660
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
An Innovative Adaptive Control Strategy for Sensorized Left Ventricular Assist Devices Antonella Verbeni, Rossella Fontana, Michele Silvestri, Giuseppe Tortora, Member, IEEE, Monica Vatteroni, Maria Giovanna Trivella, and Paolo Dario, Fellow, IEEE
Abstract—Nowadays advanced heart failure is mainly treated through heart transplantation. However, the low availability of donors makes the research of alternative therapies urgent. Continuous-flow left ventricular assist devices (LVADs) are going to assume a more significant role in assisting the failing heart. A recent challenge in clinical practice is the possibility to use LVAD as long-term therapy rather than as a bridge to transplantation. For this reason, more comfortable devices, able to dynamically adapt to the physiological cardiac demand in relation to the patient activity level, are needed in order to improve the life quality of patients with implants. Nevertheless, no control system has been developed yet for this purpose. This work proposes an innovative control strategy for a novel sensorized LVAD, based on the continuous collection of physical and functional parameters coming from implantable sensors and from the LVAD itself. Thanks to the proposed system, both the patient and the LVAD conditions are continuously monitored and the LVAD activity regulated accordingly. Specifically, a Proportional Integrative (PI) and a threshold control algorithms have been implemented, respectively based on flow and pressure feedbacks collected from the embedded sensors. To investigate the feasibility and applicability of this control strategy, an on-bench platform for LVADs sensing and monitoring has been developed and tested. Index Terms—Blood pumps, heart failure, left ventricular assist devices (LVAD), mechanical circulatory support system, remote monitoring and control.
I. INTRODUCTION
H
EART FAILURE (HF) is the highest cause of mortality and morbidity worldwide. Moreover, due to the limited availability of donor organs and the growing number of patients who run the risk of heart failure, heart transplantation is the main therapy for the treatment of this disease [1]. In this scenario, mechanical circulatory support systems are going to assume a growing and important role to assist the failing heart [2].
Manuscript received February 14, 2014; revised June 09, 2014; accepted August 01, 2014. Date of publication September 11, 2014; date of current version November 06, 2014. The work presented in this paper was developed in the framework of the European project SensorART (a remote controlled Sensorized ARTificial heart enabling patients empowerment and new therapy approaches; GA 248763). This paper was recommended by Associate Editor R. F. Yazicioglu. A. Verbeni, R. Fontana, M. Silvestri, G. Tortora, M. Vatteroni, and P. Dario are with The BioRobotics Institute, Scuola Superiore Sant’Anna, 56025 Pisa, Italy (e-mail: [email protected]). M. G. Trivella is with the Clinical Physiology Institute, National Research Council, 56025 Pisa, Italy (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBCAS.2014.2346015
Since the first application of a ventricular assist device (VAD), in 1963 by DeBakey [3], a large variety of VADs have been developed and are currently in use to treat and assist patients affected by severe heart failure with the prospect of a future heart transplantation [4]–[6]. The most recent improvements in VAD technology lead to the development of a new era of blood pumps: the continuous-flow LVADs [7]–[10]. These LVADs show several advantages, with respect to the previous pulsatile ones: smaller size, lighter weight and consequently easier implantation. As such, they represent a dramatic step forward, both in terms of durability and survival time, opening the possibility to consider LVADs not only as a bridge to transplantation, but also a long-term therapy, as well as destination therapy [11]–[13]. Slaughter et al. performed a randomized trial in patients with advanced heart failure comparing outcomes of pulsatile and continuous-flow LVAD implantation. This study demonstrated a high improvement of the probability of survival without risk of stroke and reoperation at two years [14]. Furthermore, it is worth mentioning that patients assisted with LVAD can get a good functional recovery of the heart and a better life quality than in the case of heart transplantation and bridge to transplantation [15]. Nowadays, most of the commercial continuous-flow LVADs operate at fixed speeds that can be changed only manually by the specialist who sets the desired pump speed, depending on the physiological needs of the patient [16]. The VAD controller automatically adjusts the current and voltage applied to the pump to achieve and maintain the desired speed. Although the reliability of the device for automated operation has been well proven by thorough analysis and tests, a continuous, remote and automatic monitoring system could add much more consistency and safety to the management of the device and the patient. Notwithstanding, no control system for continuous-flow LVAD able to automatically respond to the physiological cardiac demand is available and only a few artificial heart systems include remote monitoring functions (e.g., [17]). Although few additional systems are reported to have network-based remote monitoring functions, continuous, automatic and remote monitoring systems and database systems are still in an early development stage with a limited number of relevant research studies [18], [19]. Considering the goal of a long-term therapy aiming to the complete reduction of patient’s dependence on clinical management, the implementation of an automatic, robust and adaptive control strategy, able to quickly adjust pump parameters according to the patient’s real-time status, is essential. A revolutionary approach to reach the aforementioned aim could be
1932-4545 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
Fig. 1. Overview of the platform within the SensorART system.
the implementation of a continuous monitoring strategy with the purpose to optimize and personalize the heart unloading degree. The main topic of the work presented in this paper is the development of an innovative control strategy for LVADs, able to automatically adapt to the patient’s cardiac demand by means of an automatic feedback directly based on physical parameters. In particular, the work presented here is an important step towards the development of a fully sensorized and miniaturized pump for the treatment of patients who suffer from heart failure: the SensorART platform (www.sensorart.eu). The SensorART platform integrates different hardware and software components to improve both the quality of life of LVAD implanted patients and the workflow of specialists. The platform is composed of different modules such as sensors, a signal acquisition module, a hardware controller, a remote control framework and a decision support system. The sensor module, measuring pressure and flow parameters, will be used to monitor both the status of the patient and the correct functionality of the LVAD. The LVAD and the embedded monitoring sensors will be integrated in a self-contained implantable platform. This platform is in a closed loop with a central control unit, called the Auto-Regulation Unit (ARU). The ARU is an external and wearable FPGA-based controller wirelessly linked to the implanted sensors and to the VAD actuators, which ensures a continuous monitoring of the parameters and implements the control algorithms. An overview of the whole system is shown in Fig. 1. Although the final goal will be to implement all of the control algorithms into the ARU controller, they have been validated in this work by means of a PC-based platform simulating the ARU functionalities as a first step. Therefore, in this work, an adaptive control strategy has been implemented and evaluated in order to assess its feasibility and to enable a future smooth implementation of the final system. Furthermore, to investigate the feasibility and applicability of the proposed control strategy, an on-bench platform as cardiac simulator has been developed and exploited. II. MATERIAL AND METHOD The primary goal of the proposed control strategy is to reduce patient dependence on clinical management enabling an
661
auto-adjustment of the blood flow provided by the LVAD to the patient’s heart. This will be achieved thanks to real time control algorithms working within certain pre-set safety thresholds and control parameters, according to the specialist’s supervision. As stated above, currently known methodologies for adjusting pump parameters employ a manual setting of the pump speed until the patient perceives a comfortable level of perfusion. Considering this clear limitation, the goal is to adopt a closed loop control strategy able to provide a constant blood perfusion even if the patient’s conditions change during his/her daily activities. In particular, the control strategy implemented is based upon two different control methods: a Proportional Integrative (PI) control on flow values and a threshold control on pressure data in order to both control the blood perfusion and check the pressure drop to the heads of the pump. Both of these controls are performed by setting the proper speed, usually expressed in revolutions per minute (rpm), of the pump and sending the corresponding command to the pump controller. The control algorithms are implemented through the LabView software (National Instruments, USA), used to acquire and store data from both implantable sensors and the pump and to process these data for controlling the LVAD functionalities. Even using the LabView software, a Human-Machine-Interface was developed to allow the physician a simpler and better managing of the pump. In particular, through this dedicated interface the physician is able to choose a proper level of perfusion and to set the corresponding flow value. According to this value, the implemented control algorithms automatically send to the pump controller the proper command to change the rpm and check the output of the flow sensor. The target value is reached thanks to a PI control that dynamically and precisely changes the pump speed in order to achieve the desired flow value. This control method has been implemented because it exhibits a fast, fine and stable attainment and maintenance of the desired flow. Concerning the threshold control on pressure, the main goal is to monitor if high, fast and/or repetitive changes in the pressure values occur. Indeed, these changes could be the signal for adverse and undesired events (such as certain pathological events and/or LVAD malfunctioning) that could negatively affect the patient condition and eventually provoke thrombus formation [text deleted]. To avoid this problem, the threshold control on drop pressure variations resets the pump if pressure oscillations are detected. It is worth mentioning that, even though these algorithms are developed and tested on a dedicated on-bench platform, they could be intended as a general methodology for monitoring and controlling continuous-flow LVADs. Indeed the final ARU will be a flexible device that could be easily set for different cardiac pumps and sensors, thus ensuring a feedback control.
