Home
Search
Collections
Journals
About
Contact us
My IOPscience
Multi-channel electrical impedance tomography for regional tissue hydration monitoring
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 Physiol. Meas. 35 1137 (http://iopscience.iop.org/0967-3334/35/6/1137) View the table of contents for this issue, or go to the journal homepage for more Download details: IP Address: 207.162.240.147 This content was downloaded on 12/06/2017 at 19:26 Please note that terms and conditions apply.
You may also be interested in: The Body Composition Monitor: a flexible tool for routine fluid management across the haemodialysis population D F Keane, P Baxter, E Lindley et al. Comparison of internal multi-electrode arrays for electrical impedance tomography D Zhu, M A Barry, D M Nguyen et al. Monitoring relative fluid balance alterations in haemodialysis of diabetic patients by electrical impedance Nencho Nenchev, Feras Hatib and Ivan Daskalov Conductivitychanges due to ventilation and perfusion Jennifer L Mueller, David Isaacson and Jonathan C Newell A novel data calibration scheme for EIT Nirmal K Soni, Hamid Dehghani, Alex Hartov et al. Bioelectrical impedance spectroscopy as a fluid management system in heart failure Sören Weyer, Matthias Daniel Zink, Tobias Wartzek et al. Sensitivity study of an ultrasound coupled TREIT system for prostate imaging Y Wan, R Halter, A Borsic et al. An optimized strategy for real-time hemorrhage monitoring with EIT Canhua Xu, Meng Dai, Fusheng You et al. A 3D reconstruction algorithm for EIT using a handheld probe for breast cancer detection Tzu-Jen Kao, D Isaacson, J C Newell et al.
Institute of Physics and Engineering in Medicine Physiol. Meas. 35 (2014) 1137–1147
Physiological Measurement
doi:10.1088/0967-3334/35/6/1137
Multi-channel electrical impedance tomography for regional tissue hydration monitoring Xiaohui Chen, Tzu-Jen Kao, Jeffrey M Ashe, Gregory Boverman, James E Sabatini and David M Davenport GE Global Research Center, One Research Circle, Niskayuna, NY 12309, USA E-mail: [email protected] Received 18 December 2013, revised 24 March 2014 Accepted for publication 7 April 2014 Published 20 May 2014 Abstract
Poor assessment of hydration status during hemodialysis can lead to underor over-hydration in patients with consequences of increased morbidity and mortality. In current practice, fluid management is largely based on clinical assessments to estimate dry weight (normal hydration body weight). However, hemodialysis patients usually have co-morbidities that can make the signs of fluid status ambiguous. Therefore, achieving normal hydration status remains a major challenge for hemodialysis therapy. Electrical impedance technology has emerged as a promising method for hydration monitoring due to its noninvasive nature, low cost and ease-of-use. Conventional electrical impedancebased hydration monitoring systems employ single-channel current excitation (either 2-electrode or 4-electrode methods) to perturb and extract averaged impedance from bulk tissue and use generalized models from large populations to derive hydration estimates. In the present study, a prototype, singlefrequency electrical impedance tomography (EIT) system with simultaneous multi-channel current excitation was used to enable regional hydration change detection. We demonstrated the capability to detect a difference in daily impedance change between left leg and right leg in healthy human subjects, who wore a compression sock only on one leg to reduce daily gravitational fluid accumulation. The impedance difference corresponded well with the difference of lower leg volume change between left leg and right leg measured by volumetry, which on average is ∼35 ml, accounting for 0.7% of the lower leg volume. We have demonstrated the feasibility of using multi-channel EIT to extract hydration information in different tissue layers with minimal skin interference. Our simultaneous, multi-channel current excitation approach provides an effective method to separate electrode contact impedance and 0967-3334/14/061137+11$33.00
© 2014 Institute of Physics and Engineering in Medicine Printed in the UK
1137
X Chen et al
Physiol. Meas. 35 (2014) 1137
skin condition artifacts from hydration signals. The prototype system has the potential to be used in clinical settings for helping optimize patient fluid management during hemodialysis as well as for home monitoring of patients with congestive heart failure, chronic kidney disease, diabetes and other diseases with peripheral edema symptoms. Keywords: electrical impedance tomography, EIT, hydration monitoring, hemodialysis, multi-channel (Some figures may appear in colour only in the online journal) Introduction Hemodialysis is the most common treatment for end-stage renal disease (ESRD) patients to remove excessive water and toxins. More than 380 000 ESRD patients receive maintenance dialysis for the treatment of the disease. In 2010, Medicare expenditures for dialysis patients totaled $32 billion, or 6% of the total Medicare budget (USRDS 2012). Fluid balance is of great importance for hemodialysis therapy. In current clinical practice, fluid management is largely based on clinical assessments to estimate dry weight (normal hydration body weight). Parameters typically used to assess fluid status include the poke test, trend in body weight, blood pressure, residual renal function, neck vein observation, breathing patterns, heart size by chest x-ray and presence of intra- or post-dialysis problems. Unfortunately, some of these parameters are subjective and easily affected by factors other than hydration (Lindley et al 2011). Furthermore, hemodialysis patients usually have co-morbidities that can make the signs of fluid status ambiguous (Lindley et al 2011). Thus, achieving normal hydration status remains a major challenge for hemodialysis therapy, and under- or over-hydration easily remains undetected (Jaeger and Mehta 1999, Cridlig et al 2011, Hecking et al 2013), resulting in increased morbidity and mortality (Banerjee et al 2007, Wabel et al 2008, Wizemann et al 2009). With a projected estimate of ∼800 000 ESRD patients by 2020 (USRDS 2012), there is an increasing clinical need for new tools to more accurately assess fluid status in dialysis patients for improved hemodialysis therapy. Electrical impedance technology has emerged as a promising method for hydration monitoring due to its non-invasive nature, low cost, and ease-of-use. During the past 50 years, electrical impedance technology has evolved from global wrist-to-ankle bioimpedance analysis (BIA) for measurement of whole-body impedance (Hoffer et al 1969, Khaled et al 1988) to segmental BIA for better impedance resolution (Zhu et al 2000). Multi-frequency BIA, also called bioelectrical impedance spectroscopy (BIS), allows estimation of both intracellular water and extracellular water (ECW), especially relevant to hemodialysis patients where excess fluid removal is mainly from ECW (Chamney et al 2002, Fisch and Spiegel 1996). In recent years, body composition monitoring (BCM) using BIS technology has been tested to estimate post-hemodialysis target weight from a fluid model (Moissl et al 2013) without the need for a reference population (Machek et al 2010). However, for conventional electrical impedance spectroscopy, including BIA, BIS, and BCM, a single-channel current source is used as the stimulus signal and only bulk impedance measurements are recorded. Drawbacks associated with these approaches are: (1) bulk impedance measurement oversimplifies the inhomogeneity of human body and therefore lacks information about the spatial distribution of excess fluid; (2) impedance measurement is unavoidably skewed by variable electrode contact impedance and skin conditions; (3) patient-to-patient variability due to gender, race, body size, body composition, medication, and other existing medical conditions is not taken into consideration when using population modeling. 1138
X Chen et al
Physiol. Meas. 35 (2014) 1137
Simultaneously applying multiple current patterns to excite tissue at different layers enables depth sensitivity for hydration monitoring. Hydration information from deeper layers can benefit from less interference from skin artifact. Information of spatial distribution of fluid within tissues can potentially provide new insight into volume flux in hemodialysis patients and offer new technology to optimize fluid management. To evaluate the effect of impedance noise from the skin and differentiate impedance in different tissue layers, we employed one set of current patterns to probe only the surface layer for estimating skin noise; we employed a second set of current patterns to penetrate into deeper tissue layers and extract hydration information with minimal skin interference. We developed a novel reconstruction algorithm to remove signal drift due to electrode contact impedance and skin conditions. Instead of using a generalized model as reference, we established a baseline in individual subjects and used their own data as reference. In this study, we demonstrated the capability to extract hydration information from different layers of the tissue using a single-frequency, multi-channel EIT system. Hydration changes in the subcutaneous soft-tissue layer are isolated from other tissue layers, eliminating confounding signals from these other tissue layers (such as skin, muscle and bone). This approach yields a method of hydration assessment particularly relevant to hydration monitoring in hemodialysis patients. Methods Prototype hardware and software development
