MICROPROCESSOR AIDED ONLINE PROCESSING OF ELECTROCARDIOGRAPHIC SIGNAL FOR DIAGNOSIS AND PROSTHESIS

Abstract

Early detection and timely treatment of serious arrhythmias in coronary care units (CCUs) reduce the risk of sudden cardiac death. Because of the human limitation in handling critical care service like this, the concept of automation has been widely accepted for CCUs. The objectives of automation are to improve patient care, to reduce avoidable cardiac deaths and to lighten the work load of the medical and paramedical staff. During the initial years of automation in this field, electronic circuits with capabilities of heart rate metering and low/high he art-rate alarm triggering, were used. The years that followed have seen a major developmental phase due to the availability of digital computers for such applications During the I960's, the simple heart-rate meters with alarm annunciation were replaced by sophisticated computer-based electrocardiographic diagnosis units. However, because of the high capital outlay, the usefulness of the mainframe computers is restricted to large hospitals. Minicomputers, although relatively cheaper, are still too expensive to be afforded by small health care centres. Consequently small centres were deprived of computerized cardiac care service, until a beginning was made in the application of micropro cessors in CCUs in mid seven ties. The small-size low-cost microprocessor system are economically justified for umibed service generally required for the small health centres in rural and semirural areas. ii Automated arrhythmia diagnosis requires continuous monitoring of patient' s ECG. This calls for a real-time and online digital signal processing. Monitoring, which forms a subsystem of the complete CCU management, has to perform the most important function of accurate and reliable identification of QRS-complex. The major diffi culty in this regard is the wide variability of the input signal. Presence of noise and artefact in the signal further hampers the detection process. Once a positive QRS identification has been made, the processing continues with the wave recognition and then proceeds to measurement of ECG parameters. Real-time analysis of single-lead ECG on beat-to-beat basis demands that a few clinically significant parameters, that are identifiable and easily measurable, be made available for diagnostic classification. The directly measured para meters are the interbeet and intrabeat intervals and dura tions. From these measurements a large number of other para meters of interest are derived. All these parameters which are clinically related with the status of the heart are grouped to form a pattern that characterizes an arrhythmia. On mainframe and the minicomputers diagnostic classifica tion has been performed using complex algorithms based on involved mathematical relations. However, for microproce ssor implementation, classification programme should be simple to implement and speedy to execute. The work presented in this thesis has been aimed at developing a microprocessor based on-line ECG processor iii that would meet the need of unibed coronary care units. The developed system provides prosthetic and therapeutic support also. They are cardiac pacing and d.c. cardiovers ion. This additional automation should further improve the management of cardiac care in unibed CCUs. The high rate of false positives and false negatives in the detection of QRS-complex has been minimized by incorporating two features in the QRS-detection model; immunity to various artefacts and noise, and adaptability to sign&l variability. Locating p fiducial point on the cardiac cycle for establishing a time reference in relation to which all temporal parameters are measured, is a significant step in the processing of ECG signal for arrhythmia analysis. In the present work an algorithm has been developed to locate R-peak as a stable and accurate time reference. Immunity to notches aid bursts of noise that may occur around the R-peak has been achieved by using a slope - polarity pattern approach. Parameter measurement and feature extraction for diagnostic classification is an application dependent processing stage. For rhythm analysis and diagnostic classification on microprocessor, the main considerations are the simplicity of approach and speed of execution of algorithm. In view of this, only timing data has been used in the classification algorithm. Basically only three parameters are identified, which carry maximum rhythm iv information and are R_R interval, PR-segment and QRSduration. Even though the basic parameters are only 3, the classification programme uses a number of other fea tures extracted from them. The features selected are such that they are related to the measured parameters through computationally simple arithmetic relations. Diagnostic classification of arrhythmias is complex decision making process^clinical classification of arrhy thmias being logical rather than mathematical, a decision logic approach has been adopted here. Out of the various approaches to logical decision making, the one using deci sion table has been found suitable for microprocessor implementation. ECG history plotting, which helps in retrospective analysis of premonitory arrhythmias, temporary cardiac pacing for restorative prosthesis and cardioverter control for electric-countershock therapy are the additional tasks assigned to the system. To accomplish this a distributed processing approach has been adopted. Two processors in master-slave configuration share the tasks. The basic task of ECG monitoring for arrhythmia classification has been assigned to the master processor and the remaining tasks viz. respiration rate monitoring, ECG history plott ing, cardioversion and external cardiac pacing have been assigned to the slave processor. The system has been evaluated in laboratory using two data bases. One of the data-base used is created by recording on audio cassettee recorder ECG signals from cardiac patients. The other data-base is digitally simu lated ECG waveforms representing various arrhythmias. The former data base has been used for evaluating the per formance of QRS-detection and R-peak location algorithms and the latter for arrhythmia classification algorithm and the tasks assigned to the slave processor.