The dominant T wave cluster and One-Class SVM based analysis of multilead ECG for classification of myocardial infarction

Abstract:
In this paper, we propose a novel algorithm for detecting myocardial infarction based on 12-lead ECG signals. Because of the strong reaction of myocardial infarction in the ST-T segment of the ECG signals, we introduce the method of the dominant T wave to describe the repolarization of the ventricular myocardium as a whole. In order to obtain robust and meaningful diagnostic features, The ST-T segment of the 12-lead ECG signals is synthesized as the dominant T wave for overall analysis. Then, in order to select the decisive heartbeats from the ECG signals, we identify some clusters over all the unlabeled heartbeats on the feature of the dominant T wave and then the result is fed to the classifier as input feature. The public ECG dataset (PTB diagnostic database) is used to evaluate the effectiveness of the proposed method. Since the number of positive samples (myocardial infarction) and negative samples (health control) in the database is not balanced, we introduce the One-Class-SVM, which is also adapted to the practical situation. Compared with the existing supervised learning algorithms, the proposed algorithm can efficiently and automatically detect myocardial infarction and improve the performance in the sensitivity and specificity.