Detection of myocardial infarction and arrhythmia from single-lead ECG data using Bagging Trees classifier

Abstract

Myocardial Infarction (MI) and Arrhythmia (AR) are serious heart diseases. Human identification of these diseases from raw electrocardiography (ECG) signals is tedious and expensive. In recent studies, these diseases have been identified and localized individually using multi-channel ECG data. We explored for the robust detection of both MI and AR from ECG signal using a minimalistic (single channel) ECG data. In this study, a total of 440 records from 12-lead ECG signals (79 Healthy Person, 346 MI, and 15 AR) have been used from “PTB Diagnostic ECG Database” of PhysioBank. The proposed algorithm automatically identifies P, Q, R, S, and T points of ECG signals, and then extracts 33 features: 15 interval type and 18 amplitude type. Bagging Tree, an ensemble method which is computationally efficient and at the same time can deal with the class imbalance problem within the data, is used for classification. During training, 10-fold cross validation is used to ensure generalization of the classifier and to remove over-fitting. Our study demonstrates that Bagging Tree is able to identify MI, AR and normal patients with the cross-validation accuracy of 99.7%, sensitivity of 99.4%, specificity of above 99.5%, precision of 99.32%, and F1 score of 99.36% from a single lead ECG data (Lead V4). The proposed algorithm uses a minimalistic single-channel ECG for robust detection of MI and AR that can be utilized in wearables for real-time patient monitoring.