We have developed an experimental EIT system. The prototype General Electric non-invasive electrical impedance spectroscopy imaging system (GENESIS) is a parallel drive EIT instrument employing modified-Howland current sources compensated with negative impedance converters (NIC). The parallel drive architecture allows for the coherent excitation of up to 32 simultaneous current sources and measurement from 32 simultaneous voltmeters. The Howland-NIC topology simultaneously provides high sensitivity and high dynamic range over roughly a 30% fractional bandwidth. The system is designed for stability when operating over a range of biological loads and is suitable for clinical use. The prototype system was configured for eight simultaneous channels operating at a carrier frequency of 10 kHz for this study. We employed a linear reconstruction algorithm to extract impedance changes due to fluid changes as summarized below. Similar algorithms have been described in previously published papers (Mueller et al 1999, Kao et al 2006). (1) At two time instants, apply designed current patterns Ilk and measure the corresponding voltages Vlk on electrodes labeled with the subscript l = 1, . . . , L for k = 1, . . . , L − 1. Use the notation δσ = σ − σ0 and the approximation U = U0 + O(δσ ) to obtain the equations relating measured current voltages to moments of the unknown conductivity, L D(k, x) = Vlk (σ0 )Ilx − Vlx (σ )Ilk = δσ ∇U0k · ∇U0x dp + O(δσ 2 ) B
l=1
where D(k, x) is the ‘data’ matrix used to reconstruct the change in conductivity between the two time points. (2) Choose a basis function {n (p)}Nn=1 where n is the number of voxels in the predefined N reconstruction mesh for the approximation δσ (p) = n=1 δσn n (p) and compute a coefficient matrix A, Ak,x,n = Vn
n (p)∇U0k · ∇U0x dp
where Vn is the spatial extent of voxel n. 1139
X Chen et al
Physiol. Meas. 35 (2014) 1137
(3) Solve the equation δσ = (AT A + εR)−1 × AT D and display δσ on the mesh. Here, R denotes the regularization matrix and ε denotes the regularization parameter. (4) Apply the next current pattern set and measure the resulting voltage for next time frame and repeat the process. We have observed that the largest component of the changes in the voltages between two sets of measurements is due to contact impedances rather than internal processes. Therefore, we remove the changes in voltages due to this confounding factor prior to the image reconstruction. Specifically, we compute the Jacobian matrix with respect to contact impedance changes in a manner analogous to that described in the previously published paper (Boverman et al 2009). (1) Compute the Jacobian matrix, Jc , with respect to complex changes in contact impedance. (2) Estimate δzc , the vector of complex changes in contact impedances, as follows: −1 δzc = JcH Jc JcH (yd − yreft ) where yd is the differential measurement and yref is the reference measurement. (3) Compute changes in voltages due to contact impedances: yc = Jc δzc
Electrode array development for chicken breast studies
A specially designed electrode array was developed for chicken breast. The array was modeled and optimized using COMSOL. Cadence/OrCAD PCB Designer Professional was used to create the Gerber and Excellon NC manufacturing design files. Standard printed circuit board fabrication processes were used to build two electrode breakout test boards (Cleveland Circuits Corp DBA Instrumatics, Cleveland, OH, USA). The boards were fabricated using G10 fiberglass 1.575 mm thick with two sided 1 oz. copper clad printed circuit materials with plated through holes. Electroless nickel immersion gold was used for the final metal plating of the exposed electrodes and breakout post pads. Green solder mask was applied to protect the remaining circuit traces from electrical leakage due to conductive samples and/or biological contaminants. Two 8-electrode linear arrays were sized and spaced at pitches of 8 mm (4 mm pad and 4 mm space).The breakout interconnections from the electrode pads to the test posts were connected with 1.27 mm traces at 8 mm minimum pitch. The final board size was cut to 177.8 × 177.8 mm. Gold plated test connection posts were hand soldered to the breakout pads to complete the assembly. The GENESIS instrument was connected to the test connection posts using driven-shield coaxial cables terminated with alligator clips. A separate reference potential electrode was connected using an additional connection post on one edge of the PCB board. Electrode array development for watermelon and healthy human subject studies
A linear, flexible, 8-electrode array was developed in collaboration with Intelesens (Belfast, Northern Ireland, UK). Eight Ag/Ag-Cl electrodes with 1.6 cm diameter and 1 cm spacing were fabricated in a single patch containing individual snap-on button connections. The electrodes were formed by placing conductive/adhesive gel over screen printed contacts on a polyester film. The polyester film was then sandwiched between sheets of medical-grade foam with the bottom side cutouts to expose the electrode regions to the skin. 1140
X Chen et al
Physiol. Meas. 35 (2014) 1137
Watermelon experiment protocol design
We attached a linear 8-electrode array to the surface of a hollowed-out watermelon filled with saline. The electrode array was connected to the GENESIS system with snap-on button connections. We positioned a stainless steel ball (diameter = 0.55 in) with higher conductivity at known depths in the saline, injected multiple current patterns and recorded the voltages on each electrode. We reconstructed the position of the metal ball by impedance contrast and compared to the known positions for algorithm validation. Chicken breast experiment protocol designs
(1) Hydration monitoring in chicken breast at different tissue layers. We injected conductive saline mixed with food dye into different depths of a chicken breast positioned on top of a linear 8-electrode array. We first injected saline three-fold more conductive compared to the chicken breast into a location close to the surface of one chicken breast and recorded impedance change during the process. We then injected the same saline into a deeper location of another chicken breast and recorded impedance change during the process. To avoid potential anisotropic effect of impedance measurement due to directionality of muscle fibers, we made sure that we measured impedance at the same location and orientation before and after injection. Impedance estimates at different tissue layers were obtained by reconstruction and the relative conductivity values for individual tissue layers were calculated and compared. (2) Hydration monitoring in chicken breast with different skin conditions. We modeled a chicken breast to have four evenly-spaced tissue layers. As the control, we measured and reconstructed the averaged conductivity of each layer of a fresh chicken breast by putting the chicken breast in contact with an 8-electrode array on a PCB board. Then we wiped dry the surface of the chicken breast in contact with the electrodes and repeated the measurement. Afterwards, we sprayed saline once to the same chicken breast surface and repeated the measurement, followed by spraying saline again to the surface and measuring again. To avoid potential anisotropic effect of impedance measurement due to muscle fiber directionality, we carefully lifted chicken breast for saline spray and put it back to the same location for impedance measurement. Averaged conductivity of each layer under these skin conditions was calculated and the difference of the relative conductivity change compared to the control in each layer was calculated and compared. Healthy human subject experiment protocol design
The human subject study was approved by the Institutional Review Board of Ethical & Independent Review Services (Corte Madera, CA, USA). Five healthy human subject volunteers were recruited from GE Global Research Center for the study. Subjects reported to the lab at the beginning of their workday where we placed 8-electrode linear arrays on the outside calf of each leg. The electrode arrays were connected to the prototype EIT system and excitation frames, each containing seven current patterns, were sequentially applied to each leg. Each current pattern was designed to deliver less than 1 mA of current at a carrier frequency of 10 kHz. Impedance measurements were recorded for each leg at a sampling rate of 20 frames per second for a period of 10 min. After the impedance measurements, the electrode arrays were removed and the volume of each leg was measured by weighing the amount of water displaced from an edema gauge (model 12–3508 volumeter, Fabrication Enterprises Incorporated, Elmsford, NY, USA) using a laboratory balance (MS3002S, Mettler Toledo, Columbus, OH, USA). The subjects were asked to return to their normal daily work 1141
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
(c)
(d)
(e)
(f)
Figure 1. Validation of reconstruction algorithm in watermelon saline tank. (a) Placement of linear 8-electrode array on an empty watermelon skin; (b) the empty watermelon was filled with saline and a steel ball target was placed near the bottom of the watermelon and moved upward toward saline surface at three different depths while taking impedance measurement. Reconstructed target tomography shows (c) no target, (d) target at 1 cm to the bottom, (e) target at 2 cm to the bottom, and (f) target at 3 cm to the bottom.