A. Experimental Setup: The On-Bench Platform An on-bench platform has been designed and exploited in order to implement and consecutively assess the suitability and feasibility of the proposed control algorithms.
662
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
cular pathological conditions will be exploited to perform further assessment. B. Control Design
Fig. 2. (Left) On-bench platform setup and (right) overall schematics.
The platform includes an LVAD, embedded pressure and flow sensors and the conditioning electronics, which are physically connected to the PCs (Fig. 2). The core of the integrated test-bench is the Synergy™ Device, a VAD pump provided by CircuLite (Aachen, Germany) with a maximum blood-flow of 4.25 liters per minute. A wired electronic controller allows a manual setting of the pump when the controls are off. The pump operates in a closed hydraulic circuit that has been set up by using a silicone tube with an internal diameter of inches (9.52 mm) and a wall thickness of inches (1.59 mm). The hydraulic circuit is connected to a tank containing the fluid (composed of a solution of 48.3 % glycerin). This solution has been selected in order to reproduce the blood viscosity and to prevent damage to the pump. The platform has been equipped with commercial pressure and flow sensors. In particular, two catheters with piezoresistive tips (Mikro-Tip® by the Millar Instruments, www.millar. com) have been positioned at the pump input and output, respectively. These sensors were employed because they are suitable to be used in the implantable platform. Additionally, they present high linearity and sensitivity (up to 5 V/V/mmHg) and can measure pressures ranging from mmHg to mmHg (from to 40 kPa). This pressure range is compatible with the range of interest for the application, typically from 40 to 200 mmHg. In order to measure the flow, a non-invasive clamp-on transducer provided by em-tec (Finning, Germany) based on the principle of the transit time difference from ultrasonic acoustic waves in up and downstream direction, was integrated. This ultrasonic flowmeter is characterized by an excellent accuracy and stability; moreover it has been calibrated for flow measurement in Polyvinyl Chloride (PVC) and silicone tubes. The transducer is able to measure a maximum flow of 10000 ml/min, which matches the application flow range of 3000–6000 ml/min. In the final setup the pressure sensors were located at the input and output of the pump at 1 mm distance, whereas the flowmeter was positioned 15 cm downstream from the pump. The presented experimental setup has been exploited with the aim to implement and assess the control algorithms since it allows for easy simulation of pathological events such as vasoconstriction (i.e., hypertension or suction), achieved by manual incrementing or decrementing the peripheral resistance through a manual action on the hydraulic circuit (e.g., by squeezing the silicon tube). However, a more advanced cardiovascular simulator [20] allowing simulation of a higher set of cardiovas-
The two control algorithms have been implemented using the LabView development environment resulting in a simple controller and a user-friendly interface. In order to decrease the overall computational load, the LabView interfaces, respectively developed for acquiring data and controlling the LVAD, run on two different dedicated PCs. For this purpose, the hardware components of the platform are connected to the first PC, named “Acquiring PC”, which acquires and stores on-line data. On the other hand, the algorithms devoted to process these data and to implement the commands to be sent to the pump controller run on the second PC, named “Control PC”. The logic work flow that has been used to implement the control unit and all the communications are graphically represented in Fig. 3. In particular, the PC-based platform communicates with the hardware module composing the on-bench platform through serial communication lines and the National Instruments data acquisition (DAQ) board. The Acquiring PC continuously acquires, visualizes and stores the data. In particular, the data acquired by the platform are pump speed and current, flow and pressure parameters. Pump speed and current are directly taken from the pump controller and transferred to the control board by a serial communication line for checking the device status. Similarly, the flow data from the ultrasonic transducer are transferred to the PC by a second serial communication line. On the other hand, pressure data are acquired using the DAQ board, transferred to the PC via USB and acquired through the LabView dedicated functions. All the data are visualized and stored in the Acquiring PC thanks to the dedicated interface exploiting the National Instrument VISA and DAQmx palettes. These data are constantly collected and converted into format strings in order to be transmitted to the Controlling PC. At every running cycle, the latter continuously controls pressure and flow values. In particular, regarding the flow control, the algorithm checks, at every cycle, if a new desired flow value is set in the interface. When a new value is inserted, the system compares it with the current flow value and discerns two different conditions: 1) the desired flow is higher than the actual read one; 2) the desired flow is lower than the actual read one. A PI control was selected in order to obtain the increment or decrement needed for reaching the desired flow, and directly change the pump speed. The theoretic basis of the PI control are reported in [21]. The PI control was implemented in LabView with the transfer function form according to the following law (1): (1) where and represent, respectively, the proportional gain and the integral time. A manual tuning was used to define their proper values. U(s) represents the Laplace transform of the controller output and E(s) the Laplace transform of the inlet. In our implementation E(s) is calculated depending on the situation deas the detected by the case structure. In fact, considering
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
663
Fig. 3. Logical work flow of the control strategy.
Fig. 4. Experimental data and linear fitting of flow versus rpm of the on-bench platform.
sired flow value inserted by the clinician and as the actual read flow value: when is 1) E(s) is the Laplace transform of higher than ; when is 2) E(s) is the Laplace transform of . higher than In order to minimize the difference between flow values, the PI control will return an output U(s), which is the quantity in rpm that must be added or subtracted (according to the calibration curve reported in Fig. 4) to the actual value. As explained above, the activation of the PI control depends on the insertion of a flow value into the interface. Conversely, the control of pressure is an automatic control, continuously active for every running cycle. A threshold control on the standard deviation of the average pressure drop values is implemented. More in detail, a dynamic array is assembled and updated as soon as a new pressure value is available, and its standard deviis calculated according to the following law (2): ation (2) in our case) and is where n is the number of samples ( their mean value. The number of samples of the dynamic array has been set to 100 to enable quick responses to sudden changes
related to adverse and undesired events while remaining unaffected by slower physiological changes. The standard deviation of the dynamic array is continuously monitored and compared to a pre-defined threshold. In order to define an accurate and suitable threshold value for the implemented pump, a specific test was carried out keeping the pump in a steady condition (i.e., no suction or unstable events occurring). The standard deviation in these conditions has been evaluated in order to define the pressure variations during normal operation of the pump. Then, the same test was carried out simulating a suction event to define the typical standard deviation in this case. By these measurements it was possible to define a proper threshold which must be higher than the one associated with the normal working condition of the pump and, at the same time, lower with respect to the one measured during a suction event. The typical value of standard deviation to foresee a suction event has been experimentally assessed. In working conditions, when the measured standard deviation exceeds the set threshold, the algorithm processes the command string to reset the pump. After the reset command is sent, the control unit waits a minimum amount of time for the system to reach the normal conditions (i.e., standard deviations below the threshold). Every time a control algorithm is activated, the resulting command must be primarily transmitted via the serial communication line from the Controlling PC to the Acquiring PC and the latter communicates via another serial line with the pump controller. Furthermore, the system is able to produce alert signals if one of the monitored parameters (flow, pressure and rpm) exceeds its specified range. In fact, to meet safety requirements, alarms based on threshold controls have been implemented. It is worth noting that, if the control units stop working properly, the pump is able to guarantee a constant blood perfusion to the patient. C. Human-Machine-Interface Two simple, intuitive and user-friendly Human-Machine-Interfaces (HMIs) were implemented with the LabView software, respectively, for the Acquiring and Controlling PCs. The aim of the HMIs is to help the final user to simply and efficiently control the LVAD. The Acquiring PC interface allows the user to
664
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
Fig. 5. Human-Machine-Interface.
