The dyes and cleaning agent (water) were applied with a flow rate of 0.5 ml/ min. The measurements on the selected tubes showed that depending on the tube diameter the cleaning efficiency of air bubble is high. It could be proved that the contamination of the next following solution is less than 2% of the supplied dye in a 1 mm inner diameter PVC tube. This means that it could be possible

Fig. 3. Two solutions separated by tiny air bubble inside a Tube (1 mm inner and 1.5 mm outer diameter).

5

Outlook

More measurements are needed to characterise the system, for example characterization of flow rate and precision of the applied drug volumes. There already measurements done using CO2 as separation medium. Measurements with CO2 as separating gas show an interesting effect. During the way through the tube the bubbles reduce their size by a factor of 3, due to its high solubility in water [20,22] and the quasi turbulent flow in the tube. These effects need deeper investigation for its application. A functional test model with 6 dosing units was also designed and built. Further future work will be to test clinically used medications with the new functional model.



Acknowledgements The authors would like to express thankfulness to the Federal Ministry of Education and Research (BMBF), Germay, for the project sponsorship.

6

References

[1] Journal of Patient Safety September 2013: 9(3); 122128. [2] Lamsdale et al., “A usability evaluation of an infusion pump by nurses using a patient simulator”, proceedings of the human factors and ergonomics society 49th annual meeting, 2005. [3] Ontario Health Technology Assessment Series; Vol. 12: No. 16, pp. 1–132, May 2012. [4] Interactions/Incompatibilities, B.Braun, Internet: http:/ /www.safeinfusiontherapy.com/cps/rde/xchg/hc-safein fusion-en-int/hs.xsl/7854.html (Accessed: 27th June 2014). [5] Timmerman A.M.D et al., “Development and validation of a new method for real-time measurement of fluid dynamics during simulated multi-infusion setups”, IFMBE Proceedings Vol. 25, 2009. [6] Abdul-karim S., “Design and development of a test bench to separate drugs in a multi-infusion system using gas bubbles”, Master thesis, Lübeck University of Applied Sciences, Lübeck, Germany 2013. [7] Nestler B., “Drug Multiplexing project description”, Lübeck University of Applied Sciences, 2011 [8] Burke G. et al., "Electronic control means for a plurality of intravenous infusion sets." U.S. Patent 4094318 A, Jul. 9, 1976. [9] Doehn M., "Device for infusing solutions from several infusion bottles." EP Patent 0013334 A2, Dec. 4, 1979. [10] Wunsch R. E., "Manifold for controlling administration of multiple intravenous solutions and medications." U.S. Patent 4512764 A, Sep. 27, 1982. [11] Rosskopf G. et al., "Device for dosed simultaneous infusing solutions." EP Patent 0189491 A1, Jan. 26, 1985. [12] Wunsch R. E., "Apparatus for controlling administration of multiple intravenous solutions and medications." U.S. Patent 4559036 A, Dec. 14, 1983. [13] Brown W., Henry T. Tai. "Method for sequential intravenous infusion of multiple fluids." U.S. Patent 4687475 A, Apr. 7, 1986. [14] Huntley A. et al., "Closed multi-fluid delivery system and method." U.S. Patent 4925444 A, Aug. 7, 1987. [15] Polaschegg H., "Multiple infusion system." DE Patent 3817411 A1, May. 21, 1988. [16] Neuder K., "Multiple infusion system." EP Patent 0543259 A1, Nov. 11, 1992. [17] Ebert H. et al., "Centralized multichannel fluid delivery system." U.S. Patent 5507412 A, Jun. 14, 1994.

S877

[18] Jerman J. et al., "Electrofluidic standard module and custom circuit board assembly." U.S. Patent 5640995 A, Jun. 24, 1997. [19] Jerman J. et al., "Apparatus and method for dispensing solutions." DE Patent102005045393 A1, Sep. 23, 2005. [20] Marek M. et. al, “Diagnosis and Treatment of Vascular Air embolism”, Anesthesiology, 106:164 – 77, 2007. [21] Abdul-Karim S., “Design and implementation of a test bed to separate different drugs in multi-infusion system using gas bubbles”, Student Conference 2014 on Medical Engineering Science in Lübeck, Proceedings ISBN 978-3-656-59648-6. [22] Culb W., Anesthesiology, Volume 107 - Issue 5 - pp 850-851, 2007.



S878

Modeling diffusion of gentamicin eluted from a coated intramedullary nail T. F. Klepsch, Laboratory of Medical Imaging, Lübeck University of Applied Sciences, and Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Lübeck, Germany, [email protected] G. Rau, Institute of Pharmaceutical Technology, Christian-Albrechts-Universität zu Kiel, Kiel, Germany, [email protected] J. Krieger, Laboratory of Medical Sensors and Devices, Lübeck University of Applied Sciences, and Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Lübeck, Germany, [email protected] H. Botterweck, Laboratory of Medical Imaging, Lübeck University of Applied Sciences, Lübeck, Germany, [email protected]

Introduction Coated bone implants play a steadily increasing role in bone fracture treatment. However, the effects controlling drug eluting and drug distribution in tissues surrounding bone implants still require further investigation. Furthermore, there is a poor correlation between in vitro laboratory eluting experiments and the drug eluting behaviour in vivo. In this research project we develop a simulation model for predicting drug distribution in tissues after drug eluting from an intramedullary tibial nail and implement it using the finite element software COMSOL Multiphysics.

Methods The initial simulation model comprises four concentric and homogeneous cylindrical compartments, namely the implant with a drug load of 50 mg gentamicin, a blood clot stemming from the implantation procedure, the medullary cavity, and the cortical bone compartment. Geometry and diffusion coefficients were taken from the literature and the periosteum as the outer border of the cortical bone was set to zero concentration. We conducted the simulation over a simulated time period of 28 days which is considered as the clinically relevant time frame.

Results The influence of the different compartments and their diffusion coefficients on the time-dependent concentrations were observed. After a simulated time of 24 hours the drug still resides mainly in the blood clot. Even after 28 days the concentration in the blood clot is still twice as high as in the medulla so that the diffusion process will still continue after this time.

Conclusion The implemented model achieves a reasonable outcome and shall serve as a starting point for future modeling attempts considering specific eluting profiles and bone healing processes. A physical in vitro compartmental bone model using hydroxyl-apatite scaffolds and hydrogels as bone and marrow compartments is currently being evaluated for model validation. Likewise, the model supports the setup of a test bench for measuring in vitro drug eluting from coated implants.



S879

A physical model of perfused, pulsating tissue compartments – Design concept B. Weber1,3, V. Hennicke2 and B. Nestler2 1 Medical Sensors and Devices Laboratory, Luebeck University of Applied Sciences, Luebeck, Germany, [email protected] 2 Medical Sensors and Devices Laboratory, Luebeck University of Applied Sciences, Luebeck, Germany 3 Graduate School for Computing in Medicine and Life Sciences, University of Luebeck, Luebeck, Germany

Abstract The calibration of pulse oxymeters is currently perfomed by empirical desaturation studies with human volunteers (controlled hypoxemia studies), which is very time consuming, elaborate and expensive. Furthermore this is not applicable for regular metrological controls of the sensors in clinical routine. A new approach for sensor tests based on the direct spectral modulation of the pulse oxymeter probe light signals is currently developed. For this approach a novel tissue model with pulsatile perfusion mimicking the optical and mechanical properties of a human finger is introduced. This paper presents measurements of the physiological dilation of the human finger due to the heart beat showing relative volume changes of V/V 1 %. A first concept of the novel tissue model is described. It uses silicone spiked with dyes and scattering particles as well as fluidic channels in combination with a pump and an adjustable flow restrictor.

1

Introduction

1.1

The principle of pulse oxymetry

1.2

Pulse oxymetry determines the arterial oxygen saturation (SaO2) by means of light absorption at two wavelengths, exploiting the pulsatile blood flow through peripheral arteries. Usually LEDs emitting at 660 nm and 940 nm are used. At these wavelengths oxygenated and deoxygenated hemoglobin show high differences in optical extinction 0. Due to the heart beat arterial blood is pumped through the vessel system and the blood volume in the illuminated body part (e.g. fingertip, earlobe, toe) increases. This leads to an increased absorption and eventually the detected photocurrent decreases periodically with each cardiac cycle (photoplethysmographic signal, PPG). From this PPG signal the DC (caused by skin, muscle, bone, venous and non-pulsating arterial blood) and AC (caused by pulsating arterial blood) component for each cycle is determined and the AC/DC ratio is calculated for both wavelengths λ1 and λ2. From these two ratios the so called ratio-of-ratios , is calculated which is related to SaO2 0 0:

f ( SaO2 )

AC DC AC DC

1

(1)

2

The exact relation between and SaO2 has to be determined empirically by means of controlled hypoxemia studies, which is described in the following section.

Current calibration practice: Controlled hypoxemia studies (CHS) with human volunteers

Since the relation of and SaO2 cannot be calculated theoretically it has to be obtained empirically 0. SaO2 in volunteers is lowered to different plateaus using a breathing mask and controllable FiO2 levels. Blood samples are drawn from an arterial line and analyzed in blood gas analyzers as SaO2 reference. The uncalibrated sensor signal is recorded continuously during the course of the desaturation and finally the reference SaO2 can be related to the sensor raw signal. Image 1 depicts a scheme of the CHS setup.

Image 1 Controlled hypoxemia study setup Typically 10 subjects are included, performing 5-6 SaO2 plateaus between 100 % and 75 % SaO2. At each plateau 5 blood samples are drawn and analyzed. Each subject should be desaturated twice, resulting in a total number of 500 to 620 samples for a complete CHS 0.



S880

It is obvious that the current calibration procedure is very time consuming, elaborate and expensive and cannot be used for regular metrological controls of pulse oxymeters in clinical routine. Today it is only possible to perform functional and safety tests on pulse oxymeters and there is no accepted method to assess the sensor performance and uncertainty other than the above described CHS 0.

1.3

Optical spectral modulation of pulse oxymetry light signals

A novel approach to tackle the problem described above is the direct spectral modulation of the light signals emitted by the pulse oxymetry probe without a human finger placed at the sensor 00. This method basically collects the probe light with a fiberguide and feeds the light into a spectroscopic setup. The light is then decomposed into its spectral components and the resulting spectrum is imaged onto the surface of a micromirror array with 1,024 x 768 single mirror elements (0.7” XGA DMD with ALP-4 highspeed controller, Texas Instruments & ViALUX GmbH). Each mirror can be tilted either +12° or -12° along its diagonal, allowing to steer the incident light into one of these directions (ON and OFF direction). Since one column of the array covers only a small spectral range of the total incident spectrum, it is possible to control the amount of the reflected light of this particular spectral range into both directions by changing the number of mirrors in +12° and -12° direction. In total this enables the user to vary the shape of the spectrum reflected in the ON direction. The micromirror array pattern can be changed with a maximum frequency of fmax = 22.727 kHz to generate dynamic sequences. The reflected spectrum is finally recollected and fed back to the pulse oxymeter sensor. The setup is shown in Images 2 and 3.

Image 3 Optical setup for direct spectral modulation of pulse oxymeter light signals 0, photograph By recording optical finger transmission spectra during CHS over time and using these spectra as input for the micromirror pattern the desaturation data can be recycled at any time after the actual CHS without the need of volunteers.

1.4

Physical model of perfused, pulsating tissue compartments

The concept of direct spectral modulation based on recorded finger transmission spectra has to be tested under standardized conditions where crucial parameters can be controlled, e.g. pulsation amplitude and frequency, absorber concentration, scattering effects. This is not possible when human subjects are the only source for transmission spectra because of intra- and interindividual variations that are not under control of the operator. A physical tissue model could close this gap and provide a means to investigate the performance of the spectral modulator under controllable conditions. For the design process it is important, which properties of real pulsatile perfused human tissue have to be imitated. Image 4 shows a plastination specimen of the blood vessel system of a human finger 0. It is obvious that the vast number of small blood vessels down to the capillaries is not realistic to be remodeled. The goal is not to produce an anatomically realistic tissue model, but rather a model that is capable of reproducing the deciding optical and biomechanical properties that influence the response of a pulse oxymetry sensor.

Image 2 Optical setup for direct spectral modulation of pulse oxymeter light signals 0, scheme



S881

3

Results and discussion

3.1

Pulsatile tissue dilation and relative volume change of the human finger

The physiological dilation of a human finger due to the heart beat was measured contactlessly with an optical laser triangulation sensor. An exemplary result is shown in Image 6.

Image 4 Plastination of the blood vessels of a human finger 0

2

Material and methods

2.1

Non-contact measurement of pulsatile tissue dilation

With a laser triangulation sensor (Micro Epsilon, ILD 2300-20, resolution 0.3 µm, fsample = 1.5 kHz) the tissue dilation under physiological conditions at the finger tip was measured contactlessly over time. With this dilation values a relative volume change due to the heart beat was calculated as a design requirement for the tissue model. Image 5 shows the triangulation sensor placed above the finger.

