Detection and Classification of Anterior Myocardial Infarction Using Combined Linear and Nonlinear Features of ECG Signals

Abstract

Timely recognition of changes occurring in the electrocardiogram(ECG) leads is important to identify anteroseptal myocardial infraction (ASMI). Many existing detection methods extract linear feature from ECG data. However, hidden information in the ECG signals can be extracted from nonlinear features. A combination of linear and nonlinear features of ECG is proposed to improve the classification of ASMI and normal subjects. Linear features related to ECG morphology contain S and T wave amplitude, negative area of S wave and ST segment derivate. Moreover, discrete wavelet transformation(DWT) is used in ECG data, and then three types of nonlinear features are obtained from the DWT coefficients, containing approximate entropy, Shannon entropy and wavelet entropy. PTB database is used to evaluate the performance of proposed method by SVM classifier. We only use lead V2 ECG signal. Our proposed method yielded classification results of 99.76% accuracy. The sensitivity for normal is 99.83% and for ASMI is 99.71%. Our proposed method can be used for locating the ASMI by studying one lead and there is no need for analyzing other leads. Thus, our proposed algorithm can assist the physicians to locate ASMI accurately.