An EMFi-film Sensor based Ballistocardiographic Chair: Performance and Cycle Extraction Method

Transcription

1 An EMFi-film Sensor based Ballistocardiographic Chair: Performance and Cycle Extraction Method Sakari Junnila, Alireza Akhbardeh, and Alpo Värri Tampere University of Technology Institute of Signal Processing Tampere, Finland Teemu Koivistoinen Tampere University Hospital Department of Clinical Physiology Tampere, Finland Abstract New sensor technologies open possibilities for measuring traditional biosignals in new innovative ways. This, together with the development of signal processing systems and their computing power, can sometimes give new life to old measurement techniques. Ballistocardiogram is one such technique, originally promising but quickly replaced by the now very popular electrocardiogram. A ballistocardiograph chair, designed to look like a normal office chair, was built and fitted with pressure sensitive EMFi-films. The films are connected via a charge to a medical bio. The system was accepted for medical use in Tampere University Hospital and patient measurements have been performed. The system is presented and it s performance evaluated. A wireless version of the system is needed to hide the cabling from the user. This will make the chair indistinguishable from a normal office chair. Overview of first wireless prototype is given. To analyze recorded BCG, individual BCG cycles must be extracted from the signal containing respiration and movement artifacts. A method for this and results of it s application are presented. The developed system can be used for BCG measurements and it is able to automatically extract individual BCG cycles, but it has some limitations which are presented in the paper. I. INTRODUCTION New sensors offer the possibility to monitor some physiological variables from humans quite unobtrusively. One such variable is the ballistocardiogram (BCG) [1]. BCG is a noninvasive technique for the assessment of the cardiac function. BCG consists mainly of heart movement and the movement of blood in aorta, arteries, and periphery. With the development of signal processing algorithms, the ballistocardiography is again becoming interesting for researchers. In our research, we aim to develop a normal-looking chair fitted with sensitive sensors for BCG measurement, a wireless data link to transfer recorded signal forward for analysis, and signal processing algorithms for automated BCG analysis. In this paper, we used EMFi-film 1 sensors to measure BCG from body movements in sitting position. One sensor is installed in the seat and second one in the backrest. Only the seat sensor recordings are used in this paper. The hardware used for measurements is presented in this paper. The performance of the hardware is evaluated, and technical parameters such as frequency response are presented. Also, 1 EMFi is a registered trademark of Emfit Ltd. an overview of the architecture for the wireless system is given. Three major interference sources have been identified; movement of the patient, respiration, and electrical noise. The noise can usually be removed with lowpass filtering, but the movement and respiration of the patient make identification of BCG cycles difficult. Usually electrocardiogram (ECG) and its R-spike are used as a reference to extract BCG cycles from the recorded signal. We present a blind segmentation method to extract BCG cycles per cardiac period without using ECG signal from the subject as reference. This greatly enhances the user-friendliness of the measurement process compared to a system using ECG as reference, as there is no need for physical contact with the patients body surfaces. It also provides additional benefits for hardware and device designers regarding the electrical safety issues. Section II gives short introduction to BCG measurement. The wired BCG recording system, first presented in [2], is described in Section III. This wired setup was used to perform patient recordings to gather data for algorithm development, and first results of the recordings were presented in [3], [4]. The technical performance of the wired system was evaluated and figures showing system performance are given. Next, the technology choices for the wireless implementation and the layout for the proposed wireless system are presented in Section IV. Section V presents a novel blind segmentation method for BCG classification, which was developed using data recorded with the wired system of Section III. Finally, conclusions are given. A. Sensor technology related work EMFi-film has been used to detect heart rate and breathing by Technical Research Centre of Finland (VTT) and some commercial companies. To best of our knowledge, there is currently no other published work done outside Tampere University of Technology on detailed BCG recording and analysis using the EMFi-film. During the Tazawa et al. from Muroran Institute of Technology, Japan, researched avian embryos, and used various technologies to measure the BCG. They also used piezoelectric film in one of their publications [5]. Other recent publications on BCG measurement have used piezoelectric transducers and accelerometer sensors /05/$ IEEE SIPS