Image 6 Pulsation of a human finger measured with a laser triangulation sensor aimed at center of finger nail The plot shows the physiological pulse shape with the systolic peaks followed by the secondary dicrotic wave caused by reflections of the pulse wave at vessel bifurcations. The maximum dilation at the systolic peak amounts to approximately Δd = 40 µm. Assuming a circular cross section of the finger the relative volume change is calculated with (2):

V V

Vdilated Vundilated Vundilated l (r (r

r )2 r2

r)2 l r2 r2

l r2

(2)

The dilation Δd in this case is considered the change of finger diameter, so Δr = 0.5∙Δd. Assuming a finger radius of r = 6.5 mm this leads to a relative volume change of ΔV/V = 0.6 %. Taking into account a certain individual biological variation of the finger geometry the physiological volume change in a human finger due to the cardiac pulsation is in the range of approximately ΔV/V ≈ 1 %. Image 5 Setup for contactless measurement of tissue dilation with a laser triangulation sensor

3.2

Preliminary design concept for the tissue model

The design concept is divided into a solid and a fluidic component. The solid part covers the properties of the tissue itself, i.e. the mechanical and partially the optical



characteristics. It will be made from a material that can be well shaped, e.g. molded and whose mechanical properties can be varied easily, e.g. hardness, elastic module. A likely candidate material is silicone, since it is available in various mixtures to generate models of different mechanical properties. It can be handled rather easily, and it is also possible to add dyes and scattering particles leading to optical properties similar to human skin and soft tissue. To add fluidic channels to the tissue model silicone is also useful, since it can be molded and a variety of different shapes can be produced with rather simple casting molds. Connection to the pressure source and input/output ports shall be implemented by standard tubes and fittings. Image 7 shows a rough schematic of the proposed design concept.

The tissue model must be capable of generating the same optical response at the pulse oxymetry sensor as a real finger, i.e. it must have a pulsatile perfusion including the relative volume change due to the physiological circulation. It was shown, that this dilation is in the range of using a contactless approximately ΔV/V ≈ 1 % measurement method. Further detailed measurements on more subjects are necessary to give detailed information used as requirement for the design of the tissue model and the material selection. A first tissue model will implement only one single fluidic channel, which will have to be investigated regarding the optical response of the model. If more channels are needed, the proposed design concept can easily be adapted. Introducing porosity to the channels might also be an alternative to an increased number of channels.

4

Image 7 Schematic of the perfused tissue model. Pulsation is introduced by appropriate control of the pump and/or dynamic adjustment of the flow restrictor. As stated in the introduction the goal is not to exactly reproduce the vessel system’s anatomy with its huge number of vessels with very different diameters. In a first step a rough model with one single channel of a relatively large diameter is planned, with the possibility to increase the number fluidic channels and decrease the diameters to meet the conditions that generate a comparable optical response as a real finger. The fluid to perfuse the tissue model can either be an artificial fluid with optical characteristics similar to blood or even real human blood. The circulation would be provided by a pump and an additional flow restrictor that will be adjustable to achieve flow conditions similar to the physiological conditions.

4

Summary and outlook

The calibration of pulse oxymeters requires complex controlled hypoxemia studies, which are not applicable during everyday routine to check the sensor performance and uncertainty. A novel approach for sensor tests based on the recording and conservation of finger transmission spectra and direct spectral modulation of the light signal emitted by the pulse oxymetry sensor using those spectra could solve this problem. Yet this system has to be tested under controlled conditions with the described perfused tissue model.

S882

Acknowledgements

This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B). LUMEN is a joint research project of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck.

5

References

[1] W. G. Zijlstra, A. Buursma and O. W. van Assendelft, Visible and near infrared absorption spectra of human and animal haemoglobin - Determination and Application, Utrecht, Boston, Köln, Tokyo : VSP B.V., ISBN 90-6764317-3, 2000 [2] T. Aoyagi, Pulse oximetry: its invention, theory, and future, Journal of Anesthesia, vol. 17, pp. 259-266, 2003 [3] J. W. Severinghaus, Takuo Aoyagi: Discovery of Pulse Oximetry, Anesthesia and Analgesia, vol. 105, No. 6S S1S4, 2007 [4] International Organization for Standardization, ISO 806012-61:2011 „Medical electrical equipment – Part 2-61: Particular requirements for basic safety and essential performance of pulse oximeter equipment“, 2011 [5] Chr. Hornberger, Ph. Knoop, W. Nahm, H. Matz, E. Konecny, H. Gehring et al., A prototype device for standardized calibration of pulse oximeters, Journal of Clinical Monitoring and Computing, vol. 16, pp. 161-169, 2000 [6] B. Weber, B. Nestler and H. Gehring, Spectral and temporal modulation of pulse oxymetry probe light signals – improved recombination of spectrally decomposed light, Biomedical Engineering, vol. 58, issue S1, Sep. 2013 [7] Photograph by courtesy of the Institute of Anatomy, University of Lübeck



S883

Drug release from bone implants: a phenomenological modeling approach J. Krieger1,2, T. Klepsch1,2, T. Wenzel1, C. Damiani1 and St. Klein1 1 Medical Sensors and Devices Laboratory Lübeck University of Applied Science, Lübeck, Germany 2 Graduate School for Computing in Medicine and Life Sciences, Universität zu Lübeck, Lübeck, Germany Corresponding author: [email protected]

Abstract Orthopedic implants are widely used as drug carriers for local drug delivery to minimize infection risks for the patients after surgery. The drug release kinetics of those implants should ideally be assessed in-vivo through animal or human trials, which are expensive and time-consuming. In-vitro testing is a more affordable and simpler alternative. Conventional in-vitro methods are mostly too simplistic (e.g. paddle apparatus); they disregard the complex in-vivo situation and do not allow a valid in-vitro/in-vivo correlation. In this work, a new in-vitro method is proposed, in which drug diffusion and transport at the bone-implant interface can be taken into account by using a hydrogel membrane and a flow cell. The amount of released drug can be monitored continuously using a closed loop configuration. First tests using fluorescein sodium as model drug were successfully carried out and prove the ability of the method.

1

Introduction

One of the earliest and best researched applications of drug-eluting implants are coronary drug-eluting stents for the prophylactic treatment of neointimal hyperplastic, a restenose of the vessels [1]. Another important application area are orthopedic implants. Implant-associated infections are a serious condition mostly caused by bacteria, which are introduced into the patient’s body during surgery [2]. After Implantation, a layer of clotted blood forms on the implant`s surface. This layer presents ideal conditions for bacterial growth and prevents intravenously administered drugs from reaching the infection site [3]. In extreme cases the implant needs to be surgically replaced (so called revision). Also implant loosening due to poor attachment to the bone is a common condition which requires surgical revision as well [4]. Possible ways to address these problems include the use of drug-eluting implants [5, 6, 7, 8] or bone cements [9, 10, 11]. By using antibiotics like gentamicin to prevent infections or autologous growth factors to promote bone/implant bonding, the number of orthopedic implant`s revisions could be reduced. A few products, like the gentamicin-loaded, fracture fixation nails “PROtect” from Synthes, have already been marketed [12]. Before a drug-eluting implant can be placed on the market, however, a thorough assessment of the drug-release kinetics both in-vitro and in-vivo is required [16]. The most common in-vitro method is the so called paddle apparatus [17]. An implant model is immersed in a defined volume of a buffer solution (e.g. Phosphate buffered saline, PBS). Stirring the solution facilitates the homogeneous distribution of the drug in the buffer. The amount of released drug is then determined by taking samples at defined time intervals and analyzing them using standard assays (e.g. UV/VIS spectrometry).

Image 1 Bone healing Process. Evolution of tissue as a function of time [based on 14, 15]. Compared to in-vitro tests, in-vivo studies are more expensive, time-consuming and often ethically questionable. Traditional in-vitro models, however, are often too simplistic and show poor correlation of the drug-release kinetics with clinical trials and animal tests. Therefore, any substantial improvement of the current in-vitro models need to more closely take into account the physiological in-vivo conditions. The conditions at the implant`s interface, for example, are key factor determining the drug release kinetics [18, 19, 20]. Image 1 shows the stages of the bone fracture healing process from the formation of the clotted blood layer to the remodeling of the bone. In this work, the development of a phenomenological in-vitro model for continuous monitoring of the drug-release kinetics from orthopedic implants is discussed. This method is aimed at reducing the number of animal tests in the early stages of the development of new drug-eluting implants.



2

S884

Methodes

The coated implant sample or the test drug (fluorescein sodium) which is to be tested is placed inside a flow-cell separated from a flow channel by a hydrogel membrane (2-hydroxyethylmethacylate, water amount 66%). The hydrogel membranes were custom-made by mixing 2-hydroxyethyl-methacylate with an UV-initiator and polymerizing under an 18W UV lamp over 60 minutes. Image 2 shows the hydrogel membrane before Polymerization. Image 4 Photo of the flow cell. The circular chamber in the middle accommodates the implant sample and the hydrogel membrane(s). Right and left from the chamber, the inlet and outlet of the flow channel can be seen.

Image 2 The Hydrogel membrane before polymerization. The 2-hydroxyethylmethacylate is mixed with an UVinitiator. The polymerization takes place inside an 18W UV-oven. The membrane serves as a diffusion barrier. The characteristics of the hydrogel membrane can be controlled by changing its composition and thickness (1-4 mm) and depends on the biological tissue which should be emulated. In this way, different stages of bone healing process (see image 1) can be simulated. The flow cell is shown in image 3 and 4. It was built to hold the implant sample and one or more hydrogel membranes. The blood is substituted by phosphate buffered saline (PBS, pH 7.4) and is continuously pumped in a closed loop by a peristaltic pump (0.5 mm/s to approximate blood velocity in vascular capillaries). An overview of the experimental setup is shownin image 5.

Image 3 The phosphate buffered saline flows past the hydrogel membrane. The drug diffuses through the hydrogel into the flow channel.

First, the model-drug diffuses trough the hydrogel membrane into the flow channel. The concentration is monitored continuously over time using fluorescence spectroscopy (Spectrometer Avantes 2048-L, LED light source AvaLight 475, Avantes, Netherlands). The whole system is placed in a temperature-controlled chamber at (37±1)°C. Furthermore, the acrylic box is opaque to prevent photo bleaching of the fluorescein [21]. Preliminary measurements have been carried out using one diffusion barrier (hydrogel membrane) and fluorescein sodium salt (Sigma-Aldrich GmbH, Taufkirchen, Germany) as model drug. Future experiments will use fluorescein labeled gentamicin (Lumiprobe GmbH, Hannover, Germany). The analysis of the medium takes place in form of online spectroscopy. The heat element ensures constant 37°C inside the opaque box.

Image 5 The experimental setup. The pump pumps the blood substitute in circle past the hydrogel membrane.



3

Results

3.1

Calibration of the experimental setup

Image 6 displays the calibration curve of the experimental setup for fluorescein sodium. The lowest and highest quantification limits were 250 nM and 50 µM, respectively. The precision of the experimental procedure was found to be between 0.04% and 4% (concentration dependent). The accuracy lies between 6% and 18% (concentration dependent).

S885

Parameters like membrane thickness or flow rate should then be tuned to match in–vivo data reported in the literature. In a next step the results have to be compared with results from establish methods like the paddle apparatus to provide an alternative for in-vivo studies.

Image 7 Concentration vs. time.

5 Image 6 Calibration of the setup.

3.2

In−vitro-release study

A freshly prepared 40 µM fluorescein sodium aqueous solution was used as the test drug. The measured concentrations as well as the corresponding amounts of released drug vs. time are shown in image 7. The first detectable dye concentrations were measured after 12 hours. After the first week the test drug concentration in the main flow circuit was 1.1 µM, which equals an amount of released drug of 16.6 µg (40% of the available amount in the reservoir). The test drug concentration in the main flow circuit increased almost linearly within the first four days at about 0.1 µg/h, but seemed to taper off afterwards. This was due to the decrease of the concentration gradient across the hydrogel barrier due to the drug leaving the reservoir. Small oscillations (±10%) due to the night and day temperature fluctuations can be observed on the curve. A measurement that proves the temperature dependents of the spectrometer has been carried out and future test will be under constant temperature conditions (37°C to match with in-vivo conditions).

4

Conclusions

First tests have shown that the proposed method can be used for continuous monitoring of the drug release process. Future measurements will be carried out with fluorescein labeled gentamicin and multilayer hydrogel membranes.

Acknowledgements

This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B). LUMEN is a joint research project of Lübeck University of Applied Sciences and Universität zu Lübeck and represents an own branch of the Graduate School for Computing in Medicine and Life Sciences of Universität zu Lübeck.