Fig. 7. Results from the PI controller; red arrows indicate the activation of PI controller. (Top) Flow increment from 3500 ml/min to the setting value of 5500 ml/min. (Bottom) Corresponding increment of rpm from 20000 to 30000.
Fig. 6. Results from the PI controller; red arrows indicate the activation of PI controller. (Top) Flow decrement from 5500 ml/min to the setting value of 3500 ml/min. (Bottom) Corresponding decrement of rpm from 30000 to 20000.
visualize the parameters in real-time. The first page of the Tab Control in the Front Panel, named “Read” allows the user to read the numerical values of the acquired data (speed, current, flow and pressure) and to observe their temporal trend through their
respective graphs. Two string indicators are also represented in order to check the correctness of the data transmitted to the Controlling PC and of the command activating the PI control transmitted from the Controlling PC. The second page of the Tab control, named “Setting”, allows the user to set all the parameters needed for the communication between the Acquiring PC and the hardware components of the platform. The Controlling PC interface is the most critical and important, since it allows the user to insert the desired flow value, thus activating the PI control simply by pressing a button. At the same time, this interface is provided with a LED which is switched on if the pressure variation exceeds the safety threshold, indicating that the reset control is activated. All these functionalities are implemented in the first page of Tab Control, named “Control”. Even in this page a button is present enabling the reading of data transmitted by the Acquiring PC. The numeric values of these data (rpm, flow and pressure) are also shown. A “Setting” page allows the user to properly set the communication parameters between the two PCs. The developed HMIs are shown in Fig. 5.
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
Fig. 8. PI controller activated; red arrows indicate the squeeze of the silicon tube (peripheral resistance increase) and black arrows indicate the release of the silicon tube (peripheral resistance decrease). (Top) Flow is kept constant at the set value of 4000 ml/min. (Bottom) Corresponding increment of rpm.
III. RESULTS The on-bench platform has been exploited in order to assess the reliability and consistency of the control algorithms for monitoring and controlling the LVAD parameters. The PI control and the pressure threshold control methods were implemented and tested on bench; the performed experiments are described in the following sections. A. PI Control Based on the Flow Sensor The behavior of the PI control was analyzed in two experimental phases using the dedicated on-bench platform. During the first phase, the pump speed was set at 30000 rpm. As shown in Figs. 4 and 6, the corresponding flow value in this condition is 5500 ml/min. In order to activate the PI control, a desired flow value of 3500 ml/min was set through the HMI of the Controlling PC. Fig. 6 shows clearly the effects of the PI control in redirecting the flow to the desired value, by adjusting the pump speed. Indeed, as soon as a speed of 20000 rpm is set according to the implemented algorithms, the desired flow
665
Fig. 9. PI controller deactivate. (Top) Flow decreases after the silicon tube is squeezed and goes up to set value of 4000 ml/min only after the tube is released. (Bottom) Pump speed stays constant.
value of 3500 ml/min is reached. In the second experimental phase, the starting speed was 20000 rpm (corresponding to a flow value of 3500 ml/min). As in the previous case, a desired flow value of 5500 ml/min was set. The flow and speed trends in the case of flow increment are shown in Fig. 7. In both experiments the PIcontrol has demonstrated to work properly for adjusting the flow to the desired value with a suitable response time. After a small overshoot/over damping (below a threshold of ml/min), the flow goes back below a threshold of ml/min in about 20 sec These values are compatible with medical specifications, also considering that the patient’s heart is always active and able to support the circulatory system for a certain time (up to 60–120 min). A further assessment of the suitability of the implemented PI control was carried out with the aim to demonstrate its ability to face the physiological changes that can occur during the daily life (e.g., temperature increase and chest compression). The control had to maintain constant blood perfusion despite changes in the patient’s condition. In Fig. 8 a peripheral resistance change was simulated to show the effect of the PI control in this situation. The flow value was
666
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
Fig. 10. Suction event with no control activation; (top) effect on pressure and (bottom) flow, respectively.
set to 4000 ml/min and the peripheral resistance was consecutively increased twice by squeezing the silicon tube. As shown in Fig. 8 (bottom), the system reacted to this condition by increasing the pump speed in order to maintain constant perfusion. Fig. 8 (top) demonstrates that the flow was recovered to the desired value of 4000 ml/min within about 15 s and maintained constant until the tube was released. In Fig. 9 the silicon tube was squeezed three times without the activation of the PI control. As shown, no changes in pump speed occurred and the flow value returned to 4000 ml/min only when the tube was released. B. Control on Pressure Sensor A further experimental session was carried out to assess the performance of the algorithms for controlling the pressure variations within a pre-set range (i.e., a threshold). In order to fulfill this task, the silicon tube was squeezed to simulate undesired events (e.g., suction or hypertension). Fig. 10 shows the effect of two consecutive sudden events on both pressure and flow. We can observe that, as soon as an event occurs, the flow quickly drops while the pressure increases. If no control is active this condition continues until the event ends. In Fig. 11 the results of the experiment with the activation of the control are reported.
Fig. 11. Suction event with control activation; (top) effect on pressure and (bottom) flow, respectively.
Thanks to the control on pressure thresholds, when an event is detected (i.e., the pressure value exceeds the threshold) the pump is reset. This produces a sudden drop in the pressure, but is rapidly recovered (in 2 sec) to the values acquired before the event occurred. IV. CONCLUSION In this work, an innovative control strategy for controlling and modeling LVADs has been developed. In particular, two different control algorithms have been implemented: a proportional-integrative and a threshold control methods based on flow and pressure values, respectively. Furthermore, a HMI has been developed to allow for a simpler and better management of the pump by the specialist. The control unit has been experimentally evaluated exploiting a dedicated on-bench platform embedding an LVAD and implantable sensors for flow and pressure. It has been demonstrated that the algorithms are able to control the LVAD in a feedback-loop directly based upon sensors measurements. In particular, the control algorithms showed an adaptive behavior to the new condition in about 2 sec and can suddenly change the LVAD speed as well.
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
ACKNOWLEDGMENT The authors would like to thank CircuLite GmbH (www.circulite.net) and em-tec GmbH (www.em-tec.com) for providing the pump and the flow sensor.