activities while wearing one Maxar compression sock (ITA-MED Co., San Francisco, USA) with medium compression of 18–20 mmHg on their left leg. The subjects were asked to return to the lab at the end of their workday where the compression sock was removed and measurements of impedance and volume on both calves were repeated following the same protocol as in the morning. Results and discussion System validation
The prototype system and reconstruction algorithm were validated in a phantom study using a hollowed-out watermelon and a steel ball. A linear 8-electrode array was attached to the surface of the watermelon and connected to the EIT system with snap-on connections (figure 1(a)). The steel ball with higher conductivity was positioned at known depths in the saline (figure 1(b)). The reconstructed tomographic images in figures 1(c)–(f) confirm that the EIT system can detect the steel ball at correct positions. While adequate for validating the prototype system and algorithms, the steel ball represents a discrete and high impedance contrast target which is different from the continuous and smaller impedance changes produced by hydration changes in human subjects. To further confirm that the system can detect hydration changes in biological tissues, we produced hydration changes in both chicken breasts and human subjects in the following experiments. Detection of layer-based hydration change in chicken breasts
We artificially changed the hydration level of a chicken breast by injecting saline at different tissue depths. Surface injection is used to change hydration level at near-skin level and deeper tissue injection is used to change hydration level at depth. We reconstructed electrical conductivity of the chicken breast in four horizontal tissue layers using impedance tomography. Averaged relative conductivity was calculated and compared for each layer. For surface tissue injection as shown in figure 2(a), 5 ml of saline (1.6% of the total tissue weight), with three-fold conductivity compared to the chicken breast, was injected into a location close to the surface of a 310-gram chicken breast. Our reconstruction algorithm detected the largest relative conductivity change of 0.33 in the surface layer (blue dye colored portion in the picture inset of figure 2(a)). Averaged relative conductivity for layer 1 has the most significant change while adjacent layer 2 demonstrates relative smaller conductivity change; whereas for 1142
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
Figure 2. The prototype EIT system can detect hydration change at different tissue layers in chicken breast. (a) Surface tissue layer injection. (b) Deeper tissue layer injection. Note: layer 1 (0–1 cm), layer 2 (1–2 cm), layer 3 (2–3 cm), layer 4 (3–4 cm) in depth.
layer 3 and layer 4, there is minimal conductivity change during the injection. For deeper tissue injection as shown in figure 2(b), 10 ml of the same saline (3.3% of the total tissue weight) was injected into the deeper layer of a 305-gram chicken breast. Our reconstruction algorithm detected the greatest relative conductivity change of 0.20 in layer 2 (red dye colored portion in the picture inset figure 2(b)) corresponding to the injection location. These results confirm that the multi-channel excitation EIT system can detect hydration change at different tissue layers. The benefits of this layer-based hydration detection are better regional resolution and reduced skin artifacts. It is worth noting that the conductivity change for layer 1 and layer 2 crossed over at ∼3 ml injection in surface injection experiment. A possible explanation for this cross-over was that layer 1 had become saturated and continued saline injection spread into the second layer. Also it is difficult to inject saline at precise depths and control the spreading of saline in the specific layer, which can induce cross-talk between layers as well. Reduced skin artifacts in chicken breasts
We applied a set of current patterns to penetrate only the surface layer underneath the skin to evaluate the effect of impedance noise from the skin. At the same time, we applied another set of current patterns that penetrate deeper into the skin to extract information from deeper tissue layers with minimal interference from the skin. We modeled a chicken breast to have four tissue layers as specified in the caption to figure 2. We first measured and reconstructed the averaged impedance of each layer of a fresh chicken breast as the control. Then we artificially changed the skin condition of the chicken breast to make it dry, wet and wetter. Averaged impedance of each layer under these skin conditions was reconstructed and the difference of the impedance change compared to the control in four layers of the chicken breast was calculated. Reconstructed impedance tomography in figure 3(a) shows that the majority of 1143
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
Figure 3. Multi-channel excitation helps to reduce interference from different skin conditions. (a) Reconstructed tomography and relative conductivity change in chicken breast under different skin conditions, fresh skin condition as the control; (i) the surface in contact with the 8-electrode array was wiped dry before taking impedance measurement; (ii) the same surface was sprayed once with conductive saline before making impedance measurement; (iii) the same surface was sprayed again with conductive saline before making impedance measurement; (b) summary of relative conductivity change in four tissue layers within the chicken breast under three skin conditions.
conductivity changes occurred at the skin layer (layer 1) whereas much smaller changes were detected at deeper tissue layers 2 to 4. Averaged conductivity change within different layers for three skin conditions are compared in figure 3(b). Reconstructed conductivity change within layer 1 due to skin condition change is two to five times larger than layer 2. Minimal conductivity change was observed for deeper tissue layers 3 and 4. This data confirms that multi-channel excitation provides a method to reduce skin artifacts. Future work is needed to optimize the multi-channel current pattern and reconstruction algorithm for more refined separation of deeper hydration signals from skin artifacts. Ideally, it is desired to have no impedance change in layer 2 to 4 when there is skin condition change in layer 1. As you can see from the reconstructed impedance tomography, there is cross-talk between layer 1 and layer 2. This is an intrinsic spatial resolution limitation of EIT/EIS. More future work would be needed to focus on eliminating cross-talk between layers by optimizing reconstruction algorithm, electrode array design and prior calibration of spatial resolution. 1144
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
(c) Figure 4. Compression sock experiment in healthy human subjects. (a) Left: healthy human subject wears compression sock only on left leg to reduce daily accumulation of fluid in lower limb; Middle: placement of a linear 8-electrode array for impedance data acquisition; Right: volumeter to measure leg volume by water displacement; (b) daily volume change percentage in right leg (control, blue bar) and left leg (compression sock, red bar) in five healthy human subjects, there is an average of 35 ± 35.9 ml fluid reduction in lower left leg with compression sock compared to the lower right leg (control, without compression sock); (c) average daily conductivity difference between right and left leg in five tissue layers for five healthy human subjects (Conductivity change in subcutaneous tissue layers 2 and 3 is statistically significant compared to change in skin layer 1. Results for t-test (paired two sample for means): ∗ p = 0.02 for comparing layer 1 with layer 2, p = 0.03 for comparing layer 1 with layer 3.)