2 B. BCG analysis related work BCG is not a new concept, many publications were written about it between 1940 and mid-1970 s. During the past several years some recording and monitoring methods as well as pattern recognition methods for BCG have been developed as reported in [6], [7]. Also, the authors have done some research work in BCG pattern recognition [8], [9]. Almost all of these studies use the R-Component of Electrocardiogram (ECG) as the reference signal. For medical doctors the visual evaluation of BCG signal and its temporal components is the traditional method of BCG analysis. In this field of studies a lot of research work has been done by medical experts and clinicians. II. BCG MEASUREMENT Ballistocardiogram is generated by movement of the heart and blood. Thus, a sensor measuring body movements is required. In our implementation, we use pressure sensitive EMFi-films to measure movement in two separate axes in sitting position. Sitting position is used instead of traditional supine (lying) position because it is more natural position for a person during daily activities and measurements can more easily be done in that position, even without the patients notice. A measurement chair also consumes less space and is easier to transport than a bed type measurement device. It is a generally accepted fact that stress and anxiety caused by a doctor performing medical measurements on a person often affects the measurement results. An increase in heart rate, for example, is very common. Therefore, a BCG measurement performed unnoticed by the patient or at home in familiar surroundings should produce better recordings. This is possible with our system. Currently, due to ethical reasons, we still have to inform the patient when he/she is being measured. EMFi-film, used as the sensor element, is elastic permanently charged electret film material produced by Emfit Ltd ( [10] [12]. The film has a permanent charge that changes when pressure is applied to the film. Film is also pyroelectric, e.g., sensitive to temperature changes, but in a relatively stabile environment (indoor office) the effect of temperature is very small compared to the pressure sensitivity. A special type of current, e.g., a charge is needed to measure the charge changes of the EMFi-film and produce an amplified voltage signal output. III. WIRED IMPLEMENTATION For initial recordings and algorithm development, we first implemented a wired version of our measurement system [2]. In the wired version, EMFi-films mounted on the chair are connected with wires to a external charge device. The charge is then connected to CircMon circulation monitor using CircMon s external device interface connectors. With this setup (Fig. 1), CircMon is able to measure, display and record ECG, impedance cardiogram (ICG) and BCG simultaneously. CircMon is connected to a PC laptop computer, which can store data to various media for algorithm developers. All figures and results presented in this paper are Chair Pre CircMon device + Laptop PC Fig. 1. Man seated in the chair EMFi film sensor installed under the upholstery Charge Low pass filter Anti aliasing filter ADC Digital filtering Monitoring and storage to database Wired measurement setup. ECG and ICG electrodes attached to the subject EMFi film CircMon based on recordings done with this wired setup. Results of first recordings have been published in [3], [4], [8], [9]. A. System performance The system consists of two separate devices (Fig. 1). The evaluation of CircMon was straightforward. We measured the frequency response of CircMon using white noise input and point measurements with discrete sinusoidal signals. Both methods produced the same results so only figures obtained with the point measurements are presented. The frequency response of the CircMon device from 0 to 120 Hz is presented in Fig. 2. The presence of a 50 Hz notch filter can be seen from the figure, and its presence was confirmed by CircMon manufacturer, JR Medical Ltd. The notch filter is implemented as a digital filter after AD-conversion. The linearity of the CircMon was also measured using 3, 10 and 20 Hz input signals. The results for both input channels (labelled AO and BP) can be seen in Fig. 3. The attenuation of the signals with 10 Vp-p input is listed in Table I. The internal noise was measured by short-circuiting the inputs of CircMon. Noise was found to be ± 1 bit. TABLE I ATTENUATION OF A 10 VP-P SINISUOIDAL INPUT SIGNAL 3Hz 10 Hz 20 Hz AO connector 0.21 db 0.65 db 2.18 db BP connector 0.19 db 0.61 db 2.05 db 374