REFERENCES [1] A. H. H. AlOmari et al., “Developments in control systems for rotary left ventricular assist devices for heart failure patients: A review,” Phys. Meas., vol. 34, pp. R1–R27, Jan. 2013. [2] M. Strüber et al., “The current status of heart transplantation and the development of” artificial heart systems,” Dtsch Arztebl Int., vol. 106, pp. 471–477, 2009. [3] M. E. DeBakey, “Left ventricular bypass pump for cardiac assistance. Clinical experience,” Amer. J. Cardiol., vol. 27, pp. 3–11, Jan. 1971, 1971. [4] G. S. Kumpati et al., “Left ventricular assist device bridge to recovery: A review of the current status,” Ann. Thoracic Surgery, vol. 71, pp. S103–S108, Mar. 2001. [5] Ž. Sutlić et al., “Overview of new mechanical circulatory support devices,” Rad Hrvatske Akademije Znanosti i Umjetnosti. Medicinske Znanosti, vol. 509, pp. 123–132, 2011. [6] C. A. Milano and A. A. Simeone, “Mechanical circulatory support: Devices, outcomes and complications,” Heart Failure Rev., vol. 18, pp. 35–53, Jan. 2013. [7] S. Salzberg et al., “Left ventricular assist device as bridge to heart transplantation—Lessons learned with the MicroMed DeBakey axial blood flow pump,” Eur. J. Cardio-Thoracic Surgery, vol. 24, pp. 113–118, July 2003. [8] S. Westaby et al., “Destination therapy with a rotary blood pump and novel power delivery,” Eur. J. Cardio-Thoracic Surgery, vol. 37, pp. 350–356, Feb. 2010. [9] D. Esmore et al., “A prospective, multicenter trial of the VentrAssist left ventricular assist device for bridge to transplant: Safety and efficacy,” J. Heart Lung Transplant., vol. 27, pp. 579–588, Jun. 2008. [10] M. Strueber et al., “Multicenter evaluation of an intrapericardial left ventricular assist system,” J. Amer. College Cardiol., vol. 57, pp. 1375–1382, Mar. 22, 2011. [11] T. Hrobowski and D. E. Lanfear, “Ventricular assist devices: Is destination therapy a viable alternative in the non-transplant candidate?,” Current Heart Failure Rep., vol. 10, pp. 101–107, Mar. 2013, 2013. [12] E. A. Rose et al., “Long-term use of a left ventricular assist device for end-stage heart failure—N Engl J Med 2001;345 : 1435–1443,” J. Cardiac Failure, vol. 8, pp. 59–60, Apr. 2002. [13] D. Goldstein et al., “Comparison of costs, readmissions and days out of hospital at one year between heart transplantation and continuous flow LVAD,” J. Heart Lung Transplant., vol. 32, p. S98, 2013. [14] M. S. Slaughter et al., “Advanced heart failure treated with continuous-flow left ventricular assist device,” New Eng. J. Med., vol. 361, pp. 2241–2251, 2009. [15] E. J. Birks et al., “Left ventricular assist device and drug therapy for the reversal of heart failure,” New Engl. J. Med., vol. 355, pp. 1873–1884, 2006. [16] C. A. Giridharan et al., “Modeling and control of a brushless DC axial flow ventricular assist device,” Asaio J., vol. 48, pp. 272–289, May–Jun. 2002. [17] T. Mussivand et al., “Wireless monitoring and control for implantable rotary blood pumps,” Artif. Organs, vol. 21, pp. 661–664, 1997. [18] S. Hübler et al., “Development of a database of patients supported by ventricular assist devices,” Asaio J., vol. 49, pp. 340–344, 2003. [19] R. Kosaka et al., “Tsukuba remote monitoring system for continuous-flow artificial heart,” Artif. Organs, vol. 27, pp. 897–906, 2003. [20] M. Kozarski et al., “A hybrid (hydro-numerical) cardiovascular model: Application to investigate continuous-flow pump assistance effect,” Biocybern. Biomed. Eng., vol. 32, pp. 77–91, 2012, 2012. [21] Modern Control Engineering, K. Ogata and Y. Yang, Eds. Upper Saddle River, NJ, USA: Prentice Hall, 1990.
667
Antonella Verbeni was born in Reggio Calabria, Italy, in 1986. She received her Master’s degree in biomedical engineering from the University of Pisa, Pisa, Italy, in 2011. Currently, she is working toward the Ph.D. degree in biorobotics at The BioRobotics Institute of the Scuola Superiore Sant’Anna, Pisa, Italy. In September 2010, she joined the Scuola Superiore Sant’Anna, working on surgical robotics and focusing her activity on endoscopic capsules. Her main research interests include innovative devices for cardiovascular disease with particular attention on the study and the application of ultrasound.
Rossella Fontana was born in Agrigento, Italy, in 1986. She received the Master’s degree in biomedical engineering from the University of Pisa, Pisa, Italy, in May 2011. Currently, she is working toward the Ph.D. degree at The BioRobotics Institute of Scuola Superiore Sant’Anna, Pisa, Italy. She is working on the development of sensorized devices for implantable cardiovascular applications.
Michele Silvestri was born in Pisa, Italy, in 1984. He received the Master’s degree in biomedical engineering (cum laude) from the University of Pisa, Pisa, Italy, in 2010. Currently, he is working toward the Ph.D. degree in biorobotics at The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy. His current research interests include biomechatronics and medical imaging.
Giuseppe Tortora (S’09–M’13) was born in Isernia, Italy, in 1983. He received the Master’s degree in biomedical engineering from the University of Pisa, Pisa, Italy, and the Ph.D. degree in biorobotics from The BioRobotics Institute of Scuola Superiore Sant’Anna, Pisa, Italy, in 2008 and 2012, respectively. In April 2007 he joined the Scuola Superiore Sant’Anna, focusing his activity on medical robotics and biomechatronic systems. He spent a period as Visiting Researcher at Imperial College London, London, U.K., and Carnegie Mellon University, Pittsburgh, PA, USA, in 2011. His main research interests are in the field of biorobotics and advanced systems for minimally invasive diagnosis, surgery, and therapy.
Monica Vatteroni was born in La Spezia, Italy, in 1975. She received the M.S. degree in electrical engineering from the University of Pisa, Pisa, Italy, and the Ph.D. degree in physics from the University of Trento, Trento, Italy, in 2001 and 2008, respectively. Currently, she is with the Scuola Superiore Sant’Anna, Pisa, Italy as a Postdoctoral Fellow, where she is responsible for research and development of image sensors, vision systems, and sensorized biomedical platforms.
668
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
Maria Giovanna Trivella received the medical doctor degree from the University of Pisa, Pisa, Italy, in 1976 and majored in cardiology in 1981. Currently, she is a Senior Investigator of the Italian National Research Council (CNR), Head of the Experimental Laboratory of IFC-CNR, and Head of Experimental Surgery Laboratory, a joint laboratory in cooperation between CNR and the University of Pisa. She is the coordinator of SensorART—a remote controlled sensorized artificial heart enabling patients empowerment and new therapy approaches (FP7-ICT-248763), 2010–2014. Her current research fields of interest are interdisciplinary and translational research in medicine. Dr. Trivella was awarded by the European Science Foundation for the Exploratory Workshop, “Molecular signaling in cardiovascular and oncological disease: Similar and shared pathways,” in 2007.
Paolo Dario (F’02) received the Master’s degree in mechanical engineering from the University of Pisa, Pisa, Italy, in 1977. He is Professor of Biomedical Robotics at the Scuola Superiore Sant’Anna, Pisa, Italy, where he supervises a team of approximately 150 young researchers. He has the role of Scientific-Technical manager of SensorART. His main research interest is biorobotics, including mechatronic and robotic systems for rehabilitation, prosthetics, surgery, and microendoscopy. He is the author of more than 160 ISI journal papers, many international patents, and several book chapters on medical robotics. He was the recipient of the Joseph Engelberger Award as a Pioneer of Biomedical Robotics.
