A WIRELESS PORTABLE ELECTROCARDIOGRAM MONITORING SYSTEM FOR CONTINUOUS CARDIAC ACTIVITY RECORDING

Transcription

1 Proceedings of the Third IASTED International Conference Telehealth and Assistive Technology (TAT 2016) October 6-7, 2016 Zurich, Switzerland A WIRELESS PORTABLE ELECTROCARDIOGRAM MONITORING SYSTEM FOR CONTINUOUS CARDIAC ACTIVITY RECORDING 1 Kampanart Preechawai, 2 Chatchai Neatpisarnvanit, 3 Tachapong Ngarmukos, 1 Phornphop Naiyanetr 1 Department of Biomedical Engineering, Faculty of Engineering, Mahidol University 2 Department of Electrical Engineering, Faculty of Engineering, Mahidol University 3 Department of Medicine, Faculty of Medicine Ramathibodi Hospital, Mahidol University 999 Phuttamonthon 4 Road, Salaya, Nakhon Pathom, 73170, Thailand Corresponding phornphop.nai@mahidol.ac.th ABSTRACT In this paper, a wireless portable electrocardiogram (ECG) monitoring system that suits for continuous cardiac activity recording is proposed. An analog front end (AFE) integrated circuit is the solution for biopotential recording. The microcontrollers are embedded with the program in order to control the wireless data transmission, and store data in the flash memory. To display and analyze the result, C# base software has been developed to communicate with the device via Bluetooth interface. A modified real time QRS detection algorithm with auto threshold method is implemented for calculating the heart rate in real time and offline application. The performance of developed prototype is evaluated in amplitude and heart rate accuracy with standard calibration equipment. To evaluate the QRS detection algorithm, standard 24-hr MIT/BIH normal sinus rhythm database are applied to compare with the result. The algorithm sensitivity and +P is 99.85% and 97.74%, respectively. This proposed system will be foundation for uncomplicated wireless portable ECG monitoring system in the future. KEY WORDS Electrocardiogram, Biopotential recording, Analog Front End (AFE), R-peak detection, System-on-Chip (SOC). 1. Introduction Nowadays, a wireless portable monitoring system is widely used in many medical-purpose applications such as Electrocardiography (ECG), Electromyography (EMG), etc. For research purpose, a portable monitoring system is very useful for collecting data. Different applications may require different controlled parameters. Therefore, the system with adjustable parameters configuration or parameters modification advantages over the commercial product. The design issues of portable ECG system with battery-powered are described in previous papers [1]. A main problem when using battery-powered is the quality of signal [2]. Due to the mismatch of impedances between patient s body and amplifier, the high common mode rejection ratio (CMRR) and low input voltage noise amplifier are essentially required [3]. Moreover, a low voltage application mostly requires single supply, causing low CMRR as a result. For achieving the high CMRR and solving the mismatch impedances, the Driven Right Leg (DRL) technique is able to reduce common mode voltage in ECG signal by driving potential into patient s body [4]. Currently, many analog front end (AFE) integrated circuits have the DRL feature to reduce a number of components on the circuit. A wireless monitoring system is widely described in many papers [5, 6]. There are variety types of wireless technology [7, 8]. Because of low power consumption, Bluetooth, one of the wireless technologies is applied. Another widespread technology is Radio Frequency (RF) or Wireless Local Area Network (WLAN) that has benefit over Bluetooth in transferring rate and signal distance. Some previous papers proposed wireless monitoring system which transfers data between portable device and smart phone via Bluetooth [9, 10]. Many papers proposed that the wireless system in clinical use can transfers the physiological information between devices by using WLAN. Some researchers propose ZigBee as the wireless solution [11, 12]. Due to the forward development of mobile network protocol, some papers propose the complex system which transfers data via mobile protocol such as GSM, WAT and HSPDA [13, 14]. A bigger scale of monitoring