Detection of layer-based hydration change in human subjects
We conducted a study to induce in vivo fluid changes in five healthy human subjects under IRB approval. This study manipulated the normal gravitational fluid accumulation that occurs in lower limbs from morning to afternoon. The study was designed such that fluid accumulation was restricted in the left leg by use of a compression sock. Baseline measurements of impedance and volume on both calves were taken upon presentation to the lab. Subjects then wore one compression sock on left leg while conducting normal daily activities (figure 4(a)). After 1145
X Chen et al
Physiol. Meas. 35 (2014) 1137
wearing the sock for 6 h, the sock was removed, and impedance and volume on both calves were measured again. The volumetric data in figure 4(b) demonstrates a mean volume difference of 35 ml (0.7% of gauge-measured leg volume) between the lower right leg volume change (no compression sock) and the lower left leg volume change (with compression sock) from morning to evening (Out of five healthy human subjects, subject P91 demonstrated a different volume change pattern than other human subjects, with slightly more fluid accumulation in the left leg with compression sock. One possible explanation might be that the pressure of compression sock for this subject is not strong enough to restrict fluid accumulation). Using our reconstruction algorithm with electrode drift compensation, we reconstructed the mean conductivity difference of right leg conductivity change and left leg conductivity change from morning to evening at five tissue layers in five subjects. The volume change in each subject shown in figure 4(b) confirm our hypothesis that there is more fluid accumulation in the right leg than in the left leg, indicating restriction effect of compression sock for healthy human subjects. Average relative conductivity change in each layer (figure 4(c))demonstrates that conductivity changes in subcutaneous tissue layers 2 and 3 are statistically different than that in skin layer 1. It is observed that there is more conductivity variability in the skin layer 1 indicating an intrinsic impedance distortion on the skin due to skin contact impedance and skin conditions. By dividing the calf into different tissue layers and enabling tomographic spatial reconstruction of impedance measurements with multi-channel EIT, we are able to differentiate skin artifacts from hydration signals at subcutaneous tissue layers. Another observation is that conductivity changes in layer 4 and 5 are small, which may result from smaller signal or insufficient sensitivity at deeper layers. Since excess fluid tends to accumulate in subcutaneous layers, we are more interested in monitoring subcutaneous layers. Nevertheless, it is important to address the issue of deeper layers in future work as well. Summary
We have demonstrated the feasibility of using multi-channel EIT to extract hydration information in different tissue layers with minimal skin interference. Our simultaneous, multichannel current excitation approach provides an effective method to separate electrode contact impedance and skin condition artifacts from hydration signals. Further studies are underway to integrate multi-frequency capability into the prototype system to extract extracellular and intracellular fluid information and to validate the system in a clinical setting with hemodialysis patients. Acknowledgment The authors would like to thank Dr Allen Garner for PCB board design simulation and Ms Maxine Gibeau for PCB board design assistance. References Banerjee D, Ma J Z, Collins A J and Herzog C A 2007 Long-term survival of incident hemodialysis patients who are hospitalized for congestive heart failure, pulmonary edema, or fluid overload Clin. J. Am. Soc. Nephrol. 2 1186–90 Boverman G, Isaacson D, Saulnier G J and Newell J C 2009 Methods for compensating for variable electrode contact in EIT IEEE Trans. Biomed. Eng. 56 2762–72 Chamney P W et al 2002 A new technique for establishing dry weight in hemodialysis patients via whole body bioimpedance Kidney Int. 61 2250–8 1146
X Chen et al
Physiol. Meas. 35 (2014) 1137
Cridlig J et al 2011 Formulation of a dry weight bioimpedance index in hemodialysis patients Int. J. Artif. Organs 34 1075–84 Fisch B J and Spiegel D M 1996 Assessment of excess fluid distribution in chronic hemodialysis patients using bioimpedance spectroscopy Kidney Int. 49 1105–9 Hecking M et al 2013 Significance of interdialytic weight gain versus chronic volume overload: consensus opinion Am. J. Nephrol. 38 78–90 Hoffer E C, Meador C K and Simpson D C 1969 Correlation of whole body impedance with total body water volume J. Appl. Physiol. 27 531–4 Jaeger J Q and Mehta R L 1999 Assessment of dry weight in hemodialysis: an overview J. Am. Soc. Nephrol. 10 392–403 Kao T J, Isaacson D, Newell J C and Saulnier G J 2006 A 3D reconstruction algorithm for EIT using a handheld probe for breast cancer detection Physiol. Meas. 27 S1–11 Khaled A M, McCutcheon M J, Reddy S, Pearman P L, Hunter G R and Weinsier R L 1988 Electrical impedance in assessing human body composition: the BIA method Am. J. Clin. Nutr. 47 789–92 Lindley E, Lynne A, Claire G and Garthwaite E 2011 Management of fluid status in haemodialysis patients: the roles of technology and dietary advice Technical Problems in Patients on Hemodialysis ed M G Penido (Rijeka: InTech) (www.intechopen.com/books/technical-problemsin-patients-on-hemodialysis/management-of-fluid-status-in-haemodialysis-patients-the-roles-oftechnology-and-dietary-advice) Machek P et al 2010 Guided optimization of fluid status in haemodialysis patients Nephrol. Dial. Transplant. 25 538–44 Moissl U, Arias-Guill´en M, Wabel P, Fontser´e N, Carrera M, Campistol J M and Maduell F 2013 Bioimpedance-guided fluid management in hemodialysis patients Clin. J. Am. Soc. Nephrol. 8 1575–82 Mueller J L, Isaacson D and Newell J C 1999 A reconstruction algorithm for electrical impedance tomography data collected on rectangular electrode array IEEE Trans. Biomed. Eng. 46 1379–86 US Renal Data System (USRDS) 2012 Annual Data Report: Atlas of Chronic Kidney Disease and EndStage Renal Disease in the United States (Bethesda, MD: National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases) Wabel P et al 2008 Towards improved cardiovascular management: the necessity of combining blood pressure and fluid overload Nephrol. Dial. Transplant. 23 2965–71 Wizemann V, Wabel P, Chamney P, Zaluska W, Moissl U, Rode C, Malecka-Masalska T and Marcelli D 2009 The mortality risk of overhydration in haemodialysis patients Nephrol. Dial. Transplant. 24 1574–9 Zhu F, Schneditz D, Kaufman A M and Levin N W 2000 Estimation of body fluid changes during peritoneal dialysis by segmental bioimpedance analysis Kidney Int. 57 299–306
1147
Search
Collections
Journals
About
Contact us
My IOPscience
Multi-channel electrical impedance tomography for regional tissue hydration monitoring
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 Physiol. Meas. 35 1137 (http://iopscience.iop.org/0967-3334/35/6/1137) View the table of contents for this issue, or go to the journal homepage for more Download details: IP Address: 207.162.240.147 This content was downloaded on 12/06/2017 at 19:26 Please note that terms and conditions apply.