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
An Innovative Adaptive Control Strategy for Sensorized Left Ventricular Assist Devices Antonella Verbeni, Rossella Fontana, Michele Silvestri, Giuseppe Tortora, Member, IEEE, Monica Vatteroni, Maria Giovanna Trivella, and Paolo Dario, Fellow, IEEE
Abstract—Nowadays advanced heart failure is mainly treated through heart transplantation. However, the low availability of donors makes the research of alternative therapies urgent. Continuous-flow left ventricular assist devices (LVADs) are going to assume a more significant role in assisting the failing heart. A recent challenge in clinical practice is the possibility to use LVAD as long-term therapy rather than as a bridge to transplantation. For this reason, more comfortable devices, able to dynamically adapt to the physiological cardiac demand in relation to the patient activity level, are needed in order to improve the life quality of patients with implants. Nevertheless, no control system has been developed yet for this purpose. This work proposes an innovative control strategy for a novel sensorized LVAD, based on the continuous collection of physical and functional parameters coming from implantable sensors and from the LVAD itself. Thanks to the proposed system, both the patient and the LVAD conditions are continuously monitored and the LVAD activity regulated accordingly. Specifically, a Proportional Integrative (PI) and a threshold control algorithms have been implemented, respectively based on flow and pressure feedbacks collected from the embedded sensors. To investigate the feasibility and applicability of this control strategy, an on-bench platform for LVADs sensing and monitoring has been developed and tested. Index Terms—Blood pumps, heart failure, left ventricular assist devices (LVAD), mechanical circulatory support system, remote monitoring and control.
I. INTRODUCTION
H
EART FAILURE (HF) is the highest cause of mortality and morbidity worldwide. Moreover, due to the limited availability of donor organs and the growing number of patients who run the risk of heart failure, heart transplantation is the main therapy for the treatment of this disease [1]. In this scenario, mechanical circulatory support systems are going to assume a growing and important role to assist the failing heart [2].
Manuscript received February 14, 2014; revised June 09, 2014; accepted August 01, 2014. Date of publication September 11, 2014; date of current version November 06, 2014. The work presented in this paper was developed in the framework of the European project SensorART (a remote controlled Sensorized ARTificial heart enabling patients empowerment and new therapy approaches; GA 248763). This paper was recommended by Associate Editor R. F. Yazicioglu. A. Verbeni, R. Fontana, M. Silvestri, G. Tortora, M. Vatteroni, and P. Dario are with The BioRobotics Institute, Scuola Superiore Sant’Anna, 56025 Pisa, Italy (e-mail: [email protected]). M. G. Trivella is with the Clinical Physiology Institute, National Research Council, 56025 Pisa, Italy (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBCAS.2014.2346015
Since the first application of a ventricular assist device (VAD), in 1963 by DeBakey [3], a large variety of VADs have been developed and are currently in use to treat and assist patients affected by severe heart failure with the prospect of a future heart transplantation [4]–[6]. The most recent improvements in VAD technology lead to the development of a new era of blood pumps: the continuous-flow LVADs [7]–[10]. These LVADs show several advantages, with respect to the previous pulsatile ones: smaller size, lighter weight and consequently easier implantation. As such, they represent a dramatic step forward, both in terms of durability and survival time, opening the possibility to consider LVADs not only as a bridge to transplantation, but also a long-term therapy, as well as destination therapy [11]–[13]. Slaughter et al. performed a randomized trial in patients with advanced heart failure comparing outcomes of pulsatile and continuous-flow LVAD implantation. This study demonstrated a high improvement of the probability of survival without risk of stroke and reoperation at two years [14]. Furthermore, it is worth mentioning that patients assisted with LVAD can get a good functional recovery of the heart and a better life quality than in the case of heart transplantation and bridge to transplantation [15]. Nowadays, most of the commercial continuous-flow LVADs operate at fixed speeds that can be changed only manually by the specialist who sets the desired pump speed, depending on the physiological needs of the patient [16]. The VAD controller automatically adjusts the current and voltage applied to the pump to achieve and maintain the desired speed. Although the reliability of the device for automated operation has been well proven by thorough analysis and tests, a continuous, remote and automatic monitoring system could add much more consistency and safety to the management of the device and the patient. Notwithstanding, no control system for continuous-flow LVAD able to automatically respond to the physiological cardiac demand is available and only a few artificial heart systems include remote monitoring functions (e.g., [17]). Although few additional systems are reported to have network-based remote monitoring functions, continuous, automatic and remote monitoring systems and database systems are still in an early development stage with a limited number of relevant research studies [18], [19]. Considering the goal of a long-term therapy aiming to the complete reduction of patient’s dependence on clinical management, the implementation of an automatic, robust and adaptive control strategy, able to quickly adjust pump parameters according to the patient’s real-time status, is essential. A revolutionary approach to reach the aforementioned aim could be
1932-4545 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
Fig. 1. Overview of the platform within the SensorART system.
the implementation of a continuous monitoring strategy with the purpose to optimize and personalize the heart unloading degree. The main topic of the work presented in this paper is the development of an innovative control strategy for LVADs, able to automatically adapt to the patient’s cardiac demand by means of an automatic feedback directly based on physical parameters. In particular, the work presented here is an important step towards the development of a fully sensorized and miniaturized pump for the treatment of patients who suffer from heart failure: the SensorART platform (www.sensorart.eu). The SensorART platform integrates different hardware and software components to improve both the quality of life of LVAD implanted patients and the workflow of specialists. The platform is composed of different modules such as sensors, a signal acquisition module, a hardware controller, a remote control framework and a decision support system. The sensor module, measuring pressure and flow parameters, will be used to monitor both the status of the patient and the correct functionality of the LVAD. The LVAD and the embedded monitoring sensors will be integrated in a self-contained implantable platform. This platform is in a closed loop with a central control unit, called the Auto-Regulation Unit (ARU). The ARU is an external and wearable FPGA-based controller wirelessly linked to the implanted sensors and to the VAD actuators, which ensures a continuous monitoring of the parameters and implements the control algorithms. An overview of the whole system is shown in Fig. 1. Although the final goal will be to implement all of the control algorithms into the ARU controller, they have been validated in this work by means of a PC-based platform simulating the ARU functionalities as a first step. Therefore, in this work, an adaptive control strategy has been implemented and evaluated in order to assess its feasibility and to enable a future smooth implementation of the final system. Furthermore, to investigate the feasibility and applicability of the proposed control strategy, an on-bench platform as cardiac simulator has been developed and exploited. II. MATERIAL AND METHOD The primary goal of the proposed control strategy is to reduce patient dependence on clinical management enabling an
661
auto-adjustment of the blood flow provided by the LVAD to the patient’s heart. This will be achieved thanks to real time control algorithms working within certain pre-set safety thresholds and control parameters, according to the specialist’s supervision. As stated above, currently known methodologies for adjusting pump parameters employ a manual setting of the pump speed until the patient perceives a comfortable level of perfusion. Considering this clear limitation, the goal is to adopt a closed loop control strategy able to provide a constant blood perfusion even if the patient’s conditions change during his/her daily activities. In particular, the control strategy implemented is based upon two different control methods: a Proportional Integrative (PI) control on flow values and a threshold control on pressure data in order to both control the blood perfusion and check the pressure drop to the heads of the pump. Both of these controls are performed by setting the proper speed, usually expressed in revolutions per minute (rpm), of the pump and sending the corresponding command to the pump controller. The control algorithms are implemented through the LabView software (National Instruments, USA), used to acquire and store data from both implantable sensors and the pump and to process these data for controlling the LVAD functionalities. Even using the LabView software, a Human-Machine-Interface was developed to allow the physician a simpler and better managing of the pump. In particular, through this dedicated interface the physician is able to choose a proper level of perfusion and to set the corresponding flow value. According to this value, the implemented control algorithms automatically send to the pump controller the proper command to change the rpm and check the output of the flow sensor. The target value is reached thanks to a PI control that dynamically and precisely changes the pump speed in order to achieve the desired flow value. This control method has been implemented because it exhibits a fast, fine and stable attainment and maintenance of the desired flow. Concerning the threshold control on pressure, the main goal is to monitor if high, fast and/or repetitive changes in the pressure values occur. Indeed, these changes could be the signal for adverse and undesired events (such as certain pathological events and/or LVAD malfunctioning) that could negatively affect the patient condition and eventually provoke thrombus formation [text deleted]. To avoid this problem, the threshold control on drop pressure variations resets the pump if pressure oscillations are detected. It is worth mentioning that, even though these algorithms are developed and tested on a dedicated on-bench platform, they could be intended as a general methodology for monitoring and controlling continuous-flow LVADs. Indeed the final ARU will be a flexible device that could be easily set for different cardiac pumps and sensors, thus ensuring a feedback control.