6

References

[1] M. Merciadez, L. Alquier, R. Mehta, A. Patel, and A. Wang, “A Novel Method for the Elution of Sirolimus (Rapamycin) in Drug-Eluting Stents”, Dissolution Technologies (2011), 37-42 [2] T. H. Hoang Thia, F. Chaia, S. Leprêtrea, N. Blanchemaina, B. Martela, F. Siepmanna, H.F. Hildebranda, J. Siepmanna, and M.P. Flamenta, “Bone implants modified with cyclodextrin: Study of drug release in bulk fluid and into agarose gel”, International Journal of Pharmaceutics (2010), Vol. 400, 74–85 [3] Qiu, Y., N. Zhang, Y. H. An, and X. Wen, “Biomaterial strategies to reduce implant-associated infections”, Int J Artif Organs (2007) Vol. 30, 828–841 [4] M. A. Buttaro, R. Pusso, and F. Piccaluga, “Vancomycin-supplemented impacted bone allografts in infected hip arthroplasty”, J Bone Joint Surg (2005) Vol. 87-B, 314-319 [5] U. Brohede, J. Forsgren, S. Roos, A. Mihranyan, H. Engqvist, and M. Strømme, “Multifunctional implant coatings providing possibilities for fast antibiotics loading with subsequent slow release” J Mater Sci: Mater Med (2009) Vol. 20, 1859–1867



[6] Lee, C.G. et al. Simulation of Gentamicin Delivery for the Local Treatment of Osteomyelitis, Biotechnology and Bioengineering (2005), Vol. 91, 622-635 [7] M. Lucke, T B Wildemann, S Sadoni, C Surke, R Schiller, A Stemberger, M Raschke, NP Haas, G Schmidmaier. “Systemic versus local application of gentamicin in prophylaxis of implant-related osteomyelitis in a rat model”,Bone (2005), Vol. 36, 770– 778 [8] H. Vester, B. Wildemann, G. Schmidmaier, U. Stöckle, and M. Lucke, “Gentamycin delivered from a PDLLA coating of metallic implants in vivo and in vitro characterisation for local prophylaxis of implant-related osteomyelitis”, Injury (2010), Vol. 41, 1053–1059 [9] S. Torrado, P. Frutos, and G. Frutos, “Gentamicin bone cements: characterisation and release (in vitro and in vivo assays)”, International Journal of Pharmaceutics (2001), Vol. 217, 57–69 [10] B. A. Masri, C. P. Duncan, and C. P. Beauchamp, “Long-term Elution of Antibiotics from BoneCement”, The Journal of Arthroplasty (1998) Vol. 13, 333–1338 [11] J. G. E. Hendriks, D. Neut, J. R. van Horn, H. C. van der Mei, H. J. Busscher, “The Release of Gentamicin from Acrylic Bone Cements in a Simulated Prosthesis-Related Interfacial Gap”, Journal of Biomedical Materials Research Part B: Applied Biomaterials (2003) Vol. 64B, 1–5 [12] “http://www.synthes.com/sites/intl/Products/featured -products-solutions/pages/Expert-Tibial-NailPROtect.aspx” , online access: 29.01.2014 [13] J. E. Davies, “Understanding Peri-Implant endosseous healing”, Journal of Dental Education (2003), Vol. 67, 932–949 [14] C. T. Brighton, R. M. Hunt, "Early histologic and ultrastructural changes in microvessels of periosteal callus", Journal of Orthopaedic Trauma (1997), Vol. 11, 244-253 [15] A. W. Ham, W. R. Harris, "Repair and transplantation of bone", The biochemistry and physiology of bone, New York: Academic Press (1972), 337-399 [16] E. Sanchez, M. Baro, I. Soriano, A. Perera, C. Evora, In vitro – in vivo study of biodegradable and osteointegrable gentamicin bone implants. European Journal of Pharmaceutics and Biopharmaceutics (2001), Vol. 52, 151–158 [17] Ramchandani, M. and Robinson, D. “In vitro and in vivo release of ciprofloxacin from PLGA 50:50 implants”, Journal of Controlled Release (1998), Vol. 54, 167–175 [18] M. Zilberman, J. J. Elsner, “Antibiotic-eluting medical devices for various applications”, Journal of Controlled Release (2008) Vol. 130, 202–215 [19] A. T. Raiche and D. A. Puleo, “Triphasic Release Model for Multilayered Gelatin Coatings That Can Recreate Growth Factor Profiles During Wound Healing”, Journal of Drug Targeting (2001), Vol. 9, 449–460 [20] N. L. Ignjatovic, P. Ninkov, R. Sabetrasekh, D. P. Uskokovic, “A novel nano drug delivery system

S886

based on tigecycline-loaded calciumphosphate coated with poly-DL-lactide-co-glycolide”, J Mater Sci: Mater Med (2010), Vol. 21, 231–239 [21] L. Song, E. J. Hennink, I. T. Young, and H. J. Tanke, “Photobleaching kinetics of fluorescein in quantitative fluorescence microscopy”, Biophys J. (1995); Vol. 68, 2588–2600



S887

A Portable In-Ear Pulse Wave Measurement System R. Kusche1, A. Malhotra1, M. Ryschka1, S. Kaufmann1,2, 1 Laboratory for Medical Electronics, Lübeck University of Applied Sciences, Germany, [email protected] 2 Graduate School for Computing in Medicine and Life Sciences, University of Lübeck, Germany

Abstract The measurement of the pulse wave has proven to be a vital tool in medical diagnosis. Whereby most pulse wave measurements are carried out at extremities, this work proposes a system for measuring the pulse wave and the Pulse Arrival Time (PAT) in the interior of the ear. The developed measurement device is based on a battery powered microcontroller system. The measurement device simultaneously acquires an Einthoven Lead II Electrocardiogram (ECG), a dual wavelength Photoplethysmogram (PPG), the pressure in both ears, the temperature inside the ear canal, as well as the subject’s motion. The acquired measurement data can either be saved on a micro SD-card or can be transmitted wireless via Bluetooth or wired via USB to a host PC for further analysis. Battery powered the device can operate up to 8 hours. In addition to the system description, first measurements carried out with the system will be presented.

1

Introduction

A key to future personal monitoring is portability and adaptability. For that purpose it is desirable to record vital signs non-invasively and comfortably for a long period of time. Pulse Arrival Time (PAT) and the morphology of the pulse wave are considered as indicators of arterial stiffness and are also known as (prognostic) markers for cardiac and vascular diseases [1, 2, 3, 4, 5]. In the past various methods and sites have been demonstrated for pulse wave measurements, mostly on carotid and femoral artery through invasive and non-invasive methods. This work proposes a system for detecting the pulse wave inside the auditory canal for PAT and Pulse Wave Velocity (PWV) measurements, as proposed by [6]. The developed measurement device is based on a battery powered microcontroller system and can simultaneously acquire a single channel Einthoven Lead II Electrocardiogram (ECG), a dual wavelength Photoplethysmogram (PPG), the pressure in both the ears, the temperature inside the ear canal, as well as the subject’s motion via an accelerometer. After acquisition, the measured data can be transmitted via USB or wirelessly via Bluetooth to a host PC for further analysis. Additionally it is also possible to save the measurement data on a micro SD-card.

2

Measurement Methods

2.1

Electrocardiogram (ECG)

Via the recording of the Electrocardiogram (ECG) it is possible to detect the electrical activity of the heart, which is closely correlated with the contractions of the heart’s atria and ventricles and can therefore provide a reliable time reference for the PPG and pressure measurements.

2.2

Photoplethysmography (PPG), Pulse Arrival Time (PAT) and Pulse Wave Velocity (PWV)

Caused by the rhythmic contraction of the heart, blood is pumped into the arterial system where it propagates as a pulse wave due to the elasticity of the arteries. The blood

pulsation is time and site dependent and can be measured in terms of flow, pressure and volume. PPG simply detect the changes in volume of blood below the sensor by exploiting the absorption of light. The PPG in this work is recorded from the A. temporalis superficialis on the height of the Tragus region. The morphology of the obtained waveform, as well as thereof derived PAT are very significant and direct prognostic markers for the stiffness of the arteries [1, 2, 3, 4, 5]. Hence provides vital information regarding the overall condition of the cardiovascular system. PAT values can also be used for the calculation of the PWV, which is also used as an important marker for arterial stiffness [2, 3]. The PAT is defined as the time span between the R-peak of the ECG and the arrival of the pulse wave at a specific position. The PAT is commonly defined as one of three possibilities according to Figure 1. R-Peak

Peak Amplitude PPG

PATFoot

ECG PPG

PATDerivative PATPeak

Figure 1. Calculation principle for the Pulse Arrival Time (PAT). The PWV (1) is given by the ratio between the distance of two measurement sites (Δx) and the Pulse Transient Time (PTT) [7], which is the time required by the pulse wave to travel that distance. ∆𝑥 (1) 𝑃𝑊𝑉 = 𝑃𝑇𝑇 The PTT can also be expressed in terms of the PAT and the Pre-Ejection Period (PEP) according to (2). Where the



PEP is the time between the R-peak of the ECG and the opening of the aortic valve. 𝑃𝑇𝑇 = 𝑃𝐴𝑇 − 𝑃𝐸𝑃

2.3

(2)

In-Ear Pressure

For the measurement of the internal pressure variation the auditory canal has to be sealed against the ambient pressure. The sealing in this work is achieved via stethoscope ear olives connected to a pressure sensor. Inside the sealed cavity a pressure variation can be observed [6]. According to the ideal gas law this pressure variation corresponds to a volume change, under the assumption of a constant temperature. Figure 2 shows the principle drawing of the inear pressure model of the auditory canal. Tissue rich with arteries

Sealing

Tymphanic Membrane

Eustachian tube

Change in volume (∆V) due to pulsating blood

Figure 2. Basic model of the pressure changes in the auditory canal. The volume change can in principle be caused by two effects or their combination 1) the blood vessels around the auditory canal expand during the inrush of the pulse wave 2) the tympanic membrane moves corresponding to the heart beat.

2.4

Acceleration and Temperature

For the assessment of the subject’s position and physical activity an accelerometer is implemented in the measurement system. Additionally the temperature in the ear canal can be measured via a NTC resistor based temperature sensor.

3

Implementation

3.1

Block diagram

Figure 3 shows a block diagram of the developed portable in-ear pulse wave measurement system. ECG

Battery

SD-CARD

PRESSURE SENSOR

PPG

MICROCONTROLLER

PC

TEMPERATURE SENSOR

ACCELERATION SENSOR

USB

BLUETOOTH

iGunther V1.0 PRESSURE SENSOR

PPG-SENSOR

The microcontroller based embedded system acquires all measurement data from the subject and transmit the measured data to the host PC for further analysis and display. The data can be transmitted via Bluetooth, USB or can be stored on the micro SD-card.

3.2

ECG ELECTRODE

Figure 3. Block diagram of the portable microcontroller based measurement system.

ECG Module

The ECG module consists of an Instrumentation Amplifier (INA) with a Driven Right Leg (DRL), a driven shield, and a base line wandering rejection circuit. The ECG module measures in a frequency range of 150 mHz to 150 Hz. The analogue ECG signal is digitized via the internal 12 bit Analogue to Digital Converter (ADC) of the microcontroller.

3.3 Middle Ear

Atmospheric Pressure

S888

Photoplethysmography (PPG) Module

The implemented PPG circuit is designed for reflective mode and is based on two time multiplexed LEDs with wavelengths of 660 nm and 940 nm respectively. The reflected light is sensed by a photodiode, which is connected to a transimpedance amplifier (OPA2380 from Texas Instruments). Whose output signal is used on one hand to control the current of the LEDs by a digital control loop to ensure the reflected light intensity is in the optimal reception range of the detector circuit. On the other hand the signal is band-pass filtered with a pass-band of ≈ 15 mHz to ≈ 30 Hz to remove the DC-offset due to the dark current of the photo diode and to mitigate 100 Hz flickering of the ambient light. The cut-off frequencies are also chosen to ensure the fidelity of the desired PPG waveform, whose spectral components are ranged mainly from a few mHz up to ≈ 20 Hz. Finally the PPG signal is digitized via another channel of the microcontroller’s ADC.

3.4

Pressure Measurement

The pressure sensors (HCE-M010DBE8P3 from First Sensor AG) have a calibrated and compensated pressure measurement range of ±1000 Pa with a digital Serial Peripheral Interface (SPI) output. The pressure sensor supports up to 1 kSPS with 14 bit precision, which leads to a theoretic resolution of 122 mPa.

3.5

Acceleration Measurement

The acceleration measurement is based on a LIS3DH (from ST Microelectronics) accelerometer. The sensor can measure in three directions with full scale ranges from ±2 g up to ±16 g with 16 bit precision per direction at a sample rate from 1 SPS to 5 kSPS and is connected to the microcontroller via the SPI interface.

3.6

Temperature Measurement

The temperature measurement is based on a NTC resistor sensor with dimensions of 1.0 x 0.5 x 0.5 mm3, which allows the sensor be located inside the ear canal. The NTC resistance is evaluated with a quarter Wheatstone bridge and its calibration is optimized for the temperature range



around 37 °C. The output signal of the bridge is amplified and digitized with a further channel of the internal ADC of the microcontroller.