system causes the problem in both power consumption and signal stability. A correct detection of R-peak location is the important process for data analyzing. For real time processing, Pan and Tompkins [15, 16] proposed a QRS detection algorithm, that has the advantages of the uncomplicated processing and required minimum processing performance. The QRS detection algorithm is used for signal which contains interferences such as power-line interference, body movement, etc. For the better result, applying the automatic threshold to calculate each segment of data [17]. PhysioNet, a digital recording physiological database, is the accessible resource for researchers in biomedical community [18]. MIT/BIH Normal Sinus Rhythm database is one of database that the subjects have no significant arrhythmia. In this paper, a wireless portable ECG monitoring system that suits for continuous cardiac activity recording DOI: /P

2 is proposed. The system design is described in the next section, Methodology, including development of wireless portable ECG monitoring prototype and signal processing. After Methodology, the verification of the developed algorithm is described in evaluation part. 2. Methodology 2.1 Analog Front End (AFE) As a result of semiconductor development, the features of integrated circuits (ICs) have been improved for specific application. AFE is the ICs that composed of many instrumentation amplifiers and built-in circuit of DRL or Wilson Central Terminal (WCT) for ECG application. Thereby, the overall circuit dimension and power consumption can be significantly reduced. In this system, the commercial AFE module, ADS1298 (Texas Instrument), is used. ADS1298 has a 24-bit resolution for Analog to Digital Converter (ADC) with built-in programmable gain amplifier (PGA). It has eight channels of each independent instrumentation amplifiers as their positive and negative inputs. To acquire the 12-leads standard ECG directly from AFE, the author calculate by measuring eight leads ECG from ADS1298: two limb leads (Lead I, Lead II) and six chest leads (V1-V6). The residue leads are mathematically calculated from other leads. A calculation for the 12-leads establishing is processed in C# software. To control and receive the data from ADS1298, the microcontroller is used to interface AFE by SPI communication. In this system, a sampling rate is set at 500 samples/second with the high resolution mode (24-bit). two microcontrollers to do the different tasks. The first microcontroller manages the interface with AFE and Bluetooth module by using Serial Peripheral Interface (SPI), and Universal Synchronous/Asynchronous Receiver/Transmitter (USART) interface respectively. The second microcontroller does a task as the data logger application. It receives the data from another microcontroller via USART interface, and then converts the data and stamps the time by using internal Real Time Clock (RTC) registers. All data with time-stamp is saved in the flash memory via Secure Digital Input Output (SDIO) interface. A low power and high performance microcontroller in this system is STM32F401RB (STMicroelectronics). An operational voltage is 3.3V. Two microcontrollers share the same oscillator at 8 MHz. Both microcontrollers operate with the different clock frequency; because the frequency requirement in each interface is dissimilar. 2.4 Peripheral Interface A small size and low power Bluetooth module is one of the wireless solutions that suits for uncomplicated and low power application. In this system, there is RN42 (Microchip) that is a small form integrated the on-board PCB trace antenna. It is a cost-effective class 2 Bluetooth version. It also supports multiple interfaces. For Bluetooth module interface, a microcontroller transmits the AT commands to configure the module in the power-on duration via USART interface. To apply the SDIO/MMC interface feature in the microcontroller, the author use 16 GB class 4 micro SD card as a flash memory for a stand-alone system. 