3 Fig. 2. CircMon frequency response. Fig. 4. Frequency response of complete system. Fig. 3. CircMon linearity. Fig. 5. Complete system and charge linearity. The charge is designed to amplify a charge change signal. Simulation with voltage or current input produces a frequency response where 20 db/dec falling response equals flat response. We used a signal generator with a series 10 MΩ resistor, and scaled the results to obtain frequency response shown in Fig. 4. Because the has a huge gain for low-frequency signals and the signal generator was limited to minimum of 50 mvp-p output, the output of the charge saturates at low frequencies. The frequency response could therefore only be evaluated from 3 Hz forwards, and the sharp attenuation below 3 Hz does not represent true system behaviour. Other tests ans Spice simulations have shown that the charge has small attenuation between 0 to 1.5 Hz, but this does not greatly affect the BCG recordings, as the BCG signal components are mostly between 1.5 to 20 Hz, according to our experience. The linearity of the charge was also measured with and without CircMon. The results (Fig. 5, dashed lines) show that the is very linear with 3, 10 and 20 Hz test signals. The comparison to linearity measured with CircMon (Fig.5, fixed lines) shows again the attenuating behaviour of CircMon. IV. WIRELESS IMPLEMENTATION We have chosen the IEEE standard as a base for the wireless communication. IEEE is the lower level protocol used by Zigbee, and it is therefore possible to make the system Zigbee compatible in the future by adding more protocol layers. Initially, we will only use a partial implementation of the standard. We have also chosen to use Zigbee radio chips by ChipCon (CC2420). The first prototype system uses the charge developed for the wired system (Section III), which is connected to the internal AD-converter of the AVR microcontroller on the CC2420DB (included in ChipCon CC2420DBK kit) via a small level-converter circuitry. In our initial tests we have successfully sampled one BCG channel (8-bit resolution), and transmitted 4 bytes of data (8-bit BCG, 8-bit temperature and two 8-bit test signals) with approximately 250 Hz sampling rate. The RF-transmitter, even if placed next to the EMFi-film leads or the charge unit, does not generate such levels of interference that could be visually seen on the PC screen of the receiving device. The next step is to implement a +5 to 0 V output range version of the charge together with a low-power ECG. The ECG is implemented, as we intend to fit small ECG electrode plates to the armrests of the chair. This would provide us with a coarse ECG signal for R-spike position detection. Also, an external AD-converter is needed, because the 10-bit resolution offered by the internal ADC in the AVR is insufficient. In the final version of the hardware, we hope to implement our own mainboard with an AVR microcontroller and the RF-circuitry, s and AD-converter as in Fig. 6. The electronics will then be placed 375

4 Fig. 7. Typical raw and filtered BCG signals for a young healthy man. Fig. 9. Typical four BCG cycles of a young healthy man. Fig. 8. Typical raw and filtered BCG signals for an old man with a cardiac infarct in his past. Fig. 10. Typical four BCG cycles of an old man with a past cardiac infarct. within the chairs upholstery, so that it will look like a normal office chair. V. BLIND SEGMENTATION OF BCG DATA The recorded BCG signal consists of components attributable to cardiac activity, respiration, and body movements. To have a pure BCG signal and to remove additional components including background noises as well as respiration, we used a band pass filter with passband from 2 to 20 Hz. Figures 7 and 8 show raw and filtered BCG signal for a young healthy man and for an old man with cardiac infarct in his past. However, as can be seen especially in Fig. 8, body movements during recording cause loss of information and destroy individual BCG cycles. The parts of recorded BCG containing movement artifacts are useless and must be eliminated. For elimination, we used threshold values (3 and -3 V), using only signal between these values. This method was found useful as body movements generate stronger signal values than normal cardiac activities. To perform BCG data segmentation in a blind way, without using any other synchronization signal such as ECG, we extracted an additional coarse BCG signal using a narrow bandpass filter with 0.1 and 1 Hz -3 db cutting frequencies. The signal obtained after filtering the recorded BCG signal is coarse BCG signal riding on top of respiration signal. Then, we used absolute values of this coarse BCG signal and its peaks between 0.1 and 3 V as synchronization points. Peaks outside this range are usually not related to BCG cycles, being background noise or motion artifacts. Computed synchronization points of the coarse BCG signal are used to find central points of uniform windows with length of 250 samples (1.2 seconds). 125 samples before and 124 samples after these central points are taken from the bandbass filtered (2 to 20 Hz) BCG signal to create BCG cycles with the same lengths (1.2 seconds for every cycle). Figures 9 and 10 show extracted cycles for one typical healthy man and for one typical man with a past heart infarct. 376