3.7

Housing and Battery

The housing for the measurement system was designed in SolidWorks (Dassault Systems) and manufactured with a 3d-printer (MakerBot Replicator 2X). The housing consists of ABS plastic and has a size of 71.5 x 71.5 x 38 mm3. Figure 4 shows an exploded assembly drawing of the housing inclusive the Printed Circuit Board (PCB) and the used battery.

S889

ing to the microcontroller is implemented as asynchronous serial connection. Alternatively to the battery the system can be supplied externally via USB. Whereas the digital components operate with +3.3 V only, the analogue components are supplied with ±3.3 V. For electrical safety considerations it is recommended to operate the system without USB connection in wireless or SD-card storage mode while running from the system battery.

3.9

Manufactured System

Figure 5 shows a photograph of the manufactured and populated Printed Circuit Board (PCB) of the measurement system. The PCB has a size of about 60 x 60 mm2 and contains about 200 components.

Figure 5. Manufactured and populated Printed Circuit Board (PCB) of the developed measurement system. The PCB has dimensions of 60 x 60 mm2 and contains about 200 components.

3.10 Figure 4. Exploded assembly drawing of the housing of the measurement system inclusive the Printed Circuit Board (PCB) and the used battery. The li-ion battery has 3.7 V by 1.25 Ah with dimensions of about 53 x 34 x 5.5 mm3 and weighs about 23 g. The expected run-time with the battery is about 8 hours.

3.8

Microcontroller System and Communication Interfaces

The microcontroller (ATxMega128A4U from Atmel Corp.) acquires the measurement data from the different sensors and handles data transmission to the host PC via Bluetooth or USB. Optionally the measurement data can be stored on the micro SD-card. While USB communication is realized via the internal USB stack of the microcontroller, Bluetooth is implemented with a commercially available module (RN42-I/RM from Roving Networks). The Bluetooth module is certified according to Bluetooth V2.1 incl. Enhanced Data Rate (EDR) mode and supports master and slave mode with up to 300 kbps. The interfac-

Software

The firmware of the measurement system is based on the Atmel Software Framework (ASF) and is written in C language. The PC interface software is written in C# language and is based on the .NET framework from Microsoft. It is able to display the measured waveforms in real-time with a latency of about 50 ms and can be used to configure the embedded measurement systems in terms of active channels and sample rates. With the interface software, waveforms can be recorded and exported to MathWorks MATLAB or Microsoft Excel. Figure 6 shows an image of the Graphical User Interface (GUI) of the software with the acquired waveforms. Based on the recorded PPG, ECG and pressure waveforms, PAT and PWV values are derived.



Age Sex Height Weight Heart Rate Mean PATDerivative, PPG Mean PATDerivative, Press.

Subject 1 27 male 190 cm 92 kg 86 BPM 197 ms 154 ms

S890

Subject 2 29 male 188 cm 134 kg 64 BPM 201 ms 167 ms

Table 1. Results of the PAT measurements

6

Figure 6. Screenshot of the user interface of the developed C# based PC software.

4


Table 1 shows preliminary measurement results acquired over 30 heart beats on two healthy male subjects. For simplicity reasons only measurements on the left ear are shown. The measurement signals are sampled with 1 kSPS and are digitally processed with zero-phase filters in MathWorks MATLAB. Afterwards peak detections and derivation calculations are performed on the filtered signals. The PPG and pressure waveforms are referenced to the ECG’s R-peak for evaluating the PAT values. The PAT derived from the pressure measurements are in a range of about 150 ms to 170 ms and the PAT derived from the PPG measurements are in a range of about 190 ms to 210 ms. These results are reasonable taking into account that usually reported PAT values on extremities are in the range of 240 ms [8]. With an assumed PEP of 70 ms and an estimated time difference of 50 ms between PATFoot and PATDerivative the results of the pulse wave velocities are PWVPPG ≈ 3.8 m/s and PWVPressure ≈ 7.4 m/s for Δx = 30 cm. This indicates that the pressure wave in the ear measured through a pressure sensor occurs before the change of blood volume in the arteries, measured via PPG.

5

Conclusion

A prototype of an in-ear pulse wave measurement system was developed and tested. The results are very promising and showing a good performance. In future the prototype has to be miniaturized and the interface software must be enhanced by the possibilities to automatically measure PAT and PWV, as well as the heart rate. Furthermore a second PPG channel for measurements on the extremities should be added to have an additional reference measurement side for comparisons.

Acknowledgment

The authors would like to thank F. Adam, G. Ardelt, K. Breßlein, and N. Hunsche for supporting this work and Linear Technology and Texas Instruments for the provision of free samples during the development process. This publication is a result of the ongoing research within the LUMEN research group, which is funded by the German Federal Ministry of Education and Research (BMBF, FKZ 13EZ1140A/B).

7

References

[1] M. Middeke, Pulswellenanalyse, Renaissance einer alten Methode als moderner Biomarker der Gefäßsteifigkeit. Deutsche Medizinische Wochenzeitschrift. 2010, 135, S3. [2] S. Laurent et al., Expert consensus document on arterial stiffness: methodological issues and clinical applications. European Heart Journal. 2006, 27, p. 25882605. [3] S. Laurent et al., Aortic stiffness is an independent predictor of all-cause and cardiovascular mortality in hypertensive patients. Hypertension. 2001, 37, pp. 1236. [4] A. P. Guerin et al., Impact of aortic stiffness attenuation on survival of patients in end-stage renal failure. Circulation. 2001, 103, p. 987-992. [5] J. Baulmann et al., Arterielle Gefäßsteifigkeit und Pulswellenanalyse - Positionspapier zu Grundlagen, Methodik, Beeinflussbarkeit und Ergebnisinterpretation. Deutsche Medizinische Wochenschrift. 2010, 135, S4-S14. [6] S. Kaufmann, G. Ardelt, A. Malhotra, M. Ryschka, Inear pulse wave measurements: A pilot study. Biomedical Engineering / Biomedizinische Technik. 2013. [7] F. S. Cattivelli and H. Garudadri, Noninvasive Cuffless Estimation of Blood Pressure from Pulse Arrival Time and Heart Rate with Adaptive Calibration. IEEE computer society. 2009, p. 114-119. [8] M. C. Kortekaas et al., Comparison of bilateral pulse arrival time before and after induced vasodilation by axillary block. Physiological Measurements. 2012, 33, p. 1993-2002.



S891

Development of short-pulsed high-field electromagnetic dipoles for laser-based proton therapy Michael Schürer1, Thomas Herrmannsdörfer2, Leonhard Karsch1, Florian Kroll2, Umar Masood1, Jörg Pawelke1,2 1 2

OncoRay – National Center for Radiation Research in Oncology, Technische Universität Dresden, Germany Helmholtz-Zentrum Dresden - Rossendorf, Dresden, Germany

Introduction During the last years, the new technology of laser based particle acceleration was developed at such a rate that medical application for cancer therapy becomes conceivable. Promising more compact and economic accelerators, the laser technology however generates intense ultra-short (~ ps) pulsed proton beams with large divergence and broad energy spectrum. Within the German joint research project “onCOOPtics” the clinical applicability of such pulsed proton beams is investigated including the development of a laser accelerator and a suitable beam transport.

Methods A compact beam transport system was designed enabling an efficient transport of proton pulses from generation to treatment site. The initially divergent proton beam is captured by a cylindrical electromagnet (solenoid), deflected by 45° dipole magnets and formed by quadrupole magnets, whereas the spectrum is shaped by adaptable lead apertures. For realization, electromagnetic dipoles with magnetic fields of up to 10 T are required to deflect up to 220 MeV protons. These field strengths are achieved by in-house developed non-ferrous dipoles that consist of 80 copper coils in 12 layers and are operated at peak currents of up to 20 kA. To handle the high currents and the generated heat the dipoles are externally cooled and operated in 1 ms short pulses synchronized with the laser repetition frequency.

Results The prototype of a short-pulsed electromagnetic dipole magnet was designed and manufactured. Results of the experimental characterization and first performance tests at a conventional Tandem accelerator are under way.

Conclusion Pulsed electromagnetic dipoles as crucial components of a compact beam line for laser-accelerated protons are engineered. Following validation of their suitability at a conventional accelerator the dipoles will be implemented and further tested at a laser accelerator. Together with improvement of the dipole the design of quadrupoles will start. The work was supported by the German Federal Ministry of Education and Research (BMBF), grant no. 03Z1N511. Contact: [email protected]



S892

Comparative study of different translational methods for the development of medicinal products inside the research collaboration BIOFABRICATION for NIFE in Hannover N. Lüdtke1, F. Duda2, H. Voigt2, A. Loos5, M. Wilhelmi3, A. Kampmann4, C. Schubert6, M. Elff1, T. Lenarz2, A. Haverich3 1

NIFE, Feodor-Lynen-Str. 27, 30625 Hannover, Germany Lü[email protected]

2

Department of Otolaryngology, 3Department of Cardiothoracic, Transplantation and Vascular Surgery, 4Clinic for Cranio-Maxillo-Facial Surgery, Hannover Medical School, Carl-Neuberg-Str. 1, 30625 Hannover, Germany 5

Biocompatibility Laboratory BioMedimplant, Hannover Medical School, Feodor-Lynen-Str. 31, 30625 Hannover, Germany 6

University of Siegen, Artur Woll-Haus, Am Eichenhang 50, 57076 Siegen, Germany

Introduction The innovation of medicinal products is a complex cost- and time-intensive process. Only few products pass successfully from basic research through the approval procedure to the market. To organize the innovation trajectories of medicinal products from first ideas to final products more effectively, there is a need for novel strategies to translate research data into clinical practice. The socio-technical process of translation demands heterogeneous cooperation and is often hindered by conflicting interests, e.g. between scientific acceptance and economic benefit. This study analyses current translational strategies in the field of biomedical engineering, using the research collaboration BIOFABRICATION for NIFE as a model. For the development of novel medicinal products, it is indispensable to identify barriers of innovation and factors of success for research and development.

Methods In close cooperation with sociologists, we use qualitative social research methods, such as ethnography and interviews, to analyse, compare, and evaluate structures and concepts of diverse projects from BIOFABRICATION for NIFE associated institutes (Clinic for Dental Prosthetics, Department of Cardiothoracic, Transplantation and Vascular Surgery and Department of Otolaryngology) which are analysed in detail over a period of three years. To cover a broad range of translation stages, the observed projects are situated at different states of progress (in vitro study, animal study and clinical trial).

Results This close view on translational methods will supply knowledge and skills to optimize the innovation trajectories of medicinal products from bench to bedside. The observed projects may share results to avoid previous mistakes and raise efficacy. The results of this study will also provide a broad support of local and nationwide research projects in the future.

Conclusion Translational knowledge is fundamental for successful and effective research and development of new medicinal products. Gathering and evaluating experiences in this area means to support the innovation progress. Unauthenticated Download Date | 4/21/17 2:48 PM


S893

Development of vessel phantoms using intravascular ultrasound (IVUS) datasets T. Hoffmann1, S. Glaßer2, F. Klink3, A. Boese4, M. Skalej1 1 Department of Neuroradiology, Otto-von-Guericke University, Magdeburg, Germany 2 Department of Simulation and Graphics, Otto-von-Guericke University, Magdeburg, Germany 3 Department of Machine Design, Otto-von-Guericke University, Magdeburg, Germany 4 Department of Medical Engineering, Otto-von-Guericke University, Germany Supported by Forschungscampus STIMULATE (BMBF 03FO16101A), Magdeburg, Germany, [email protected]

Introduction Phantoms of neurovascular structures are used e.g. for hemodynamic research, training of catheter interventions and as visual models in education [1], [2].The use of medical imaging and rapid prototyping (RP) methods enables an individual manufacturing. The mechanical properties and structural shape of vessel walls cannot be reproduced by these techniques so far [3]. Therefore, we investigated the ability for modelling vessel phantoms based on an IVUS dataset.

Methods A section of a porcine artery from a kidney was dissected and integrated in an artificial blood flow circle. Data acquisition was carried out with an intravascular ultrasound system (Volcano s5i™ Imaging System, Eagle Eye® Platinum Catheter). After segmentation and visualization of the artery structure, the surface mesh was transformed into a computer aided design compatible data format by surface reconstruction. In the design process a tripartite negative form was generated. The upscaled parts were manufactured from wax using multi-jet modeling (ThermoJet®). The surface was finished by varnish. The vacuum casting method was used for filling the assembled mould with transparent silicone (ELASTOSIL® RT 601). After cracking and melting the mould and core, the phantom was free from wax. The resulting phantom was scanned by 3D computed tomography (CT) and compared to the original digitized mesh.

Results It was shown that upscaled artery phantoms with varying vessel wall thickness can be segmented, visualized, designed and manufactured based on IVUS datasets. Geometrical deviations of the original preparation occur during IVUS dataset generation caused by an inconstant pullback velocity. Additionally, deviations emerge during surface reconstruction in high curvature regions.

Conclusion To raise the geometrical accuracy (e.g. structure, length, shape) of the vessel phantom, a combinded data acquisition of IVUS and CT seems to be advantageous. Scaled vessel phantoms of flexible silicone with a wall thickness of approximately 2 mm can be manufactured by the presented process.