You may also be interested in: The Body Composition Monitor: a flexible tool for routine fluid management across the haemodialysis population D F Keane, P Baxter, E Lindley et al. Comparison of internal multi-electrode arrays for electrical impedance tomography D Zhu, M A Barry, D M Nguyen et al. Monitoring relative fluid balance alterations in haemodialysis of diabetic patients by electrical impedance Nencho Nenchev, Feras Hatib and Ivan Daskalov Conductivitychanges due to ventilation and perfusion Jennifer L Mueller, David Isaacson and Jonathan C Newell A novel data calibration scheme for EIT Nirmal K Soni, Hamid Dehghani, Alex Hartov et al. Bioelectrical impedance spectroscopy as a fluid management system in heart failure Sören Weyer, Matthias Daniel Zink, Tobias Wartzek et al. Sensitivity study of an ultrasound coupled TREIT system for prostate imaging Y Wan, R Halter, A Borsic et al. An optimized strategy for real-time hemorrhage monitoring with EIT Canhua Xu, Meng Dai, Fusheng You et al. A 3D reconstruction algorithm for EIT using a handheld probe for breast cancer detection Tzu-Jen Kao, D Isaacson, J C Newell et al.
Institute of Physics and Engineering in Medicine Physiol. Meas. 35 (2014) 1137–1147
Physiological Measurement
doi:10.1088/0967-3334/35/6/1137
Multi-channel electrical impedance tomography for regional tissue hydration monitoring Xiaohui Chen, Tzu-Jen Kao, Jeffrey M Ashe, Gregory Boverman, James E Sabatini and David M Davenport GE Global Research Center, One Research Circle, Niskayuna, NY 12309, USA E-mail: [email protected] Received 18 December 2013, revised 24 March 2014 Accepted for publication 7 April 2014 Published 20 May 2014 Abstract
Poor assessment of hydration status during hemodialysis can lead to underor over-hydration in patients with consequences of increased morbidity and mortality. In current practice, fluid management is largely based on clinical assessments to estimate dry weight (normal hydration body weight). However, hemodialysis patients usually have co-morbidities that can make the signs of fluid status ambiguous. Therefore, achieving normal hydration status remains a major challenge for hemodialysis therapy. Electrical impedance technology has emerged as a promising method for hydration monitoring due to its noninvasive nature, low cost and ease-of-use. Conventional electrical impedancebased hydration monitoring systems employ single-channel current excitation (either 2-electrode or 4-electrode methods) to perturb and extract averaged impedance from bulk tissue and use generalized models from large populations to derive hydration estimates. In the present study, a prototype, singlefrequency electrical impedance tomography (EIT) system with simultaneous multi-channel current excitation was used to enable regional hydration change detection. We demonstrated the capability to detect a difference in daily impedance change between left leg and right leg in healthy human subjects, who wore a compression sock only on one leg to reduce daily gravitational fluid accumulation. The impedance difference corresponded well with the difference of lower leg volume change between left leg and right leg measured by volumetry, which on average is ∼35 ml, accounting for 0.7% of the lower leg volume. We have demonstrated the feasibility of using multi-channel EIT to extract hydration information in different tissue layers with minimal skin interference. Our simultaneous, multi-channel current excitation approach provides an effective method to separate electrode contact impedance and 0967-3334/14/061137+11$33.00
© 2014 Institute of Physics and Engineering in Medicine Printed in the UK
1137
X Chen et al
Physiol. Meas. 35 (2014) 1137
skin condition artifacts from hydration signals. The prototype system has the potential to be used in clinical settings for helping optimize patient fluid management during hemodialysis as well as for home monitoring of patients with congestive heart failure, chronic kidney disease, diabetes and other diseases with peripheral edema symptoms. Keywords: electrical impedance tomography, EIT, hydration monitoring, hemodialysis, multi-channel (Some figures may appear in colour only in the online journal) Introduction Hemodialysis is the most common treatment for end-stage renal disease (ESRD) patients to remove excessive water and toxins. More than 380 000 ESRD patients receive maintenance dialysis for the treatment of the disease. In 2010, Medicare expenditures for dialysis patients totaled $32 billion, or 6% of the total Medicare budget (USRDS 2012). Fluid balance is of great importance for hemodialysis therapy. In current clinical practice, fluid management is largely based on clinical assessments to estimate dry weight (normal hydration body weight). Parameters typically used to assess fluid status include the poke test, trend in body weight, blood pressure, residual renal function, neck vein observation, breathing patterns, heart size by chest x-ray and presence of intra- or post-dialysis problems. Unfortunately, some of these parameters are subjective and easily affected by factors other than hydration (Lindley et al 2011). Furthermore, hemodialysis patients usually have co-morbidities that can make the signs of fluid status ambiguous (Lindley et al 2011). Thus, achieving normal hydration status remains a major challenge for hemodialysis therapy, and under- or over-hydration easily remains undetected (Jaeger and Mehta 1999, Cridlig et al 2011, Hecking et al 2013), resulting in increased morbidity and mortality (Banerjee et al 2007, Wabel et al 2008, Wizemann et al 2009). With a projected estimate of ∼800 000 ESRD patients by 2020 (USRDS 2012), there is an increasing clinical need for new tools to more accurately assess fluid status in dialysis patients for improved hemodialysis therapy. Electrical impedance technology has emerged as a promising method for hydration monitoring due to its non-invasive nature, low cost, and ease-of-use. During the past 50 years, electrical impedance technology has evolved from global wrist-to-ankle bioimpedance analysis (BIA) for measurement of whole-body impedance (Hoffer et al 1969, Khaled et al 1988) to segmental BIA for better impedance resolution (Zhu et al 2000). Multi-frequency BIA, also called bioelectrical impedance spectroscopy (BIS), allows estimation of both intracellular water and extracellular water (ECW), especially relevant to hemodialysis patients where excess fluid removal is mainly from ECW (Chamney et al 2002, Fisch and Spiegel 1996). In recent years, body composition monitoring (BCM) using BIS technology has been tested to estimate post-hemodialysis target weight from a fluid model (Moissl et al 2013) without the need for a reference population (Machek et al 2010). However, for conventional electrical impedance spectroscopy, including BIA, BIS, and BCM, a single-channel current source is used as the stimulus signal and only bulk impedance measurements are recorded. Drawbacks associated with these approaches are: (1) bulk impedance measurement oversimplifies the inhomogeneity of human body and therefore lacks information about the spatial distribution of excess fluid; (2) impedance measurement is unavoidably skewed by variable electrode contact impedance and skin conditions; (3) patient-to-patient variability due to gender, race, body size, body composition, medication, and other existing medical conditions is not taken into consideration when using population modeling. 1138
X Chen et al
Physiol. Meas. 35 (2014) 1137
Simultaneously applying multiple current patterns to excite tissue at different layers enables depth sensitivity for hydration monitoring. Hydration information from deeper layers can benefit from less interference from skin artifact. Information of spatial distribution of fluid within tissues can potentially provide new insight into volume flux in hemodialysis patients and offer new technology to optimize fluid management. To evaluate the effect of impedance noise from the skin and differentiate impedance in different tissue layers, we employed one set of current patterns to probe only the surface layer for estimating skin noise; we employed a second set of current patterns to penetrate into deeper tissue layers and extract hydration information with minimal skin interference. We developed a novel reconstruction algorithm to remove signal drift due to electrode contact impedance and skin conditions. Instead of using a generalized model as reference, we established a baseline in individual subjects and used their own data as reference. In this study, we demonstrated the capability to extract hydration information from different layers of the tissue using a single-frequency, multi-channel EIT system. Hydration changes in the subcutaneous soft-tissue layer are isolated from other tissue layers, eliminating confounding signals from these other tissue layers (such as skin, muscle and bone). This approach yields a method of hydration assessment particularly relevant to hydration monitoring in hemodialysis patients. Methods Prototype hardware and software development