2.2 Preamplifier and Noise reduction To reduce the common-mode interference in ECG signal, a DRL circuit is implemented in the circuit by using the RLD features of AFE. The input of RLD amplifier can be controlled by AFE registers. In this system, the feedback is the WCT signal, generated from the WCT pin of AFE. This potential is the average potential from RA, LA, and LL electrodes, passing three built-in buffers in AFE. After that, WCT signal passes the external operational amplifier which is the component of RLD feedback loop. The output of RLD feedback loop connects to the RL electrode on patient s body. A guarding with the average of the input signals is applied in order to suppress the interference from the voltage difference between shield wire and inner core [19]. In this system, the average of the input signals is WCT signal from AFE. A guarding potential is limited by two resistors, connected as voltage divider for stability purpose. 2.3 Microcontroller For low voltage application, ARM-Cortex-M architecture microcontroller is selected to control the peripheral interfaces and process the acquired signal. This system has Figure 1. The system structure diagram 2.5 Signal Processing The software is developed to interface with the device base on C# programming language to demonstrate and analyze the data. The software has two processing modes; real time and offline. For real time processing mode, this software requires Bluetooth connection to the device. After receiving correct data package, a signal is filtered by Hz Band-pass FIR filter. For offline processing, this 19

3 software requires the recorded data in flash memory or the recorded data during real time processing mode. To detect QRS location, Pan and Tompkins QRS detection algorithm is implemented with auto threshold algorithm in this software [15-17]. A band pass filtering is used to maximize the QRS complex and suppress unwanted signal such as baseline wandering, P and T wave. In this system, a band pass FIR filtering for Hz is applied. A five point derivative is used to increase the slope of QRS complex (2.1). y(nt) = (1/8 T)[-x(nT-2T)-2x(nT-T) +2x(nT+T)+x(nT+2T)] where T is the sampling period (2.1) After derivative process, a signal is squared every point to make all data points to be positive value (2.2). y(nt) = [x(nt)] 2 (2.2) Moving window integration is applied to obtain the bell-shape waveform for QRS detection. In this system, the window width is 150 milliseconds. Therefore, a window size is 75 samples wide (2.3). y(nt) = ( 1 ) [x(nt-(n-1)t)+x(nt- N (N-2)T)+ +x(nt) ] where N is the number of window sample (2.3) The auto threshold algorithm is applied to calculate the threshold level for each segment of the signal for QRS area locating [17]. An equation for calculating the threshold value is described in equation (2.4). Threshold(n) = 0.39 max(n), if RMS(n) >A and A B { 0.39 max(n-1), if RMS(n)>A and A>B 1.6RMS(n), if RMS(n)<A where A = 0.18 max(n), B = 2max(n-1) (2.4) After locate the QRS area, the results have to be added with the delay that generates from previous process, and mapped to the preprocessing signal. A location of R point is determined by marking the maximum point of QRS area in each beat. To locate Q point, a minimum point of samples between onset and R point of that segment is determined. Likewise, a minimum point of samples between R point and offset of that segment is determined 5.5 mv with 0.5 mv incrementing at a time. Five amplitude samples are measured by counting the square box on ECG paper grid which display on the developed C# base software. For heart rate testing, the normal sinus rhythm beat, simulated from MPS450, is applied by varying beats per minute parameter with the controlled amplitude at 1 mv. A calibrated heart rate varies from 40 bpm to 160 bpm, 20 bpm of incrementing at a time. Developed C# software with R-peak detection algorithm is applied to analyze the RR interval for each beats. The heart rates are calculated from the average of RR interval in one minute. The amplitude testing and heart rate experiment are tested for five times each. In order to evaluate the performance, the standard deviation of average or uncertainty type A and percentage error are represented as the statistical result. 