References [1] [2] [3]

C. Roloff, R. Bordás, R. Nickl, Z. Mátrai, N. Szaszák, S. Szilárd, and D. Thévenin, “Investigation of the velocity field in a full-scale model of a cerebral aneurysm,” International Journal of Heat and Fluid Flow, vol. 43, pp. 212–219, 2013. A. Boese, A. Fahlberg, and G. Rose, “Rapid Prototyping Phantom der arteriellen Gefäße des Kopfes,” in 10. Magdeburger Maschinenbau-Tage, Forschung und Innovation, Magdeburg, 2011. G. Lamouche, B. F. Kennedy, K. M. Kennedy, C.-E. Bisaillon, A. Curatolo, G. Campbell, V. Pazos, and D. D. Sampson, “Review of tissue simulating phantoms with controllable optical, mechanical and structural properties for use in optical coherence tomography,” Biomedical optics express, vol. 3, no. 6, pp. 1381–1398, 2012.



S894

Experimental Setup for the Analysis of the Water Jet of a WaterAssisted Liposuction Device R. Mau, C. Drobek, H. Seitz Chair of Fluid Technology and Microfluidics, University of Rostock, Germany [email protected]

Introduction The method of water-assisted liposuction (WAL) is capable of gentle extraction of human fatty tissue with the use of a specifically U-shaped water jet. Associated with it is the obtention of adipose derived stem cells (ADSC). A novel WAL-device with compact dimensions for the extraction of small quantities of ADSC for tissue engineering application is investigated. This is done in cooperation with human med AG, Schwerin, Germany. Accordingly an experimental setup for the analysis of the water jet of the novel WAL-device and established WAL-systems was developed.

Methods The experimental setup uses a piezoresistive cantilever as a force sensor to investigate the water jet of the WAL-device. Specifically designed sensor heads allow for analysing either the complete or the partial cross-section of the water jet. To ensure precise positioning of the nozzle of the WAL-device in front of the sensor head, a fine adjustment screw is used. Additionally, a pressure sensor measures the pressure of the water supply to analyse the relationship between force and pressure.

Results An experimental setup for the analysis of the water jet was successfully established. Measurements were taken from the complete cross-section of the water jet of established WAL-devices. As expected there is a linear relationship of the impact force of the water jet and the pressure in the water supply. The effect of various nozzle geometries on the impact force of the water jet could be shown.

Conclusion Further work will include experimental investigations of the water jet of the novel WAL-device in order to optimize the nozzle for a gentle extraction of the ADSC. Measurements of the partial cross-section of the water jet will be performed to identify the distribution of the impact force over the shape of the jet. An evaluation of the results referring the stress on stem cells of human fatty tissue caused by the water jet is in progress.



S895

Specific flow rate of infusion filter systems A. Buchholz, J.-B. Matthies, W. Schmidt, N. Grabow, K.-P. Schmitz Institut for Biomedical Engineering, Rostock University Medical Center, Rostock, Germany [email protected]

Abstract The specific flow rate is an important value to characterize membrane filters for intravenous infusion. A test setup was realized and the transmembrane pressure and the flow rate are measured. The specific flow rate of 5 filter systems manufactured by the RoweMed AG is evaluated. It varies between 7.5 and 19.5 ml/(min*bar*cm²) for the different filter systems. The specific flow rate is not constant at different measurement points with different flow rates. It differs for certain filter membranes and is influenced by the geometry of the filterhousing.

1

Introduction

The usage of filters is advisable in intravenous infusion to avoid the contamination of the medical solution with particles, bacteria and the formation of air-bubbles. The filtersystems contain certain porous polymer membranes made of different materials and with different structures. The specific flow rate provides information about the flow dependent on the pressure and the surface area and it is one important value for the characterization of the filter system.

2

Methods

To evaluate the specific flow rate of different infusion filtersystems a test setup was realized according to DIN 58355-1. [1] As shown in figure 1 the setup consists of a reservoir with clean water and a peristaltic pump to deliver the fluid through the testobject. The water flows through a second reservoir only half filled with water to attenuate the pulsation. To avoid a blocking of the tested filter system by particles during the test the water is filtered by a pre-filter with a pore size of 0.2 µm. The flow rate is measured by an ultrasonic flow sensor system (Levitronix, LFS-008). A pressure sensor (PCE-910) is placed before and after the filter system to measure the transmembrane pressure. The clean water is lead back to the reservoir. The parts are assembled by silicone tubes with an inner diameter between 1.5 and 3.0 mm.

Figure 1: test setup for the evaluation of the specific flow rate of infusion filters The pressure and the flow rate are recorded during the flow of a volume of at least 100 ml. During the measurement the pressure has to be constant about at least 0.01 bar.

The specific flow rate of the filter is defined as 𝐿=

𝑉 𝑡∗𝐴∗𝑃

L is the specific flow rate, V is the volume lead through the filter, t is the time, A is the surface area of the membrane and P is the transmembrane pressure. The pressure, the volume and the time are measured while the surface area is given by the manufacturer of the filter systems. Five different types of filter systems for intravenous infusion by the company RoweMed AG are analysed. They have a pore size of 0,2 µm. Three different types of filter membranes are used in these systems. They differ in the material, the structure and the charge of the surface. They are assembled in a transparent filter housing with luerconnectors and tubing. Three different sizes are available for the application on adults (A=13.5 cm²), paediatrics (A=4.0 cm²) and neonats (A=2.0 cm²). The shape of the filter housings of systems with the same size is equal. Table 1 shows the relevant properties of the testet filter systems. Table 1: Filtersystems used as test samples for the evaluation of the specific flow rate Filtersystem RoweFil RoweFil RoweFil RowePaed RowePaedM

Membrane PET PES PA66+ PET PET

Charge positive -

Surface area 13.5 cm² 13.5 cm² 13.5 cm² 4.0 cm² 2.0 cm²

The pressure of the test setup is limited by the tubing and the connectors and has a maximum value of about 1 bar. Every filter type is analysed at four individual flow rate measurement points where the highest corresponds to the maximum pressure value. The measure points are shown in table 2. FR stands for the flow rate in ml/min. 6 test samples are analysed for each filter system.



20,00

Table 2: Values of the flow rate measurement points

3

Membrane PET PES PA66+ PET PET

FR1 20 10 10 5 4

FR2 60 45 30 30 10

FR3 100 75 50 55 16

FR4 120 110 70 80 22

Results

15,00 10,00 5,00

0,00

50 ml/min

70 ml/min

Figure 2: values of the specific flow rate of the RoweFil PA66+ The filter system RoweFil with a PES-membrane has a specific flow rate between 7.5 and 15.6 ml/(min*bar*cm²) in the measurement range. The results are shown in figure 3.

15,00 10,00 5,00

15,00 10,00 5,00 0,00

45 ml/min

75 ml/min

110 ml/min

Figure 3: values of the specific flow rate of the RoweFil PES A specific flow rate between 12.3 and 15.8 ml/(min*bar*cm²) was calculated for the filter system RoweFil with the PET-membrane. Figure 4 shows the results at the 4 measurement points.

30 ml/min

55 ml/min

80 ml/min

20,00 15,00 10,00 5,00 0,00 4 ml/min

0,00

120 ml/min

Figure 5: values of the specific flow rate of the RowePaed PET Figure 6 shows the specific flow rate of the infusion filter for neonats, RowePaedM with a PET-membrane. It has a specific flow rate between 12.5 and 16.2 ml/(min*bar*cm²).

specific flow rate / (ml/min cm² bar)

20,00

100 ml/min

20,00

5 ml/min 30 ml/min

60 ml/min

The filter system for the application on children RowePaed PET has a higher specific flow rate than the according filter system for the application on adults. The specific flow rate varies between 16.6 and 19.5 ml/(min*bar*cm²). The results are shown in figure 5.

0,00

10 ml/min

5,00

Figure 4: values of the specific flow rate of the RoweFil PET



20,00

10 ml/min

10,00

20 ml/min

The specific flow rate was analysed for 6 test samples of each filter system at 4 measurement points. The average value of the specific flow rate was calculated of the 6 results and is shown in the diagrams for each filter system together with the standard deviation. Figure 2 shows the values of the specific flow rate for the filter system RoweFil with positively charged PA66+filter membrane. The specific flow rate varies between 5.2 and 9.1 ml/(min*bar*cm²).


15,00 specific flow rate / (ml/min cm² bar)

Filtersystem RoweFil RoweFil RoweFil RowePaed RowePaedM

S896

10 ml/min

16 ml/min

22 ml/min

Figure 6: values of the specific flow rate of the RowePaedM PET To compare the results of the different filter systems to each other a measurement point with similar pressure is chosen. The value of the specific flow rate at a pressure around 300 to 400 mbar corresponds to the second measurement point of each filter system. The measurement points and the results for this comparison are shown in table 3.



Table 3: values of the flow rate and the specific flow rate at a pressure between 300 mbar and 400 mbar Filtersystem RoweFil RoweFil RoweFil RowePaed RowePaedM

Membrane PES PA66+ PET PET PET

Flow rate / Specific flow rate / ml/min ml/(min*bar*cm²) 60 11.0 30 7.7 60 13.9 30 19.3 10 14.0


An overview about the specific flow rates at the chosen measurement points is given in figure 7.

branes. It depends on the porosity of the filter membrane [2]. The results of the 3 filter systems which have the same PET membrane but differ in size and shape show that the geometry of the housing has an influence too. The reason could be that the flow is not laminar and turbulences are different in different housings. For the determination of a specific flow rate of a complete filter system it is necessary to involve the housing and not only the filter membrane.

5

Acknowledgement

This project was partially funded by the European Regional Development Fund (ERDF) and the European Social Fund (ESF) within the collaborative research between economy and science of the state MecklenburgVorpommern (V-630-F-132-2011/254; V-630-S-1322011/253)

20,00 15,00 10,00

6

5,00 0,00 RoweFil PES

RoweFil PA66

RowePaed PET

RowePaedM PET

RoweFil PET

Figure 7: comparison of the specific flow rate values of all test samples at a pressure between 300 and 400 mbar The specific flow rate for the filters with the same size and different membrane materials and structures is different. Comparing the 3 filter systems for the application on adults RoweFil PET has the highest and RoweFil PA66+ the lowest value. The 3 filter systems with the same PETmembrane and different filter sizes have also different values for the specific flow rate. The filter system RowePaed PET has the highest specific flow rate of 19.3 ml/(min*bar*cm²) at a pressure between 300 and 400 mbar whereas the filter with the smallest surface area (RowePaedM PET) and the one with the highest (RoweFil PET) have almost the same specific flowrate at this point.

4

S897

References

[1] Norm DIN 58355-1 (2005), Membranfilter – Teil 1: Prüfung der spezifischen Durchflussrate von Flüssigkeiten für Flachfilter, Deutsches Institut für Normung e.V., Beuth Verlag GmbH [2] Klaus Ohlrogge, Katrin Ebert (2006): „Membranen: Grundlagen, Verfahren und industrielle Anwendungen“, p. 201, WILEY-VCH Verlag, Weinheim [3] X. Wang, F. Thauvin, K.K. Mohanty (1998): NonDarcy flow through anisotropic porous media, Chemical Engineering Science 54, p. 1859-1869

Conclusion

After the theory of Darcy’s law the flow rate of membrane filters is proportional to the transmembrane pressure [2]. But the results of the measurement show that the specific flow rate of infusion filters depends on the flow rate. It is not a constant value for each membrane filter. Inspite of the filter system RowePaed PET the specific flow rate becomes smaller with an increasing flow rate and pressure. That means that the transmembrane pressure increases more than the flow rate. This effect could be caused by turbulences in the fluid which vary with the flow rate. [3] Comparing the filter systems with the same size and the same geometry it is obvious that the specific flow rate varies because of the different structure of the mem-



A novel diagnostic tool for therapeutic monitoring during the treatment of Pectus Excavatum with the vacuum bell D. Hradetzky1, S. Weiss1, F.M. Haecker2, S.B. Sesia2 1 University of Applied Sciences and Arts Northwestern Switzerland, Institute for Medical and Analytical Technologies, Muttenz, Switzerland [email protected] 2 University Children’s Hospital of Basel, Department of Paediatric Surgery, Basel, Switzerland

Abstract Pectus Excavatum represents the most common congenital chest wall deformity in children and adolescents. Surgical treatments based on Ravitch technique or minimal-invasive repair by Nuss are well-established procedures. A noninvasive treatment utilizing the vacuum bell represents a non-surgical alternative. For monitoring the efficiency of the procedure as well as to select appropriate patients for this treatment, currently no suitable diagnostic tool is available. To classify the mechanical properties of the chest wall a new measurement system was designed and tested, providing the pressure related elevation of the sternum during the application of the conventional vacuum bell. First in vivo measurements on selected patients indicate a correlation of age and pressure related to chest wall elevation.