A. Experimental Setup: The On-Bench Platform An on-bench platform has been designed and exploited in order to implement and consecutively assess the suitability and feasibility of the proposed control algorithms.
662
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
cular pathological conditions will be exploited to perform further assessment. B. Control Design
Fig. 2. (Left) On-bench platform setup and (right) overall schematics.
The platform includes an LVAD, embedded pressure and flow sensors and the conditioning electronics, which are physically connected to the PCs (Fig. 2). The core of the integrated test-bench is the Synergy™ Device, a VAD pump provided by CircuLite (Aachen, Germany) with a maximum blood-flow of 4.25 liters per minute. A wired electronic controller allows a manual setting of the pump when the controls are off. The pump operates in a closed hydraulic circuit that has been set up by using a silicone tube with an internal diameter of inches (9.52 mm) and a wall thickness of inches (1.59 mm). The hydraulic circuit is connected to a tank containing the fluid (composed of a solution of 48.3 % glycerin). This solution has been selected in order to reproduce the blood viscosity and to prevent damage to the pump. The platform has been equipped with commercial pressure and flow sensors. In particular, two catheters with piezoresistive tips (Mikro-Tip® by the Millar Instruments, www.millar. com) have been positioned at the pump input and output, respectively. These sensors were employed because they are suitable to be used in the implantable platform. Additionally, they present high linearity and sensitivity (up to 5 V/V/mmHg) and can measure pressures ranging from mmHg to mmHg (from to 40 kPa). This pressure range is compatible with the range of interest for the application, typically from 40 to 200 mmHg. In order to measure the flow, a non-invasive clamp-on transducer provided by em-tec (Finning, Germany) based on the principle of the transit time difference from ultrasonic acoustic waves in up and downstream direction, was integrated. This ultrasonic flowmeter is characterized by an excellent accuracy and stability; moreover it has been calibrated for flow measurement in Polyvinyl Chloride (PVC) and silicone tubes. The transducer is able to measure a maximum flow of 10000 ml/min, which matches the application flow range of 3000–6000 ml/min. In the final setup the pressure sensors were located at the input and output of the pump at 1 mm distance, whereas the flowmeter was positioned 15 cm downstream from the pump. The presented experimental setup has been exploited with the aim to implement and assess the control algorithms since it allows for easy simulation of pathological events such as vasoconstriction (i.e., hypertension or suction), achieved by manual incrementing or decrementing the peripheral resistance through a manual action on the hydraulic circuit (e.g., by squeezing the silicon tube). However, a more advanced cardiovascular simulator [20] allowing simulation of a higher set of cardiovas-
The two control algorithms have been implemented using the LabView development environment resulting in a simple controller and a user-friendly interface. In order to decrease the overall computational load, the LabView interfaces, respectively developed for acquiring data and controlling the LVAD, run on two different dedicated PCs. For this purpose, the hardware components of the platform are connected to the first PC, named “Acquiring PC”, which acquires and stores on-line data. On the other hand, the algorithms devoted to process these data and to implement the commands to be sent to the pump controller run on the second PC, named “Control PC”. The logic work flow that has been used to implement the control unit and all the communications are graphically represented in Fig. 3. In particular, the PC-based platform communicates with the hardware module composing the on-bench platform through serial communication lines and the National Instruments data acquisition (DAQ) board. The Acquiring PC continuously acquires, visualizes and stores the data. In particular, the data acquired by the platform are pump speed and current, flow and pressure parameters. Pump speed and current are directly taken from the pump controller and transferred to the control board by a serial communication line for checking the device status. Similarly, the flow data from the ultrasonic transducer are transferred to the PC by a second serial communication line. On the other hand, pressure data are acquired using the DAQ board, transferred to the PC via USB and acquired through the LabView dedicated functions. All the data are visualized and stored in the Acquiring PC thanks to the dedicated interface exploiting the National Instrument VISA and DAQmx palettes. These data are constantly collected and converted into format strings in order to be transmitted to the Controlling PC. At every running cycle, the latter continuously controls pressure and flow values. In particular, regarding the flow control, the algorithm checks, at every cycle, if a new desired flow value is set in the interface. When a new value is inserted, the system compares it with the current flow value and discerns two different conditions: 1) the desired flow is higher than the actual read one; 2) the desired flow is lower than the actual read one. A PI control was selected in order to obtain the increment or decrement needed for reaching the desired flow, and directly change the pump speed. The theoretic basis of the PI control are reported in [21]. The PI control was implemented in LabView with the transfer function form according to the following law (1): (1) where and represent, respectively, the proportional gain and the integral time. A manual tuning was used to define their proper values. U(s) represents the Laplace transform of the controller output and E(s) the Laplace transform of the inlet. In our implementation E(s) is calculated depending on the situation deas the detected by the case structure. In fact, considering
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
663
Fig. 3. Logical work flow of the control strategy.
Fig. 4. Experimental data and linear fitting of flow versus rpm of the on-bench platform.
sired flow value inserted by the clinician and as the actual read flow value: when is 1) E(s) is the Laplace transform of higher than ; when is 2) E(s) is the Laplace transform of . higher than In order to minimize the difference between flow values, the PI control will return an output U(s), which is the quantity in rpm that must be added or subtracted (according to the calibration curve reported in Fig. 4) to the actual value. As explained above, the activation of the PI control depends on the insertion of a flow value into the interface. Conversely, the control of pressure is an automatic control, continuously active for every running cycle. A threshold control on the standard deviation of the average pressure drop values is implemented. More in detail, a dynamic array is assembled and updated as soon as a new pressure value is available, and its standard deviis calculated according to the following law (2): ation (2) in our case) and is where n is the number of samples ( their mean value. The number of samples of the dynamic array has been set to 100 to enable quick responses to sudden changes
related to adverse and undesired events while remaining unaffected by slower physiological changes. The standard deviation of the dynamic array is continuously monitored and compared to a pre-defined threshold. In order to define an accurate and suitable threshold value for the implemented pump, a specific test was carried out keeping the pump in a steady condition (i.e., no suction or unstable events occurring). The standard deviation in these conditions has been evaluated in order to define the pressure variations during normal operation of the pump. Then, the same test was carried out simulating a suction event to define the typical standard deviation in this case. By these measurements it was possible to define a proper threshold which must be higher than the one associated with the normal working condition of the pump and, at the same time, lower with respect to the one measured during a suction event. The typical value of standard deviation to foresee a suction event has been experimentally assessed. In working conditions, when the measured standard deviation exceeds the set threshold, the algorithm processes the command string to reset the pump. After the reset command is sent, the control unit waits a minimum amount of time for the system to reach the normal conditions (i.e., standard deviations below the threshold). Every time a control algorithm is activated, the resulting command must be primarily transmitted via the serial communication line from the Controlling PC to the Acquiring PC and the latter communicates via another serial line with the pump controller. Furthermore, the system is able to produce alert signals if one of the monitored parameters (flow, pressure and rpm) exceeds its specified range. In fact, to meet safety requirements, alarms based on threshold controls have been implemented. It is worth noting that, if the control units stop working properly, the pump is able to guarantee a constant blood perfusion to the patient. C. Human-Machine-Interface Two simple, intuitive and user-friendly Human-Machine-Interfaces (HMIs) were implemented with the LabView software, respectively, for the Acquiring and Controlling PCs. The aim of the HMIs is to help the final user to simply and efficiently control the LVAD. The Acquiring PC interface allows the user to
664
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
Fig. 5. Human-Machine-Interface.