We have developed an experimental EIT system. The prototype General Electric non-invasive electrical impedance spectroscopy imaging system (GENESIS) is a parallel drive EIT instrument employing modified-Howland current sources compensated with negative impedance converters (NIC). The parallel drive architecture allows for the coherent excitation of up to 32 simultaneous current sources and measurement from 32 simultaneous voltmeters. The Howland-NIC topology simultaneously provides high sensitivity and high dynamic range over roughly a 30% fractional bandwidth. The system is designed for stability when operating over a range of biological loads and is suitable for clinical use. The prototype system was configured for eight simultaneous channels operating at a carrier frequency of 10 kHz for this study. We employed a linear reconstruction algorithm to extract impedance changes due to fluid changes as summarized below. Similar algorithms have been described in previously published papers (Mueller et al 1999, Kao et al 2006). (1) At two time instants, apply designed current patterns Ilk and measure the corresponding voltages Vlk on electrodes labeled with the subscript l = 1, . . . , L for k = 1, . . . , L − 1. Use the notation δσ = σ − σ0 and the approximation U = U0 + O(δσ ) to obtain the equations relating measured current voltages to moments of the unknown conductivity, L D(k, x) = Vlk (σ0 )Ilx − Vlx (σ )Ilk = δσ ∇U0k · ∇U0x dp + O(δσ 2 ) B
l=1
where D(k, x) is the ‘data’ matrix used to reconstruct the change in conductivity between the two time points. (2) Choose a basis function {n (p)}Nn=1 where n is the number of voxels in the predefined N reconstruction mesh for the approximation δσ (p) = n=1 δσn n (p) and compute a coefficient matrix A, Ak,x,n = Vn
n (p)∇U0k · ∇U0x dp
where Vn is the spatial extent of voxel n. 1139
X Chen et al
Physiol. Meas. 35 (2014) 1137
(3) Solve the equation δσ = (AT A + εR)−1 × AT D and display δσ on the mesh. Here, R denotes the regularization matrix and ε denotes the regularization parameter. (4) Apply the next current pattern set and measure the resulting voltage for next time frame and repeat the process. We have observed that the largest component of the changes in the voltages between two sets of measurements is due to contact impedances rather than internal processes. Therefore, we remove the changes in voltages due to this confounding factor prior to the image reconstruction. Specifically, we compute the Jacobian matrix with respect to contact impedance changes in a manner analogous to that described in the previously published paper (Boverman et al 2009). (1) Compute the Jacobian matrix, Jc , with respect to complex changes in contact impedance. (2) Estimate δzc , the vector of complex changes in contact impedances, as follows: −1 δzc = JcH Jc JcH (yd − yreft ) where yd is the differential measurement and yref is the reference measurement. (3) Compute changes in voltages due to contact impedances: yc = Jc δzc
Electrode array development for chicken breast studies
A specially designed electrode array was developed for chicken breast. The array was modeled and optimized using COMSOL. Cadence/OrCAD PCB Designer Professional was used to create the Gerber and Excellon NC manufacturing design files. Standard printed circuit board fabrication processes were used to build two electrode breakout test boards (Cleveland Circuits Corp DBA Instrumatics, Cleveland, OH, USA). The boards were fabricated using G10 fiberglass 1.575 mm thick with two sided 1 oz. copper clad printed circuit materials with plated through holes. Electroless nickel immersion gold was used for the final metal plating of the exposed electrodes and breakout post pads. Green solder mask was applied to protect the remaining circuit traces from electrical leakage due to conductive samples and/or biological contaminants. Two 8-electrode linear arrays were sized and spaced at pitches of 8 mm (4 mm pad and 4 mm space).The breakout interconnections from the electrode pads to the test posts were connected with 1.27 mm traces at 8 mm minimum pitch. The final board size was cut to 177.8 × 177.8 mm. Gold plated test connection posts were hand soldered to the breakout pads to complete the assembly. The GENESIS instrument was connected to the test connection posts using driven-shield coaxial cables terminated with alligator clips. A separate reference potential electrode was connected using an additional connection post on one edge of the PCB board. Electrode array development for watermelon and healthy human subject studies
A linear, flexible, 8-electrode array was developed in collaboration with Intelesens (Belfast, Northern Ireland, UK). Eight Ag/Ag-Cl electrodes with 1.6 cm diameter and 1 cm spacing were fabricated in a single patch containing individual snap-on button connections. The electrodes were formed by placing conductive/adhesive gel over screen printed contacts on a polyester film. The polyester film was then sandwiched between sheets of medical-grade foam with the bottom side cutouts to expose the electrode regions to the skin. 1140
X Chen et al
Physiol. Meas. 35 (2014) 1137
Watermelon experiment protocol design
We attached a linear 8-electrode array to the surface of a hollowed-out watermelon filled with saline. The electrode array was connected to the GENESIS system with snap-on button connections. We positioned a stainless steel ball (diameter = 0.55 in) with higher conductivity at known depths in the saline, injected multiple current patterns and recorded the voltages on each electrode. We reconstructed the position of the metal ball by impedance contrast and compared to the known positions for algorithm validation. Chicken breast experiment protocol designs
(1) Hydration monitoring in chicken breast at different tissue layers. We injected conductive saline mixed with food dye into different depths of a chicken breast positioned on top of a linear 8-electrode array. We first injected saline three-fold more conductive compared to the chicken breast into a location close to the surface of one chicken breast and recorded impedance change during the process. We then injected the same saline into a deeper location of another chicken breast and recorded impedance change during the process. To avoid potential anisotropic effect of impedance measurement due to directionality of muscle fibers, we made sure that we measured impedance at the same location and orientation before and after injection. Impedance estimates at different tissue layers were obtained by reconstruction and the relative conductivity values for individual tissue layers were calculated and compared. (2) Hydration monitoring in chicken breast with different skin conditions. We modeled a chicken breast to have four evenly-spaced tissue layers. As the control, we measured and reconstructed the averaged conductivity of each layer of a fresh chicken breast by putting the chicken breast in contact with an 8-electrode array on a PCB board. Then we wiped dry the surface of the chicken breast in contact with the electrodes and repeated the measurement. Afterwards, we sprayed saline once to the same chicken breast surface and repeated the measurement, followed by spraying saline again to the surface and measuring again. To avoid potential anisotropic effect of impedance measurement due to muscle fiber directionality, we carefully lifted chicken breast for saline spray and put it back to the same location for impedance measurement. Averaged conductivity of each layer under these skin conditions was calculated and the difference of the relative conductivity change compared to the control in each layer was calculated and compared. Healthy human subject experiment protocol design
The human subject study was approved by the Institutional Review Board of Ethical & Independent Review Services (Corte Madera, CA, USA). Five healthy human subject volunteers were recruited from GE Global Research Center for the study. Subjects reported to the lab at the beginning of their workday where we placed 8-electrode linear arrays on the outside calf of each leg. The electrode arrays were connected to the prototype EIT system and excitation frames, each containing seven current patterns, were sequentially applied to each leg. Each current pattern was designed to deliver less than 1 mA of current at a carrier frequency of 10 kHz. Impedance measurements were recorded for each leg at a sampling rate of 20 frames per second for a period of 10 min. After the impedance measurements, the electrode arrays were removed and the volume of each leg was measured by weighing the amount of water displaced from an edema gauge (model 12–3508 volumeter, Fabrication Enterprises Incorporated, Elmsford, NY, USA) using a laboratory balance (MS3002S, Mettler Toledo, Columbus, OH, USA). The subjects were asked to return to their normal daily work 1141
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
(c)
(d)
(e)
(f)
Figure 1. Validation of reconstruction algorithm in watermelon saline tank. (a) Placement of linear 8-electrode array on an empty watermelon skin; (b) the empty watermelon was filled with saline and a steel ball target was placed near the bottom of the watermelon and moved upward toward saline surface at three different depths while taking impedance measurement. Reconstructed target tomography shows (c) no target, (d) target at 1 cm to the bottom, (e) target at 2 cm to the bottom, and (f) target at 3 cm to the bottom.