1

Introduction

1.1 Pectus excavatum Pectus excavatum, also known as funnel chest, describes an abnormal development of the rib cage, leading towards a concave shaped surface of the thorax. This deformity tends to affect the 4-5 ribs on each side of the sternum, is often congenital, and occurs in 1/300-400 births [1]. Current treatment of this chest wall deformity is performed by the application of open surgery (Ravitch procedure) or the minimal-invasive repair technique by Nuss (MIRPE), latter providing 95% good to excellent outcome in long-term results [2].

Figure 1: Vacuum bell consisting of a hand pump (1) and a silicone suction cup (2) with integrated observation window (3).

In contrast to surgical therapy, an innovative non-invasive method, based on a vacuum bell (Figure 1), is gaining attention [3]. A suction cup (2, 3) is applied to the chest wall of the patient and a patient-activated hand pump (1) is reducing the pressure up to 15% below atmospheric pressure within the cup. It is recommended to apply the vacuum bell twice a day for duration of approximately 30 minutes each [4]. Due to the application the sternum will elevate as long the reduced pressure is applied, and a little residual chest elevation may be obtained up to 1cm per month, when using the suction cup repetitively [4]. Applying the treatment for a period of 18 month in some cases (13,5%) an residual elevation of the sternum to the normal level was achieved [5]. 1.2 Diagnostic tools A major lack in diagnostics and treatment of pectus excavatum is a missing suitable diagnostic tool to quantify the severity of the deformation and to monitor the effectiveness during the vacuum bell treatment. To quantify the severity of pectus excavatum the Haller Index is widely used [6], describing the ratio of the transverse diameter inside the ribcage and the shortest distance between vertebrae and sternum at the deepest point of the deformity. A severe condition relates to a Haller index larger than 3.25, while the normal value is related to an index of 2.56 [1]. These data are acquired from CT images and therefore it is less suitable for monitoring the deformity during the current treatment. A simple depth gauge arrangement utilizing a mechanical ruler is often used to measure the depth of the deformity, leading to poor accuracy of measurement. Therefore an improved ruler with a rigid reference plane was designed, fabricated, and is currently in use (Figure 2). Furthermore the success of the non-invasive treatment is expected to rely on the mechanical properties of the chest wall, in particular on the elasticity and plasticity. Up to now there is no device available, neither to measure these



flexibility parameters nor to monitor the elevation of the chest wall during the application of the suction cup.

method was chosen, detecting the height change outside the suction cup ( , ), within the suction cup ( ) and to calculate the chest elevation ( ).To detect a tilt precisely at least three detection points outside are required. For applicability issues currently only two outer points are used, in line with the third detector for measuring the height within the suction chamber ( ).

Figure 2: Improved gauge for measuring the depth of a funnel chest consisting of a stainless steel ruler with an engraved millimeter scale and a reference plane (polycarbonate). Within this paper an approach towards a pressure related chest wall elevation measurement system is presented. This device intended to be used in clinical scenario first, in order to obtain a pressure-elevation curve. From these obtained data it is expected to gather information about the effectiveness of the treatment, as well as predictive information whether this therapeutic approach will be effective for the individual patient, which procedural parameter are expected to provide the best outcome (duration of a single treatment, repetition rate, pressure to be applied) and in best case an assumption of the expected overall treatment time line. In a later stage this device might be integrated within the home used vacuum bell, to assure an optimized use of the medical device.

2

Methods

2.1 Vacuum bell – the therapeutic tool The vacuum bell according to Eckhart Klobe [4] consists of a suction cup made from a flexible material and a rigid observation window (Figure 1; item 2 and 3). During the application an elevation of the sternum up to the observation window may occur, depending on the mechanical properties of the chest wall and the applied reduced pressure. As there is no pressure sensor integrated, the pressure during the application relies on the experience of the medical supervisor and especially on the user himself. Due to the reduced pressure inside the suction cup (blue), indicated in Figure 3 with dotted lines, the chest wall (red) will be elevated ( ) towards the observation window, while the flexible suction cup also will approach towards the chest ( ). 2.2 Measurement method It is intended to determine the elevation of the chest ( ), dependent on the applied pressure ( ), within the standard instrument used for therapy. Therefore it is essential to take into account the change of height of the suction cup ( ) and furthermore a tilted change of height due to unsymmetrical body conditions. Therefore a measurement

Figure 3: Cross section of the vacuum bell (blue) applied on the sternum (red) without reduced pressure (solid lines) and during application with reduced pressure (dotted lines). The pressure dependent chest elevation ( rived according to ∑

) can be de-

(2.1)

denotes the pressure at the atmospheric level where (reference), and denotes the number of detectors outside the suction cup, in presented case ( 2). 2.3 Modification of the vacuum bell Three commercial available sensors, based on optical triangulation (Sharp GP2Y0A41SK0F), optical intensity (Vishay Semiconductor VCNL400) and ultrasonic time of flight (HC-SR04), where evaluated regarding their suitability for this application. The suitability is dedicated to the accuracy and reproducibility ( 1 ) of the distance measurement ( ) within a specified range of 0 15 . As target different gray scales level on paper were used. The measured data are referenced to the data of a commercial milling center (Picomax 20, accuracy 10µm), where the sensors were attached to. As further evaluation criteria the size of the sensor, the covered measurement range and the capability to measure through the observation window was taken into account, latter in order to modify the vacuum bell as less as possible. For measuring the pressure within and outside the suction chamber a differential pressure sensor (Honeywell 26PCBFA6D 5PSI) based on piezo resistive bridge, was applied. The pressure sensor is placed within the tubing from the hand pump towards the suction cup. Interfacing the sensors to a standard personal computer, a micro controller interface (Arduino Mega 2560) was used, correcting non-linearity of the sensors, performing calculation of the chest elevation ( ) according equation (2.1),



and transferring height and pressure data wired (USB) and wireless (Bluetooth). The mechanical set up of our measurement device, shown in Figure 5, is strongly related to the selection of the suitable sensor. An assembly on the chest wall lifer without remaining mechanical changes on the therapeutic device itself is preferred. Therefore a mechanical support was designed and produced using an additive manufacturing process on an Eden260TM system.

For the technical characterization of the set up a thorax model was realized, consisting of a tri-zone plate with three concentric zones, the inner and outer made of a rigid polymeric material, and the interconnection zone in between made from elastic polymer. With this arrangement a simulation of the pressure related changes of the vacuum bell and the chest wall itself can be simulated, in order to distinguish between deformation of the suction cup and the thorax model. The real elevation of the centre zone was evaluated from the bottom through an access whole using precision caliper (Figure 6).

Figure 6: Schematic view of the measurement arrangement with the circular tri-zone measurement plate.

3 Figure 4: Assembly of the measurement set up on a vacuum bell. The micro controller is located within a separate module (upper right). The set-up is positioned on a circular measurement plate. The two external distance sensors (Figure 5 (4)) are aligned in-line with the sensor (5) located in above the middle of the observation window. The distance of both external sensors are adjustable, in order to obtain a flexible setup, noteworthy the required symmetrical arrangement of both sensors. As there are currently three different sized vacuum bells available, a clamp fixation (1) was realized, adjustable to all vacuum bell designs.

Figure 5: Arrangement of the experimental setup consisting of three mechanical fixations (1), adjustable (2) to different sizes of the vacuum bell (3), providing two external distance sensors (4) aligned in-line with the suction cup optical distance sensor (5).

Results

3.1 Sensor evaluation Concerning the required measurement range none of the sensors is capable to cover the whole range. In particular none of them is suitable for the lower distance range, while the acoustic and triangulation sensor are suitable for distances larger than 15 cm. Only the triangulation sensor is able to provide a measurement through the observation window. Sensor (white paper)

Sharp Vishay HC-SR04

4 cm 0.1 cm 4 cm

> 15 cm < 10 cm >>15 cm

1.07 mm 2.52 mm 1.58 mm

The tow optical sensors provide none linear distanceoutput relation, while the acoustic sensor offers a linear behavior. Unfortunately the acoustic sensor is not capable to measure through the observation window, and additionally it is not suitable to be integrated within the vacuum bell due to the size. The differential pressure sensor provides linear correlation relation, with a standard deviation of 0.47 . For the instrument the optical triangulation sensors where chosen. This decision is induced to the fact, that all three sensors provide comparable performance, while only the triangulation sensor was able to measure through the observation window. Nevertheless the mechanical set up is designed to host all three types of sensors.



3.2 Measuring device – technical characterization Applying the measurement set-up shown in Figure 6, and calculating the chest wall elevation according equation (2.1) we observed an accuracy of the measurement in the sub-mm range. Furthermore the accuracy is independent from the reflectance (gray scale) of the targeted region. In contrast to the sensor evaluation a single measurement point is based on the average of six succeeding independent measurements. 3.3 Measuring device - first application on patients After institutional review board approval and written consent, the system was used during standard clinical consultation of patients suffering from pectus excavatum. The patients did use the vacuum bell by themselves, while the measuring device was monitoring the chest wall elevation, pressure and time. The measurement points were aligned in longitudinal mode on the chest, while the suction cup was placed centered on the deformity. Two typical examples of obtained data are shown in Figure 7 (11 year old patient) and Figure 8 (17).

The resulting maximum of elevation of the younger patient 1.8 at a reduced pressure of (11yrs, Figure 7) of 115 differs significantly from the results of 1 at the older patient (17yrs, Figure 8) 140 giving a first evidence for age dependency of the mechanical properties of the chest wall. Furthermore a time delay of the chest wall elevation can be observed, resulting in a still elevating chest wall if the pressure is kept constant (Figure 7) or even if the pressure is slightly reduced (Figure 8).

4

Within this paper a novel approach is presented to monitor the chest elevation during the treatment with a vacuum bell. To our knowledge for the first time physicians are enabled to gather pressure related elevation information in a clinical scenario. An age-dependent difference in elevation of the chest wall is shown and has to be proofed evaluating more patients. Furthermore a mechanical model of the chest wall has to be elaborated, in order to evaluate the obtained pressure and elevation data in more detail.

5

Figure 7: Monitored elevation (+) and pressure (x) during a single short-term application of the vacuum bell on an 11 year old patient. The signal to noise ratio (elevation) is significantly increased, compared to the technical characterization. This is expected to be caused mainly by respiration and movement of the patient during the measurement, as well as increased scattering of the light on human tissue.

Conclusion

References

[1] L. Molins, et al., "Chest wall surgery: Nuss technique for repair of pectus excavatum in adults," Multimedia Manual of Cardio-Thoracic Surgery, vol. 2007, January 1, 2007 2007. [2] D. Nuss, "Minimally invasive surgical repair of pectus excavatum," Semin Pediatr Surg, vol. 17, pp. 209-17, Aug 2008. [3] F. M. Haecker and J. Mayr, "The vacuum bell for treatment of pectus excavatum: an alternative to surgical correction?," Eur J Cardiothorac Surg, vol. 29, pp. 557-61, Apr 2006. [4] F. Schier, et al., "The vacuum chest wall lifter: an innovative, nonsurgical addition to the management of pectus excavatum," J Pediatr Surg, vol. 40, pp. 496500, Mar 2005. [5] F. M. Haecker, "The vacuum bell for conservative treatment of pectus excavatum: the Basle experience," Pediatr Surg Int, vol. 27, pp. 623-7, Jun 2011. [6] J. A. Haller, Jr., et al., "Use of CT scans in selection of patients for pectus excavatum surgery: a preliminary report," J Pediatr Surg, vol. 22, pp. 904-6, Oct 1987.

Figure 8: Monitored elevation (+) and pressure (x) during a single short-term application of the vacuum bell on a 17 year old patient.



S902

A Prototyping System for Smart Wheelchairs N. Fränzel1, F. Weichert2, A. Wenzel2 and C. Ament1 1 Inst. for Automation and Systems Engineering, Technische Universität Ilmenau, Ilmenau, Germany, [email protected] 2 Embedded Systems, Fraunhofer IOSB-AST, Ilmenau, Germany

Abstract This paper describes a mobile rapid prototyping system for the development and evaluation of assistance systems for wheelchair users. The realized system is capable of manipulating the driver’s input in real-time, which enables a shared control, without applying major changes to the original wheelchair. For this purpose, the user of the prototypting system does not need a detailed knowledge of the original control system as its input/output is available in an abstract manner. Thus novel assistance concepts can be easily developed and evaluated with patients under realistic conditions.

1

Introduction

Wheelchair control is not within its direct scope, although the projects methods might be feasible.

As the world’s population is increasingly aging [1], more people suffer from different impairments such as immobility [2] and will rely on medical aids to maintain some degree of independence and mobility. To increase the personal independence further, the use of assistance systems might be feasible to compensate the decline of physical and mental capabilities caused by geriatric diseases [3]. Various assistance systems have been developed and deployed for passenger cars and trucks such as collision avoidance systems and automated parking systems. However, regarding the field of wheelchairs hardly any assistance system is available that is not in the state of research and development. One major obstacle for the development of these systems is the communication and information exchange with the wheelchair control system. The actual integration of such systems is often as challenging as their development, evaluation and parameterization. Furthermore, some realizations are not fit to be evaluated by either real wheelchair users or outside the lab, because patients have different, often multiple, impairments or the system’s size or power demands are not appropriate respectively. To ease the development and evaluation of novel assisting technologies a system is needed that integrates with the existing power train and human-machine-interface (e.g. joystick or sip and puff control).