Fig. 7. Results from the PI controller; red arrows indicate the activation of PI controller. (Top) Flow increment from 3500 ml/min to the setting value of 5500 ml/min. (Bottom) Corresponding increment of rpm from 20000 to 30000.
Fig. 6. Results from the PI controller; red arrows indicate the activation of PI controller. (Top) Flow decrement from 5500 ml/min to the setting value of 3500 ml/min. (Bottom) Corresponding decrement of rpm from 30000 to 20000.
visualize the parameters in real-time. The first page of the Tab Control in the Front Panel, named “Read” allows the user to read the numerical values of the acquired data (speed, current, flow and pressure) and to observe their temporal trend through their
respective graphs. Two string indicators are also represented in order to check the correctness of the data transmitted to the Controlling PC and of the command activating the PI control transmitted from the Controlling PC. The second page of the Tab control, named “Setting”, allows the user to set all the parameters needed for the communication between the Acquiring PC and the hardware components of the platform. The Controlling PC interface is the most critical and important, since it allows the user to insert the desired flow value, thus activating the PI control simply by pressing a button. At the same time, this interface is provided with a LED which is switched on if the pressure variation exceeds the safety threshold, indicating that the reset control is activated. All these functionalities are implemented in the first page of Tab Control, named “Control”. Even in this page a button is present enabling the reading of data transmitted by the Acquiring PC. The numeric values of these data (rpm, flow and pressure) are also shown. A “Setting” page allows the user to properly set the communication parameters between the two PCs. The developed HMIs are shown in Fig. 5.
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
Fig. 8. PI controller activated; red arrows indicate the squeeze of the silicon tube (peripheral resistance increase) and black arrows indicate the release of the silicon tube (peripheral resistance decrease). (Top) Flow is kept constant at the set value of 4000 ml/min. (Bottom) Corresponding increment of rpm.
III. RESULTS The on-bench platform has been exploited in order to assess the reliability and consistency of the control algorithms for monitoring and controlling the LVAD parameters. The PI control and the pressure threshold control methods were implemented and tested on bench; the performed experiments are described in the following sections. A. PI Control Based on the Flow Sensor The behavior of the PI control was analyzed in two experimental phases using the dedicated on-bench platform. During the first phase, the pump speed was set at 30000 rpm. As shown in Figs. 4 and 6, the corresponding flow value in this condition is 5500 ml/min. In order to activate the PI control, a desired flow value of 3500 ml/min was set through the HMI of the Controlling PC. Fig. 6 shows clearly the effects of the PI control in redirecting the flow to the desired value, by adjusting the pump speed. Indeed, as soon as a speed of 20000 rpm is set according to the implemented algorithms, the desired flow
665
Fig. 9. PI controller deactivate. (Top) Flow decreases after the silicon tube is squeezed and goes up to set value of 4000 ml/min only after the tube is released. (Bottom) Pump speed stays constant.
value of 3500 ml/min is reached. In the second experimental phase, the starting speed was 20000 rpm (corresponding to a flow value of 3500 ml/min). As in the previous case, a desired flow value of 5500 ml/min was set. The flow and speed trends in the case of flow increment are shown in Fig. 7. In both experiments the PIcontrol has demonstrated to work properly for adjusting the flow to the desired value with a suitable response time. After a small overshoot/over damping (below a threshold of ml/min), the flow goes back below a threshold of ml/min in about 20 sec These values are compatible with medical specifications, also considering that the patient’s heart is always active and able to support the circulatory system for a certain time (up to 60–120 min). A further assessment of the suitability of the implemented PI control was carried out with the aim to demonstrate its ability to face the physiological changes that can occur during the daily life (e.g., temperature increase and chest compression). The control had to maintain constant blood perfusion despite changes in the patient’s condition. In Fig. 8 a peripheral resistance change was simulated to show the effect of the PI control in this situation. The flow value was
666
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
Fig. 10. Suction event with no control activation; (top) effect on pressure and (bottom) flow, respectively.
set to 4000 ml/min and the peripheral resistance was consecutively increased twice by squeezing the silicon tube. As shown in Fig. 8 (bottom), the system reacted to this condition by increasing the pump speed in order to maintain constant perfusion. Fig. 8 (top) demonstrates that the flow was recovered to the desired value of 4000 ml/min within about 15 s and maintained constant until the tube was released. In Fig. 9 the silicon tube was squeezed three times without the activation of the PI control. As shown, no changes in pump speed occurred and the flow value returned to 4000 ml/min only when the tube was released. B. Control on Pressure Sensor A further experimental session was carried out to assess the performance of the algorithms for controlling the pressure variations within a pre-set range (i.e., a threshold). In order to fulfill this task, the silicon tube was squeezed to simulate undesired events (e.g., suction or hypertension). Fig. 10 shows the effect of two consecutive sudden events on both pressure and flow. We can observe that, as soon as an event occurs, the flow quickly drops while the pressure increases. If no control is active this condition continues until the event ends. In Fig. 11 the results of the experiment with the activation of the control are reported.
Fig. 11. Suction event with control activation; (top) effect on pressure and (bottom) flow, respectively.
Thanks to the control on pressure thresholds, when an event is detected (i.e., the pressure value exceeds the threshold) the pump is reset. This produces a sudden drop in the pressure, but is rapidly recovered (in 2 sec) to the values acquired before the event occurred. IV. CONCLUSION In this work, an innovative control strategy for controlling and modeling LVADs has been developed. In particular, two different control algorithms have been implemented: a proportional-integrative and a threshold control methods based on flow and pressure values, respectively. Furthermore, a HMI has been developed to allow for a simpler and better management of the pump by the specialist. The control unit has been experimentally evaluated exploiting a dedicated on-bench platform embedding an LVAD and implantable sensors for flow and pressure. It has been demonstrated that the algorithms are able to control the LVAD in a feedback-loop directly based upon sensors measurements. In particular, the control algorithms showed an adaptive behavior to the new condition in about 2 sec and can suddenly change the LVAD speed as well.
VERBENI et al.: AN INNOVATIVE ADAPTIVE CONTROL STRATEGY FOR VADs
ACKNOWLEDGMENT The authors would like to thank CircuLite GmbH (www.circulite.net) and em-tec GmbH (www.em-tec.com) for providing the pump and the flow sensor.
