Multibranch 1D CNN for Detection and Localization of Myocardial Infarction from 12 Lead Electrocardiogram Signal

Abstract

This paper presents a multi-branch Convolutional Neural Network (CNN) architecture to detect Myocardial Infarction (MI) from 12 lead electrocardiogram(ECG) signals. For each of the 12 leads, a feature extraction network consisting of 1D CNN layers is used to learn the features. The learned features from the 12 leads are concatenated and fed to a feature aggregator network, which summarizes the learned features and the final classification is done using a feedforward neural network. The performance of the proposed model is evaluated using PTB diagnostic database. The model achieves an accuracy of 92.3% for the detection of MI and 90.1% accuracy for MI localization on subject-based evaluation. The performance results endorse the effectiveness of the proposed model for MI detection and localization.