2

1.1

Image 1 SAM Boxes. The opened case shows a prototyping area for new sensor interfaces.

Related Work

Technologies originally developed for mobile robots have been successfully applied to wheelchairs before, e.g. the NavChair avoids obstacles and manipulates user commands [4]. The required modifications of the system and the component’s space requirements have been further minimized in the Hephaestus Smart Wheelchair System [5]. However, these systems are rather focused on the obstacle avoidance and driving capabilities. The easy integration of different human-machine-interfaces and their evaluation in conjunction with new algorithms is not directly addressed. In contrast, the AsTeRICS project provided tools for the rapid development of user centered assistive technologies [6].

Methods

The base of the developed prototyping system consists of a wheelchair with a CAN based control and the Smart Assisted Mobility (SAM) Box, see image 1, which interferes with the wheelchair communication bus and runs the assisting algorithms, and accompanying software modules to design them either graphically or with high-level programming languages. The SAM Box is installed between the user input device and the master controller. Such a so-called man-in-the-middle attack has been successfully demonstrated for wheelchair control before [4]. However, our approach enables not only the manipulation of the driving commands, but also the acquisition of CAN bus messages, such as speed and battery voltage.

2.1

System Design

The SAM Box is a heterogeneous system with two modules for dedicated tasks respectively. The first module is a BeagleBoard-xM (BB) for data processing and control. Furthermore, this board offers USB connectivity and a network interface which enables the integration of widely available peripherals for personal computers. The second module is designed as a Daughterboard (DB) for the first one and is based on a microcontroller for CAN data reception and manipulation. Furthermore, eight auxiliary analog inputs and eight digital in-/outputs are provided for additional sensors



and actuators. The DB provides power to the BB from the vehicle bus and turns it on/off depending on the state of the wheelchair. Hence it conserves the wheelchair’s battery while not in use. The SAM Box is transparent to the user system and does not require additional modifications of the wheelchair. The power demand of only 2.7 W at full load is negligible compared to the wheelchairs idle power consumption of nearly 15 W. The advantage of the heterogeneous design is the strict separation between the actual assistance algorithms and the wheelchair interface. Even if the the assistance system fails during the evaluation by patients, the wheelchair can still be operated without the additional assistance like an unmodified system. Hence long term tests outside a lab are feasible to perform a real-life evalutation.

2.2

loop paradigms can be applied as well. Even if the model is running on the BB, internal values can be accessed from the host computer which enables an accelerated design and tuning of novel algorithms. After a successful paramterization the obtained values can be used for the stand alone operation of the SAM Box without the further need for the host computer.

Software

The software is split into three parts. The first part is the firmware of the Daughterboard. It handles the information exchange between the input device (e.g. joystick) and the master controller as well as between the CAN bus and the BB. The base functionality is self-contained, thus the basic operation of the wheelchair does not rely on additional information or the BB in general. If a message from the joystick is received, it is forwarded to the master as long as the control is not claimed by the assistance system. The second part is the software within the BB. It runs an embedded Linux (3.12.6-rt Kernel) with a minimal Ubuntu system, which enables the use of software repositories for an accelerated development by the use of open-source libraries. The third part, the actual assistance system in charge of the wheelchair’s shared control, runs on the BB and communicates with the daughterboard via an internal UART interface. The default data rate for the internal data exchange is 100 Hz, which is sufficient to capture and process the input from an input device (see image 2). Depending on the sensor interface, digital and analog output can be acquired by the Daughterboard and processed at the BB. However, more sophisticated sensor interfaces such as Ethernet or USB have to be processed by the software in the BB. By default the user input is unaltered. If the assistance system has to interfere, new commands are sent to the daughterboard overwriting the user data. Furthermore, it is feasible to process the user commands and, depending on additional sensor readings and the user commands, adjust them before they are received by the master controller in charge of the motors.

2.3

S903

Development Process

The required time to develop and implement new assisting technologies is significantly reduced by means of model based design using e.g. Matlab/Simulink [6]. Algorithms that have been verified in simulation can be evaluated in the real world with users by the use of library components specifically designed for the SAM Box. By adding those blocks to a simulation model, it can be compiled and afterwards transferred to the wheelchair. Thus the real-time processing of user input can be combined with previously developed algorithms without additional effort. Moreover, captured data can be used as input for simulations and software in the

Image 2 Data flow in the SAM Box for development and stand alone operation.

2.3

Smartphone Interface

Considering the growing penetration rate of smartphones in many countries it is obvious to deploy them in the context of a smart wheelchair. Smartphones may offer a convenient user interface as many people are familiar with them by their daily use. In order to provide a convenient interface between the SAM Box and smartphones a dedicated smartphone application has been designed. This application exchanges data with the BB using the Android Open Accessory Protocol [7], thus enabling the use of the smartphone’s sensor as well. The sensor readings and the user input are accessible from the model running in the BB via dedicated library blocks. This interface is bidirectional, i.e. the application can visualize output from the model. The data exchange is encapsuled within a Java class for the smartphone and a corresponding C function for the BB, i.e. it can be easily adapted to the actual requirements and be reused for other applications. Furthermore, the smartphone can act as a mobile WLAN hotspot, that provides internet access for the SAM Box and the model running on it likewise.

Image 3 Simple model with smartphone visualization.



3

Results

To validate the system design its timing is examined and the actual implementation of simple assistance systems as proof of concept is demonstrated.

3.1

Timing Behaviour

The model running on the BB has to process the data at a frequency of at least 100 Hz as shown in image 2. This timing requirement has been tested with a higher frequency of 1 kHz via cyclictest [8] as indicated by Claudi et.al. [9]. If a frequency of 1 kHz can be hold steadily, a 10 times lower frequency should not present a problem. Image 4 shows the results of 50 million samples wherby 548 µs is the maximum delay. This indicates that a processing frequency of 100 Hz with a jitter of approx. 5% is achievable with the BB and the Linux system.

S904

input can be modified in real-time with a frequency of 50 Hz and that the user input can be processed before it is transmitted to the master for motor control.

3.2

Data Fusion

To demonstrate the potential of the SAM Box data from different sources has been acquired simultaneously (see image 6). The user command corresponds to the joystick forward deflection, the velocity is acquired via CAN messages from the master and the wheelchair’s pitch is measured by an application running on the Android smartphone that is connected to the SAM Box via USB.

Image 6 Example of data aggregated by the SAM Box

Image 4 Results of the cyclictest. The timing for the whole SAM Box has been evaluated by a model running on the BB that toggles one digital output pin on the DB every cycle, i.e at 100 Hz. As all set points are sent within a single frame, all other values can be updated likewise.

The SAM Box enables a synchronized acquisition of data from various sources and its storage for further analysis. By combining wheelchair data with external sensors new opportunities are available for more sophisticated assistance systems as compared to using only external sensor readings.

3.3

Rear View Camera

For this application a USB webcam (see image 7) is connected to the SAM Box and the smartphone is used as a video display. As soon as the wheelchair user drives backwards, the SAM Box starts an application on the smartphone that shows the life video stream provided via the WLAN connection. The smartphone can be used for other not necessarily assistance related purposes while the wheelchair is not moving backwards. In consequence, a simple non-intrusive situation dependent system for assisting the driver can easily be created.

Image 5 Oscilloscope recording of the digital output. The oscilloscope recording (see image 5) shows one digital output toggling every 10 ms without a noticeable jitter. The DB acts as a sample and hold element and eliminates the BB induced jitter thusly. Therefore it is ensured that the user

Image 7 Wheelchair with SAM Box, webcam and ultrasonic sensors at the rear.



3.4

Collision Avoidance

The idea of the rear view camera system can be improved by not only providing information on the environment, but also by manipulating the user input in order to avoid a potential collision. Ultrasonic sensors provide the distance to obstacles via the analog input pins. As soon as the distance to an obstacle decreases below 0.8 m (red line in image 8), the user is not allowed to go any further backwards. The SAM Box takes over the control of the wheelchair and overrides the user input to prevent a collision with an obstacle that might not be visibile to the driver. However, the forward movement is unaffected by the assistance system. The set threshold of 0.8 m accounts for the wheelchair’s extent beyond the actual sensor position and the braking distance.

4

S905

Conclusion

The SAM Box is a flexible rapid prototyping tool for the accelerated developement of novel assistance systems. The provided abstraction of the vehicle bus access shifts the focus of future work to the actual implementation of such systems and away from their integration. The separation between the assisting algorithms and their interface to the wheelchair system enables a secure evaluation with real wheelchair users during real-life situations. A supervisor ensures the reliable operation of the wheelchair at all times. Furthermore, this separation offers the possibility to adapt the SAM Box to different vehicles in general. The small size and power demands of the system permit even long term tests without affecting the wheelchair’s extent and driving range. Future work will concentrate on the development of new input devices for wheelchair users, because an reliable interface to the master controller is available via the SAM Box. This work was supported by TAB (Thüringer Aufbaubank) and ESF (Europäischer Sozialfonds für Deutschland), #2011FGR0127

5

Image 8 SAM Box prevents potential collision. The combination of the rear view with the collision avoidance system supports the user while driving backwards by providing distance measurements (see image 9). If the driver ignores this information and is about to collide with an obstacle, the assistance system manipulates the user input and prevents potential accidents. This behavior can be turned of by the user to handle emergency situations.

Image 9 Screenshot of the android application (left) and smartphone attached to a wheelchair running the assistance system (right).

References

[1] U.N. Department of Economic and Social Affairs, Population Division: “World Population Ageing: 19502050”, United Nations Publications. Sales No. E.02.XIII.3, 2002 [2] Seeman, T. E., Merkin, S. S., Crimmins, E. M. and Karlamangla, A. S.: “Disability Trends among Older Americans: National Health and Nutrition Examination Surveys, 1988–1994 and 1999-2004.” American Journal of Public Health 100(1), 100-107, 2010 [3] Liang, C., Yang, Y., Lin, Y., Kang, S., Lin, P., Chen, Y.: “BotBeep — An affordable warning device for wheelchair rearward safety”, International Conference on Orange Technologies (ICOT), pp.159,163, 2013 [4] Levine, S. P., Bell, D. A., Jaros, L., Simpson, R., Koren, Y. and Borenstein, J.: “The NavChair Assistive Wheelchair Navigation System”, IEEE Transactions on Rehabilitation Engineering, vol. 7, no. 4., 1999 [5] Simpson, R. C., Poirot, D. and Baxter, F.: “The Hephaestus Smart Wheelchair System”, IEEE Trans. on Neural Systems and Rehabilitation Engineering, vol.10 no. 2, 118-122, 2002 [6] Veigl, C., Weiß, C., Kakousis, K., Ibáñez, D., SoriaFrisch, A. and Carbone A.: “Model-based Design of Novel Human-Computer Interfaces”, Biosignals and Biorobotics Conference (BRC), 2013 [7] http://source.android.com/accessories/protocol.html [8] https://rt.wiki.kernel.org/index.php/Cyclictest [9] Claudi, A.; Dragoni, A.F.: “Testing Linux-based realtime systems: Lachesis”, International Conference on Service-Oriented Computing and Applications (SOCA), 2011



S906

Cell-Shape Wizard – A Concept for User-Guidance for Active Shape Segmentation in Fluorescence Cell Micrographs 1,2 1 1,2

D. Franz , V. Wiesmann , M. Stamminger², T. Wittenberg Fraunhofer-Institute for Integrated Circuits IIS, Erlangen, Germany ² University Erlangen Nuremberg, Erlangen, Germany 1

Abstract Fluorescence micrographs are acquired for the quantitative evaluation of different types of cell experiments within the fields of e.g. virology, micro-biology, or immunology. Life-scientists either assess and evaluate these micrographs manually or automated with adequate software tools. Nevertheless, the usage and parameterization of these image processing and analysis methods is a challenging task, which is often infeasible for life scientists with no or little image processing or computer science knowledge. In order to bridge the so-called “semantic gap” between high level and complex to parameterize image processing methods and their life scientist user, we propose the use of the wizard interaction pattern. In this contribution, we will review existing wizard approaches and guidelines for interactive fluorescence micrograph segmentation and derive a practical concept for a cell segmentation wizard, using active contours and active shape models as one high level segmentation approach. The proposed concept for a Cell-Shape-Wizard integrates and couples tight user guidance with the benefits of interactive cell segmentation in fluorescent micrographs.