5 EMFi films ECG electrodes Electronics inside the chairs upholstery Charge ECG A/D converter Atmel uc RF module (ChipCon CC2420) SRAM memory Desktop PC for the doctor/nurse Fig. 6. Final wireless setup. VI. CONCLUSIONS In this paper, we presented a wired measurement setup with an innovative sensor system for measuring, displaying and recording BCG signals. The performance of the system was evaluated. The system suffers from attenuation of signal components near 50 Hz, due to a 50 Hz notch filter in one of the devices. However, the frequency response between 1.5 to 20 Hz containing the most important BCG components is relatively flat, and the system in linear. The wired system has been approved for hospital use in Tampere University Hospital (TAYS), and has been used to record BCG from different patient groups. The wireless version based on IEEE protocol and hardware is work in progress, with some promising early results. A method for extracting individual BCG cycles from the recorded signal was presented. The method was implemented in Matlab and BCG cycles were successfully extracted from recordings done with our wired measurement setup. The extracted cycles can be used for visual evaluation of the patients health and heart condition by medical experts or fed forward to automated analysis. Automated analysis tools are currently being developed within the research group. ACKNOWLEDGMENT This study was financially supported by the Academy of Finland, Proactive Information Technology Program , and the Finnish Centre of Excellence Program The authors would like to thank the whole research group and particularly Dr. Tiit Kööbi and Dr. Väinö Turjanmaa for their involvement in the development of the measurement system and organizing the test group recordings. We also thank Ms. Marjaana Ylhäinen and Mrs. Pirjo Järventausta for carrying out the test measurements and helping in system performance measurements, and all the test subjects for their participation. REFERENCES [1] D.C. Deuchar, "Ballistocardiography," Brit. Heart Journal, Vol. 29, 1967, p [2] S. Junnila, T. Koivistoinen, T. Kööbi, J. Niittylahti, and A. Värri, "A Simple Method for Measuring and Recording Ballistocardiogram," in Proc. 17th Int. EURASIP Conference BIOSIGNAL 2004, Brno, Czech Republic, Jun. 2004, pp [3] T. Koivistoinen, S. Junnila, A. Värri, and T. Kööbi, "A new method for measuring the ballistocardiogram using EMFi sensors in a normal chair," Proc. 26th Annual Internatioal Conference of the IEEE Engineering in Medicine and Biology society, San Francisco, California, USA, 1-5 September 2004, pp. 4 p, (2004) [4] M. Koivuluoma, L. C. Barna and A. Värri, "Signal processing in ProHe- Mon Project: Objectives and First Results," Proc. Proactive computing workshop PROW 2004, Helsinki, Finland, 2004, pp [5] H. Tazawa, Y. Hashimoto, M. Takami, Y. Yufu, and G.C. Whittow, "Simple, noninvasive system for measuring the heart rate of avian embryos and hatchlings by means of a piezoelectric film," Med Biol Eng Comput., 31(2), 1993, pp [6] X. Yu, D. Dent, "Neural networks in ballistocardiography (BCG) using FPGAs, Software Support and CAD Techniques for FPGAs," IEE Colloquium on, Apr. 1994, pp. 7/1 7/5. [7] B.H. Jansen, B.H. Larson, K. Shankar, "Monitoring of the ballistocardiogram with the static charge sensitive bed," IEEE Trans. on Biomedical Engineering, Vol. 38, Issue 8, Aug. 1991, pp [8] A. Akhbardeh, M. Koivuluoma, T. Koivistoinen, and A. Värri, "BCG Data Discrimination using Neuro Systems and Daubechies compactly supported Wavelet Transform towards Heart Disease Diagnosing," Proc International Symposium on Intelligent Control 13th Mediterranean Conference on Control and Automation, June 2005, Limassol, Cyprus. [9] A. Akhbardeh, M. Koivuluoma, T. Koivistoinen, and A. Värri, "BCG Data Clustering using New Method so-called Time-frequency Moments Singular Value Decomposition (TFM-SVD) and Artificial Neural Networks," Proc. International Symposium on Intelligent Control 13th Mediterranean Conference on Control and Automation, June 2005, Limassol, Cyprus. [10] K. Kirjavainen, "Electromechanical film and procedure for manufacturing same," U.S. Patent No , [11] J. Lekkala, M. Paajanen, "EMFi-New electret material for sensors and actuators," IEEE ISE, 1999, pp [12] M. Paajanen, J. Lekkala, and H. Välimäki, "Electromechanical Modeling and Properties of the Electret Film EMFI," IEEE Trans. on Dielectrics and Elec. Insulation, 8;2001, pp