2.7 Evaluating QRS detection algorithm To evaluate performance of the QRS detection algorithm, the results from modified algorithm are compared with the annotation files of MIT/BIH Normal Sinus Rhythm database. The annotation files can be accessed on PhysioNet database [18]. It indicates the beat location and classifies the ECG beat that the technician observed. The thirty minutes of 18 records are analyzed with the modified algorithm to detect R-peak location and QRS complex. The sensitivity (Se) and positive predictive value (+P) are calculated. Sensitivity = +P = TP TP+FN TP TP+FP (2.5) (2.6) Where, True Positive (TP) is referred to the normal beat being taken correctly, and predicted that the R-peak can be found. False Positive (FP) is referred to the non-beat being taken wrongly as normal beat, and predicted that the R-peak can be found. False Negative (FN) is referred to the normal beat being taken wrongly as a non-beat, and predicted that the R-peak cannot be found. 3. Results and Discussion 3.1 Hardware Prototyping The prototype and 12-leads wires that used in this system are shown in Figure 2. The result of 12-lead ECG is shown in Figure Evaluating performance of prototype The amplitude and heart rate accuracy of the measured signal from the prototype is verified. For amplitude testing, a multi-parameter simulator, Fluke Biomedical MPS450, is used to generate the calibration signal, 10 Hz sine wave. The amplitude of calibration signal varies from 0.5 mv to 20

4 3.2 Signal Processing and Software prototype The resulted signal in each QRS detection process is shown in Figure 5. Developed software is based on C# programming for analysis and real time monitoring. The user interface is shown in Figure 6. Figure 2. A prototype of wireless portable ECG device Figure 5. (a) Raw signal from stress cycling test, (b) Filtered signal at Hz, (c) Filtered signal at 1035 Hz, (d) A result from QRS detection algorithm, (e) Rpeak detection result, and (f) QRS detection with annotation beat Figure 3. Result of ECG Lead I, II, III, avr, avl, avf, and V1 - V6 A real time monitoring via Bluetooth by testing with cycling experiment is shown in Figure 4. Figure 6. Software for analyzing ECG 3.3 Evaluating performance of prototype Table 1 lists the statistical results from calibration amplitude testing. All standard deviation of average for all calibration amplitude is equal to A maximum percentage error is 8% at the 0.5 mv of calibrated amplitude. A possible problem in this experiment is the error of measurement when counting square box on ECG paper. The statistical results from heart rate testing are presented in Table 3. The overall percentage errors are equal to 0% for all calibrated heart rate. These results Figure 4. (a) Stress Cycling test, (b) Real time monitoring mode on C# software, and (c) 12-lead electrode placement 21

5 confirm that the R-peak detection algorithm for normal sinus rhythm can process and calculate the heart rate excellently. Table 1 Statistical results from calibration amplitude testing Calibrated STDEV of Percentage Average STDEV Amplitude (mv) Average Error (%) Calibrated Heart Rate (bpm) Table 2 Statistical results from heart rate testing Average STDEV STDEV of Average Percentage Error (%) Evaluating the QRS detection algorithm Table 3 lists the QRS detection results of 18 records (44182 of normal beats) from MIT/BIH Normal Sinus Rhythm database. The overall sensitivity and +P are 99.85% and 97.74%, respectively. Table 3 Results of evaluating the QRS detection algorithm using MIT/BIH Normal Sinus Rhythm database Record Beats TP FP FN Se +P % 99.53% % 90.97% % 100% % 99.92% % 99.67% % 99.92% % 100% % 100% % 100% % 100% % 100% % 98.68% % 99.92% % 80.31% % 97.19% % 97.82% % 97.41% % 97.94% Total The error of the FP results are mostly mistreat ST change characteristics beat, non-beat in the detection as positive condition. Record number has ST change, baseline wandering, and noise characteristics which were classified as non-beat. Thereby, the number of FP is higher than other record. The error in FN results is beat missing, caused by low R-amplitude. 