1

Introduction

Cell experiments within the field of life sciences, such as virology, immunology or microbiology are most often evaluated on the basis of fluorescence micrographs. As the number and image size of cell micrographs increases by the availability of high-through put scanning systems, a manual examination of the depicted cells and cellular structures in the resulting micrographs is prone to interand intra-observer errors. Furthermore, it is a timeconsuming task and typically, only a small subset of the available micrographs and depicted cells can be analyzed manually within a limited amount of time. With the help of automated or interactive image processing and analysis tools, a much larger subset of this image data can be assessed and evaluated, hence increase validity and reproducibility of fluorescence micrograph evaluation [1]. Examples for some of these image analysis approaches are available to the community in form of free micrograph image processing software tools, as e.g. CellProfiler [2], Daime [3], Blobfinder [4], ImageJ [5, 6] or MiToBo [7]. A typical fluorescence micrograph analysis consists of two steps, the detection and segmentation of cells and cell structures (e.g. nuclei, plasma, skeleton, ...) and a subsequential measurement part. In the measurement part, the segmentation of the detected cells is translated into measurements such as the size, the number, intensity, texture, colocalization, translocalization or the the ratio of cells and cellular structures. The cell segmentation step can be further divided into fore- and background separation and cell separation. In some fluorescence micrographs cells and cellular structures are well separated and do not touch or overlap. Here, often global or local thresholding approaches are sufficient, and for cell segmentation and no cell separation step is necessary. In the other case, when cells touch or overlap, a cell separation step is necessary and requires the application of more complex segmentations approaches, such as level sets, active contours (such as snakes) or active shape models (ASMs).

Image 1 Application based menu structure of MiToBo [7]. Even if such complex approaches are available in some software tools, they are difficult to parameterize and to apply to cell segmentation and cell splitting, due to their complexity, that most likely only image processing experts are able to adjust and fine-tune the parameters. Unfortunately, in these cases the functional scope of the software tools is quite challenging for non-experts. Unfortunately, many image processing tools focus on functionality rather than usability and are therefore difficult to use, especially for non-image processing experts with no or little background in this field. I other words: There exists a so-called “semantic gap” between high-end image processing tools and their addressed non-imageprocessing-expert user-group. Hence, it is our goal to bridge that gap, by the introduction, implementation and use of image processing wizards.

1.1

Objective

One well understand way to bridge the above named semantic gap is the wizard interaction pattern for software



tools. In this contribution, we will first review the application of the wizard interaction pattern in some existing medical image analysis tools, which are applicable to fluorescence micrographs. Secondly, we will transfer guidelines for the application of the wizard interaction pattern from other research areas to the application of fluorescence micrograph analysis, and specifically provide a wizard concept - the Cell-Shape Wizard - for the application of active contours and active shape models for cell segmentation and separation in fluorescence micrographs.

1.2

Related Work

As mentioned above, for fluorescence micrograph segmentation there exist many software tools with a string focus on functionality but which are less focused on user guidance. Tools that integrate user guidance are e.g. CellProfiler [2,8] or Mitobo [7]. With CellProfiler (see Image 2), a user designs own image processing pipelines from modules that encapsulate functionalities from image processing over object detection to object analysis. A “help button” for each parameter describes the parameter and related guidelines for the adjustment of the parameter. Specifically, the graphical user interface has been developed by a life-scientist and therefore matches life-scientists’ expectations [7]. MiToBo (see Image 1) is an ImageJ plugin [5,6] and groups functionalities due to life science problems rather than image processing categories. Hence, parameter tuning with MiToBo remains challenging. Both tools provide a high amount of functionality to solve various tasks in cell segmentation. They provide wizards and problem oriented menu structures to structure the functionality. In contrast to these examples, we propose another user guidance concept for cell segmentation, in which we deliberately restrict the functionality and flexibility of a tool and rather guide the user through the cell segmentation task.

S907

two important statements, (1) Less is More and (2) Less Flexibility, More Guidance. We use these statements for the concept of a user-guided fluorescence micrograph segmentation task based on active contours and active shape models and depict how the wizard interaction pattern fulfills Mühler’s statements in the following.

2.1

Wizard interaction pattern

The wizard interaction pattern [10] is a divide-and-conquer approach, that separates a complicated task into a series of steps, that are each easy to solve (see Image 3). The first advantage of the wizard interaction pattern is a shorter orientation phase with a new tool. Additionally, the standardized workflow in combination with recordable user interaction leads to reproducible results. Nevertheless, a disadvantage of the wizard interaction pattern is a reduced flexibility, as the task’s structure and parameter details are hidden. On the other side, this will lead to a higher acceptance of the users. This observation agrees well with Mühler’s statements [9]. Specifically, the wizard interaction pattern is applicable in cases, where the user does not need or want an overview or fine control over the given image processing task, which is exactly the case for nonimage-processing-experts. The wizard interaction pattern has already been adapted to various image processing problems in medical image processing. Examples are the wizard-guided segmentation of ear structures [12], wizards for filtering in AMIDE [13] or a wizard for neurosurgical intervention planning [14]. An approach similar to our approach is SNAP [15], a guided workflow for active contour segmentation in medical data. With this tool, the users set spherical seeds, and describe the parameters in terms of their impact on the behavior of the evolving contour and furthermore display the iteratively deformed contour. The speed function of the active contour is visualized and shows the effect of each parameter on the total speed force. The contour deformation can be steered in a video-player-like fashion with a “Play”, “Stop”, “Rewind” and “Single Step” button. Our approach for cell segmentation in fluorescent micrographs applies a 2D active contour instead of 3D. Additionally, the speed function is extended with a deformation term based on an active shape model (ASM). We apply the active contour for interactive segmentation and use the wizard interaction pattern to mask image processing terms.

Image 2 User interface of the CellProfiler with local parameter documentation.

2

Methods

Mühler et al. [9] gave some guidelines for the design of surgical planning software and condense their findings into

Image 3 Basic structure of a wizard page, adapted from [11].



3

Image 4 The schematic workflow of the standard segmentation workflow (top) and the proposed segmentation workflow integrated into the Cell-Shape Wizard (bottom).

2.2

Interactive cell segmentation

Olabarriaga and Smeulders have identified reasons, why automated segmentation methods might fail [16]. We have translated them into the use case of cell segmentation in fluorescence micrographs: (1.) Visual evidence is absent: A cell border is not visible, as it is either overrun and occluded by neighboring cells or there is no border. Solution: A trained observer or a shape model extrapolates missing data and repositions missing borders. (2.) Visual evidence deviates from assumptions: Either the cell shape of a new cell has not been included in the used model or the image quality is not sufficient to adapt the model to the image. The segmentation will fail even if there is visual evidence which might support the assumptions. Solution: The user adapts parameters, changes the model, or corrects the segmentation manually. Both cases occur in fluorescence micrograph segmentation and both cases are well suited to be solved with the ASMsupported active contour. In the first case, a well-trained ASM should be able to extrapolate missing cell borders. In the second case, a manual segmentation correction can be used directly to retrain the ASM and adapt the new model to cell shapes that have so far not been included in the model before. Image 4 depicts the integration of a segmentation wizard for the second case in to the standard image processing work-flow. For the training of the ASM model we suggest the application and extension of the approach of Held et al. [17] which learns the best possible combination of the parameters of the image processing steps in a pipeline from a set of given manual segmented cells. In this approach user interaction is necessary and consists of a manual annotation of cells. The manual segmentations approach can also be used to train or retrain active shape models of cells.

S908

Results

Following the design criteria discussed in the previous section we propose the concept of a Cell-Shape Wizard that guides users through ASM-supported active contour cell segmentation of fluorescence micrographs. The basic structure of a standard wizard page (see Image 3) is translated into the steps of the Cell-Shape Wizard. 5a and 5b depict a mockup of three Cell-Shape Wizard pages with title and action (on the top of the wizard page), interaction buttons (on the bottom of the page) and a combined view of the fluorescence micrograph and the segmentation results (image part of the wizard page). In comparison to the standard active contour cell segmentation pipeline, the Cell-Shape Wizard provides a control loop to iteratively adapt the training set for the ASM (Image 4). The Cell-Shape Wizard implements this loop with the “next” and “back” buttons. This integrates the users’ expertise with cells and cellular structures. A more experienced user will be quite fast to identify the worst outliers of segmentation. By correcting those outliers, the user provides the ASM with and extended and broader training set and the ASM speed term will improve. The manual segmentation and correction of the cells in the micrographs can ideally be performed on a graphics tablet, which was found to be the most accurate and fastest input method for interactive fluorescence micrograph segmentation [18].

4

Conclusion

In this contribution we reviewed wizards and user support in fluorescence micrograph segmentation and medical image processing, and based on these findings have proposed a concept for a Cell-Shape Wizard for user guidance in ASM-supported active contour cell segmentation. The application principles and necessities for the wizard interaction patterns are well met in fluorescence micrograph cell segmentation. The need for interactive segmentation is more than justified and supported by the work of Olibarriaga and Smeulders [16]. Furthermore, an ASMsupported active contour is most likely suitable to solve segmentation problems of touching, overlapping and occluding cells in fluorescence micrograph segmentation. Additionally, the proposed wizard structure can also be used for learning-based parameter adaption of other fluorescence micrograph cell segmentation pipelines [17]. Hence, the next step will be the implementation of a CellShape Wizard prototype in to our framework and conduct a user study to evaluate the user interaction concepts integrated in the proposed concept of a Cell-Shape Wizard.

Acknowledgements This work has been supported be the DFG CRC796 ’Reprogramming of host cells by microbial effectors’, subprojects A4.



S909

References

Image 5a Mockup of the “Manual Segmentation” step

Image 5b Mockup of the “Segmentation Correction” step

[1] T. Nattkemper: “Automatic segmentation of digital micrographs: A survey,” in Proc’s 11th World Congress on Med. Informatics (MedInfo) 11(Pt.2). 2004, pp. 847–851. [2] A. Carpenter et al: “CellProfiler: image analysis software for identifying and quantifying cell phenotypes,” Genome Biology, p. 7ff : R100, 2007. [3] H. Daims et al: “Daime, a novel image analysis program for microbial ecology and biofilm research,” Environ. Microbiol.8(2), pp. 200–213, 2006. [4] A. Allalou, C. Wählby: “BlobFinder, a tool for fluorescence microscopy image cytometry,” Comp. Methods & Programs in Biomedicin 94(1), pp. 58 - 65, 2009. [5] M. Abramoff et al: “Image processing with ImageJ,” Bioph. Int 11(7), pp. 36-42, 2004. [6] C. Schneider et al: “NIH Image to ImageJ: 25 years of image analysis,” Nat Meth, 9(7), pp. 671-675, 2012. [7] B. Möller, S. Posch: “A framework unifying the development of image analysis algorithms and associated user interfaces,” in Proc. 13th IAPR Inte. Conf. Machine Vision Applications, Kyoto, Japan, 2013, pp. 447–450. [8] L. Kamentsky et al : “Improved structure, function and compatibility for CellProfiler: modular high-throughput image analysis software,” Bioinf. 27(8), pp. 1179-80, 2011. [9] K. Mühler et al: “Richtlinien für die Gestaltung chirurgischer Planungs- und Trainings-Software,” in Procs’s CURAC 2008, Leipzig, pp. 241–242, 2008. [10]J. Tidwell, Designing interfaces. O’Reilly, 2007. [11] Welie - Patterns in interaction design, “www.welie.com/patterns/showpattern.php?patternid=wizar d,” website, last accessed: 2014-03. [12] D. Franz et al: “A wizard-based segmentation approach for cochlear implant planning,” in Proc. Bildverarbeitung für die Medizin, Springer, Berlin, 2014, pp. 258-262. [13]A. Loening, G. Sanjiv: “AMIDE: A free software tool for multimodality medical image analysis,” Int J Mol Imag, pp. 131–137, 2003. [14]S. Ahmadi et al: “User friendly graphical user interface for workflow management during navigated robotic-assisted keyhole neurosurgery,” in Procs’s CARS 2010, Geneva, p. 186, 2010. [15] P. Yushkevich et al: “User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability,” NeuroImage 31(3) pp. 11161128, 2006. [16] S. Olabarriaga, A. Smeulders: “Setting the mind for intelligent interactive segmentation: Overview, requirements, and framework.” Springer, 1997, pp. 417- 422. [17] C. Held et al: “Comparison of Parameter-Adapted Segmentation Methods for Fluorescence Micrographs,” Cytometry A 79(11), pp. 933-945, 2011. [18]C. Dach et al: “Evaluation of input modalities for the inter-active image segmentation of fluorescent micrographs,” Biomed Tech 1, p. 56, 2011.

Image 5c Mockup of the “Measurements” step Image 5 Mockup of the “Manual Segmentation”, “Segmentation Results” and “Measurement” wizard page from Image 4 bottom


Miscellaneous.

Miscellaneous.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous News.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous Notes.

Miscellaneous Notes, &c.

Miscellaneous Jottings from Daily Practice.

Cough, exertional, and other miscellaneous headaches.

'On-track' and 'off-track' shoulder lesions.

Miscellaneous indications for extracorporeal photochemotherapy (ECP).

Miscellaneous adverse events with biologic agents (excludes infection and malignancy).

Miscellaneous acquired cystic lesions of the kidney and retroperitoneum.

Miscellaneous colitides: what the surgeon needs to know.

Human immunodeficiency virus as a risk factor in miscellaneous cancers.

BioC interoperability track overview.

Track 4: informatics.

Walking path or track.