activities while wearing one Maxar compression sock (ITA-MED Co., San Francisco, USA) with medium compression of 18–20 mmHg on their left leg. The subjects were asked to return to the lab at the end of their workday where the compression sock was removed and measurements of impedance and volume on both calves were repeated following the same protocol as in the morning. Results and discussion System validation
The prototype system and reconstruction algorithm were validated in a phantom study using a hollowed-out watermelon and a steel ball. A linear 8-electrode array was attached to the surface of the watermelon and connected to the EIT system with snap-on connections (figure 1(a)). The steel ball with higher conductivity was positioned at known depths in the saline (figure 1(b)). The reconstructed tomographic images in figures 1(c)–(f) confirm that the EIT system can detect the steel ball at correct positions. While adequate for validating the prototype system and algorithms, the steel ball represents a discrete and high impedance contrast target which is different from the continuous and smaller impedance changes produced by hydration changes in human subjects. To further confirm that the system can detect hydration changes in biological tissues, we produced hydration changes in both chicken breasts and human subjects in the following experiments. Detection of layer-based hydration change in chicken breasts
We artificially changed the hydration level of a chicken breast by injecting saline at different tissue depths. Surface injection is used to change hydration level at near-skin level and deeper tissue injection is used to change hydration level at depth. We reconstructed electrical conductivity of the chicken breast in four horizontal tissue layers using impedance tomography. Averaged relative conductivity was calculated and compared for each layer. For surface tissue injection as shown in figure 2(a), 5 ml of saline (1.6% of the total tissue weight), with three-fold conductivity compared to the chicken breast, was injected into a location close to the surface of a 310-gram chicken breast. Our reconstruction algorithm detected the largest relative conductivity change of 0.33 in the surface layer (blue dye colored portion in the picture inset of figure 2(a)). Averaged relative conductivity for layer 1 has the most significant change while adjacent layer 2 demonstrates relative smaller conductivity change; whereas for 1142
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
Figure 2. The prototype EIT system can detect hydration change at different tissue layers in chicken breast. (a) Surface tissue layer injection. (b) Deeper tissue layer injection. Note: layer 1 (0–1 cm), layer 2 (1–2 cm), layer 3 (2–3 cm), layer 4 (3–4 cm) in depth.
layer 3 and layer 4, there is minimal conductivity change during the injection. For deeper tissue injection as shown in figure 2(b), 10 ml of the same saline (3.3% of the total tissue weight) was injected into the deeper layer of a 305-gram chicken breast. Our reconstruction algorithm detected the greatest relative conductivity change of 0.20 in layer 2 (red dye colored portion in the picture inset figure 2(b)) corresponding to the injection location. These results confirm that the multi-channel excitation EIT system can detect hydration change at different tissue layers. The benefits of this layer-based hydration detection are better regional resolution and reduced skin artifacts. It is worth noting that the conductivity change for layer 1 and layer 2 crossed over at ∼3 ml injection in surface injection experiment. A possible explanation for this cross-over was that layer 1 had become saturated and continued saline injection spread into the second layer. Also it is difficult to inject saline at precise depths and control the spreading of saline in the specific layer, which can induce cross-talk between layers as well. Reduced skin artifacts in chicken breasts
We applied a set of current patterns to penetrate only the surface layer underneath the skin to evaluate the effect of impedance noise from the skin. At the same time, we applied another set of current patterns that penetrate deeper into the skin to extract information from deeper tissue layers with minimal interference from the skin. We modeled a chicken breast to have four tissue layers as specified in the caption to figure 2. We first measured and reconstructed the averaged impedance of each layer of a fresh chicken breast as the control. Then we artificially changed the skin condition of the chicken breast to make it dry, wet and wetter. Averaged impedance of each layer under these skin conditions was reconstructed and the difference of the impedance change compared to the control in four layers of the chicken breast was calculated. Reconstructed impedance tomography in figure 3(a) shows that the majority of 1143
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
Figure 3. Multi-channel excitation helps to reduce interference from different skin conditions. (a) Reconstructed tomography and relative conductivity change in chicken breast under different skin conditions, fresh skin condition as the control; (i) the surface in contact with the 8-electrode array was wiped dry before taking impedance measurement; (ii) the same surface was sprayed once with conductive saline before making impedance measurement; (iii) the same surface was sprayed again with conductive saline before making impedance measurement; (b) summary of relative conductivity change in four tissue layers within the chicken breast under three skin conditions.
conductivity changes occurred at the skin layer (layer 1) whereas much smaller changes were detected at deeper tissue layers 2 to 4. Averaged conductivity change within different layers for three skin conditions are compared in figure 3(b). Reconstructed conductivity change within layer 1 due to skin condition change is two to five times larger than layer 2. Minimal conductivity change was observed for deeper tissue layers 3 and 4. This data confirms that multi-channel excitation provides a method to reduce skin artifacts. Future work is needed to optimize the multi-channel current pattern and reconstruction algorithm for more refined separation of deeper hydration signals from skin artifacts. Ideally, it is desired to have no impedance change in layer 2 to 4 when there is skin condition change in layer 1. As you can see from the reconstructed impedance tomography, there is cross-talk between layer 1 and layer 2. This is an intrinsic spatial resolution limitation of EIT/EIS. More future work would be needed to focus on eliminating cross-talk between layers by optimizing reconstruction algorithm, electrode array design and prior calibration of spatial resolution. 1144
X Chen et al
Physiol. Meas. 35 (2014) 1137
(a)
(b)
(c) Figure 4. Compression sock experiment in healthy human subjects. (a) Left: healthy human subject wears compression sock only on left leg to reduce daily accumulation of fluid in lower limb; Middle: placement of a linear 8-electrode array for impedance data acquisition; Right: volumeter to measure leg volume by water displacement; (b) daily volume change percentage in right leg (control, blue bar) and left leg (compression sock, red bar) in five healthy human subjects, there is an average of 35 ± 35.9 ml fluid reduction in lower left leg with compression sock compared to the lower right leg (control, without compression sock); (c) average daily conductivity difference between right and left leg in five tissue layers for five healthy human subjects (Conductivity change in subcutaneous tissue layers 2 and 3 is statistically significant compared to change in skin layer 1. Results for t-test (paired two sample for means): ∗ p = 0.02 for comparing layer 1 with layer 2, p = 0.03 for comparing layer 1 with layer 3.)