4. Conclusion and Future Work A wireless portable ECG monitoring system has been proposed in this paper. An analog front end integrated circuit is the solution to reduce the complication of bioamplifier circuit. To achieve a standalone purpose, two microcontrollers are embedded the program to interface the peripheral module. A cost-effective Bluetooth module is the solution for wireless interface. To receive and analyze the data, C# software has been developed. In order to evaluate the performance, the standard calibration equipment is applied to evaluate the amplitude and heart rate accuracy of prototype. The results of two evaluating performance are satisfied. A modified QRS detection algorithm with auto threshold method is implemented in this software to calculate the heart rate. The sensitivity and +P, comparing with MIT/BIH Normal Sinus Rhythm database are 99.85% and 97.74% respectively. The possible future work is the smartphone-based application. The classification algorithm can be applied to improve QRS detection performance in order to classify the abnormal beat. Acknowledgements This work was supported in part by the Health System Research Institute (HSRI and HSRI ). The author appreciates the support, and would like to acknowledge Phornphop Naiyanetr, Dr. scient. med. and Assoc. Prof. Chatchai Neatpisarnvanit, PhD for helpful discussion. Our lab members are also wished to thank for help in experiment. References [1] Jyoti Bali AN. Design Issues of Portable, Low-Power & High-Performance ECG Measuring System. International Journal of Engineering Science and Innovative Technology (IJESIT). 2013;2(5). [2] Spinelli EM, Martinez NH, Mayosky MA. A single supply biopotential amplifier. Medical Engineering and Physics.23(3): [3] Burke MJ, Gleeson DT. A micropower dry-electrode ECG preamplifier. IEEE Transactions on Biomedical Engineering. 2000;47(2): [4] Winter BB, Webster JG. Driven-right-leg circuit design. IEEE Transactions on Biomedical Engineering. 1983;BME- 30(1):

6 [5] S. Suave Lobodzinski MML. New devices for very longterm ECG monitoring. Cardiology Journal. 2012;19(2): [6] Walsh JA, 3rd, Topol EJ, Steinhubl SR. Novel wireless devices for cardiac monitoring. Circulation Aug 12;130(7): PubMed PMID: Pubmed Central PMCID: [7] Tsai TH, Hong JH, Wang LH, Lee SY. Low-Power Analog Integrated Circuits for Wireless ECG Acquisition Systems. IEEE Transactions on Information Technology in Biomedicine. 2012;16(5): [8] Nemati E, Deen MJ, Mondal T. A wireless wearable ECG sensor for long-term applications. IEEE Communications Magazine. 2012;50(1): [9] Pandya UT, Desai UB. A Novel Algorithm for Bluetooth ECG. IEEE Transactions on Biomedical Engineering. 2012;59(11): [10] Lerdwuttiaugoon K. NP, editor Wireless Electrocardiogram Monitoring using Mobile Network Communication. Biomedical Engineering International Conference 2014 (BMEiCON2014); ; Fukuoka, Japan. [11] Apostu O, Hagiu B, Pa S, x015f, ca, editors. Wireless ECG monitoring and alarm system using ZigBee. Advanced Topics in Electrical Engineering (ATEE), th International Symposium on; May [12] Chen SK, Kao T, Chan CT, Huang CN, Chiang CY, Lai CY, et al. A Reliable Transmission Protocol for ZigBee-Based Wireless Patient Monitoring. IEEE Transactions on Information Technology in Biomedicine. 2012;16(1):6-16. [13] Lee RG, Chen KC, Hsiao CC, Tseng CL. A Mobile Care System With Alert Mechanism. IEEE Transactions on Information Technology in Biomedicine. 2007;11(5): [14] Hung K, Yuan-Ting Z. Implementation of a WAP-based telemedicine system for patient monitoring. IEEE Transactions on Information Technology in Biomedicine. 2003;7(2): [15] Pan J, Tompkins WJ. A Real-Time QRS Detection Algorithm. IEEE Transactions on Biomedical Engineering. 1985;BME-32(3): [16] Hamilton PS, Tompkins WJ. Quantitative Investigation of QRS Detection Rules Using the MIT/BIH Arrhythmia Database. IEEE Transactions on Biomedical Engineering. 1986;BME- 33(12): [17] Arzeno NM, Deng ZD, Poon CS. Analysis of First- Derivative Based QRS Detection Algorithms. IEEE Transactions on Biomedical Engineering. 2008;55(2): [18] Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PC, Mark RG, et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation June 13, 2000;101(23):e215-e20. [19] Metting van Rijn AC, Peper A, Grimbergen CA. Highquality recording of bioelectric events. Medical and Biological Engineering and Computing. 1991;29(4):