Pattern Recognition in the Clinical Electrocardiogram

Abstract

This study was undertaken to demonstrate the feasibility of use of computers in extracting clinically useful parameters from electrophysiologic waveforms. The ECG leads were recorded on magnetic tape. The analog signal was sampled 625 times per second and that data was converted to a form suitable for a general-purpose digital computer. Criteria for clinically significant voltage fluctuations of the signal from the baseline within specified time intervals were determined. The computer was programmed to identify those fluctuations automatically. For an output, the computer produces a set of measurements of ECG waveforms from one cardiac cycle in any random 5-sec. portion of a lead. The program can be expanded to include any measurement, but for present purposes it permits determination of amplitude of P, Q, R, S and T waves, ST and PQ segments, and QT and RR intervals. Variables are measured to an accuracy of 1 part in 1000. Measurements conform to those obtainable by careful hand measurement of magnified tracings of the original records. Automatically derived data can be simultaneously used in computer-programmed statistical analysis to permit classification of the tracings into categories of normality or abnormality. The entire automated system can thus become a diagnostic aid to the physician. This paper deals with one phase of a project directed at the development of an automated system to aid in the diagnosis of heart disease.