Detection of layer-based hydration change in human subjects
We conducted a study to induce in vivo fluid changes in five healthy human subjects under IRB approval. This study manipulated the normal gravitational fluid accumulation that occurs in lower limbs from morning to afternoon. The study was designed such that fluid accumulation was restricted in the left leg by use of a compression sock. Baseline measurements of impedance and volume on both calves were taken upon presentation to the lab. Subjects then wore one compression sock on left leg while conducting normal daily activities (figure 4(a)). After 1145
X Chen et al
Physiol. Meas. 35 (2014) 1137
wearing the sock for 6 h, the sock was removed, and impedance and volume on both calves were measured again. The volumetric data in figure 4(b) demonstrates a mean volume difference of 35 ml (0.7% of gauge-measured leg volume) between the lower right leg volume change (no compression sock) and the lower left leg volume change (with compression sock) from morning to evening (Out of five healthy human subjects, subject P91 demonstrated a different volume change pattern than other human subjects, with slightly more fluid accumulation in the left leg with compression sock. One possible explanation might be that the pressure of compression sock for this subject is not strong enough to restrict fluid accumulation). Using our reconstruction algorithm with electrode drift compensation, we reconstructed the mean conductivity difference of right leg conductivity change and left leg conductivity change from morning to evening at five tissue layers in five subjects. The volume change in each subject shown in figure 4(b) confirm our hypothesis that there is more fluid accumulation in the right leg than in the left leg, indicating restriction effect of compression sock for healthy human subjects. Average relative conductivity change in each layer (figure 4(c))demonstrates that conductivity changes in subcutaneous tissue layers 2 and 3 are statistically different than that in skin layer 1. It is observed that there is more conductivity variability in the skin layer 1 indicating an intrinsic impedance distortion on the skin due to skin contact impedance and skin conditions. By dividing the calf into different tissue layers and enabling tomographic spatial reconstruction of impedance measurements with multi-channel EIT, we are able to differentiate skin artifacts from hydration signals at subcutaneous tissue layers. Another observation is that conductivity changes in layer 4 and 5 are small, which may result from smaller signal or insufficient sensitivity at deeper layers. Since excess fluid tends to accumulate in subcutaneous layers, we are more interested in monitoring subcutaneous layers. Nevertheless, it is important to address the issue of deeper layers in future work as well. Summary
We have demonstrated the feasibility of using multi-channel EIT to extract hydration information in different tissue layers with minimal skin interference. Our simultaneous, multichannel current excitation approach provides an effective method to separate electrode contact impedance and skin condition artifacts from hydration signals. Further studies are underway to integrate multi-frequency capability into the prototype system to extract extracellular and intracellular fluid information and to validate the system in a clinical setting with hemodialysis patients. Acknowledgment The authors would like to thank Dr Allen Garner for PCB board design simulation and Ms Maxine Gibeau for PCB board design assistance. References Banerjee D, Ma J Z, Collins A J and Herzog C A 2007 Long-term survival of incident hemodialysis patients who are hospitalized for congestive heart failure, pulmonary edema, or fluid overload Clin. J. Am. Soc. Nephrol. 2 1186–90 Boverman G, Isaacson D, Saulnier G J and Newell J C 2009 Methods for compensating for variable electrode contact in EIT IEEE Trans. Biomed. Eng. 56 2762–72 Chamney P W et al 2002 A new technique for establishing dry weight in hemodialysis patients via whole body bioimpedance Kidney Int. 61 2250–8 1146
X Chen et al
Physiol. Meas. 35 (2014) 1137
Cridlig J et al 2011 Formulation of a dry weight bioimpedance index in hemodialysis patients Int. J. Artif. Organs 34 1075–84 Fisch B J and Spiegel D M 1996 Assessment of excess fluid distribution in chronic hemodialysis patients using bioimpedance spectroscopy Kidney Int. 49 1105–9 Hecking M et al 2013 Significance of interdialytic weight gain versus chronic volume overload: consensus opinion Am. J. Nephrol. 38 78–90 Hoffer E C, Meador C K and Simpson D C 1969 Correlation of whole body impedance with total body water volume J. Appl. Physiol. 27 531–4 Jaeger J Q and Mehta R L 1999 Assessment of dry weight in hemodialysis: an overview J. Am. Soc. Nephrol. 10 392–403 Kao T J, Isaacson D, Newell J C and Saulnier G J 2006 A 3D reconstruction algorithm for EIT using a handheld probe for breast cancer detection Physiol. Meas. 27 S1–11 Khaled A M, McCutcheon M J, Reddy S, Pearman P L, Hunter G R and Weinsier R L 1988 Electrical impedance in assessing human body composition: the BIA method Am. J. Clin. Nutr. 47 789–92 Lindley E, Lynne A, Claire G and Garthwaite E 2011 Management of fluid status in haemodialysis patients: the roles of technology and dietary advice Technical Problems in Patients on Hemodialysis ed M G Penido (Rijeka: InTech) (www.intechopen.com/books/technical-problemsin-patients-on-hemodialysis/management-of-fluid-status-in-haemodialysis-patients-the-roles-oftechnology-and-dietary-advice) Machek P et al 2010 Guided optimization of fluid status in haemodialysis patients Nephrol. Dial. Transplant. 25 538–44 Moissl U, Arias-Guill´en M, Wabel P, Fontser´e N, Carrera M, Campistol J M and Maduell F 2013 Bioimpedance-guided fluid management in hemodialysis patients Clin. J. Am. Soc. Nephrol. 8 1575–82 Mueller J L, Isaacson D and Newell J C 1999 A reconstruction algorithm for electrical impedance tomography data collected on rectangular electrode array IEEE Trans. Biomed. Eng. 46 1379–86 US Renal Data System (USRDS) 2012 Annual Data Report: Atlas of Chronic Kidney Disease and EndStage Renal Disease in the United States (Bethesda, MD: National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases) Wabel P et al 2008 Towards improved cardiovascular management: the necessity of combining blood pressure and fluid overload Nephrol. Dial. Transplant. 23 2965–71 Wizemann V, Wabel P, Chamney P, Zaluska W, Moissl U, Rode C, Malecka-Masalska T and Marcelli D 2009 The mortality risk of overhydration in haemodialysis patients Nephrol. Dial. Transplant. 24 1574–9 Zhu F, Schneditz D, Kaufman A M and Levin N W 2000 Estimation of body fluid changes during peritoneal dialysis by segmental bioimpedance analysis Kidney Int. 57 299–306
1147