REFERENCES [1] A. H. H. AlOmari et al., “Developments in control systems for rotary left ventricular assist devices for heart failure patients: A review,” Phys. Meas., vol. 34, pp. R1–R27, Jan. 2013. [2] M. Strüber et al., “The current status of heart transplantation and the development of” artificial heart systems,” Dtsch Arztebl Int., vol. 106, pp. 471–477, 2009. [3] M. E. DeBakey, “Left ventricular bypass pump for cardiac assistance. Clinical experience,” Amer. J. Cardiol., vol. 27, pp. 3–11, Jan. 1971, 1971. [4] G. S. Kumpati et al., “Left ventricular assist device bridge to recovery: A review of the current status,” Ann. Thoracic Surgery, vol. 71, pp. S103–S108, Mar. 2001. [5] Ž. Sutlić et al., “Overview of new mechanical circulatory support devices,” Rad Hrvatske Akademije Znanosti i Umjetnosti. Medicinske Znanosti, vol. 509, pp. 123–132, 2011. [6] C. A. Milano and A. A. Simeone, “Mechanical circulatory support: Devices, outcomes and complications,” Heart Failure Rev., vol. 18, pp. 35–53, Jan. 2013. [7] S. Salzberg et al., “Left ventricular assist device as bridge to heart transplantation—Lessons learned with the MicroMed DeBakey axial blood flow pump,” Eur. J. Cardio-Thoracic Surgery, vol. 24, pp. 113–118, July 2003. [8] S. Westaby et al., “Destination therapy with a rotary blood pump and novel power delivery,” Eur. J. Cardio-Thoracic Surgery, vol. 37, pp. 350–356, Feb. 2010. [9] D. Esmore et al., “A prospective, multicenter trial of the VentrAssist left ventricular assist device for bridge to transplant: Safety and efficacy,” J. Heart Lung Transplant., vol. 27, pp. 579–588, Jun. 2008. [10] M. Strueber et al., “Multicenter evaluation of an intrapericardial left ventricular assist system,” J. Amer. College Cardiol., vol. 57, pp. 1375–1382, Mar. 22, 2011. [11] T. Hrobowski and D. E. Lanfear, “Ventricular assist devices: Is destination therapy a viable alternative in the non-transplant candidate?,” Current Heart Failure Rep., vol. 10, pp. 101–107, Mar. 2013, 2013. [12] E. A. Rose et al., “Long-term use of a left ventricular assist device for end-stage heart failure—N Engl J Med 2001;345 : 1435–1443,” J. Cardiac Failure, vol. 8, pp. 59–60, Apr. 2002. [13] D. Goldstein et al., “Comparison of costs, readmissions and days out of hospital at one year between heart transplantation and continuous flow LVAD,” J. Heart Lung Transplant., vol. 32, p. S98, 2013. [14] M. S. Slaughter et al., “Advanced heart failure treated with continuous-flow left ventricular assist device,” New Eng. J. Med., vol. 361, pp. 2241–2251, 2009. [15] E. J. Birks et al., “Left ventricular assist device and drug therapy for the reversal of heart failure,” New Engl. J. Med., vol. 355, pp. 1873–1884, 2006. [16] C. A. Giridharan et al., “Modeling and control of a brushless DC axial flow ventricular assist device,” Asaio J., vol. 48, pp. 272–289, May–Jun. 2002. [17] T. Mussivand et al., “Wireless monitoring and control for implantable rotary blood pumps,” Artif. Organs, vol. 21, pp. 661–664, 1997. [18] S. Hübler et al., “Development of a database of patients supported by ventricular assist devices,” Asaio J., vol. 49, pp. 340–344, 2003. [19] R. Kosaka et al., “Tsukuba remote monitoring system for continuous-flow artificial heart,” Artif. Organs, vol. 27, pp. 897–906, 2003. [20] M. Kozarski et al., “A hybrid (hydro-numerical) cardiovascular model: Application to investigate continuous-flow pump assistance effect,” Biocybern. Biomed. Eng., vol. 32, pp. 77–91, 2012, 2012. [21] Modern Control Engineering, K. Ogata and Y. Yang, Eds. Upper Saddle River, NJ, USA: Prentice Hall, 1990.
667
Antonella Verbeni was born in Reggio Calabria, Italy, in 1986. She received her Master’s degree in biomedical engineering from the University of Pisa, Pisa, Italy, in 2011. Currently, she is working toward the Ph.D. degree in biorobotics at The BioRobotics Institute of the Scuola Superiore Sant’Anna, Pisa, Italy. In September 2010, she joined the Scuola Superiore Sant’Anna, working on surgical robotics and focusing her activity on endoscopic capsules. Her main research interests include innovative devices for cardiovascular disease with particular attention on the study and the application of ultrasound.
Rossella Fontana was born in Agrigento, Italy, in 1986. She received the Master’s degree in biomedical engineering from the University of Pisa, Pisa, Italy, in May 2011. Currently, she is working toward the Ph.D. degree at The BioRobotics Institute of Scuola Superiore Sant’Anna, Pisa, Italy. She is working on the development of sensorized devices for implantable cardiovascular applications.
Michele Silvestri was born in Pisa, Italy, in 1984. He received the Master’s degree in biomedical engineering (cum laude) from the University of Pisa, Pisa, Italy, in 2010. Currently, he is working toward the Ph.D. degree in biorobotics at The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy. His current research interests include biomechatronics and medical imaging.
Giuseppe Tortora (S’09–M’13) was born in Isernia, Italy, in 1983. He received the Master’s degree in biomedical engineering from the University of Pisa, Pisa, Italy, and the Ph.D. degree in biorobotics from The BioRobotics Institute of Scuola Superiore Sant’Anna, Pisa, Italy, in 2008 and 2012, respectively. In April 2007 he joined the Scuola Superiore Sant’Anna, focusing his activity on medical robotics and biomechatronic systems. He spent a period as Visiting Researcher at Imperial College London, London, U.K., and Carnegie Mellon University, Pittsburgh, PA, USA, in 2011. His main research interests are in the field of biorobotics and advanced systems for minimally invasive diagnosis, surgery, and therapy.
Monica Vatteroni was born in La Spezia, Italy, in 1975. She received the M.S. degree in electrical engineering from the University of Pisa, Pisa, Italy, and the Ph.D. degree in physics from the University of Trento, Trento, Italy, in 2001 and 2008, respectively. Currently, she is with the Scuola Superiore Sant’Anna, Pisa, Italy as a Postdoctoral Fellow, where she is responsible for research and development of image sensors, vision systems, and sensorized biomedical platforms.
668
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 8, NO. 5, OCTOBER 2014
Maria Giovanna Trivella received the medical doctor degree from the University of Pisa, Pisa, Italy, in 1976 and majored in cardiology in 1981. Currently, she is a Senior Investigator of the Italian National Research Council (CNR), Head of the Experimental Laboratory of IFC-CNR, and Head of Experimental Surgery Laboratory, a joint laboratory in cooperation between CNR and the University of Pisa. She is the coordinator of SensorART—a remote controlled sensorized artificial heart enabling patients empowerment and new therapy approaches (FP7-ICT-248763), 2010–2014. Her current research fields of interest are interdisciplinary and translational research in medicine. Dr. Trivella was awarded by the European Science Foundation for the Exploratory Workshop, “Molecular signaling in cardiovascular and oncological disease: Similar and shared pathways,” in 2007.
Paolo Dario (F’02) received the Master’s degree in mechanical engineering from the University of Pisa, Pisa, Italy, in 1977. He is Professor of Biomedical Robotics at the Scuola Superiore Sant’Anna, Pisa, Italy, where he supervises a team of approximately 150 young researchers. He has the role of Scientific-Technical manager of SensorART. His main research interest is biorobotics, including mechatronic and robotic systems for rehabilitation, prosthetics, surgery, and microendoscopy. He is the author of more than 160 ISI journal papers, many international patents, and several book chapters on medical robotics. He was the recipient of the Joseph Engelberger Award as a Pioneer of Biomedical Robotics.
