Convolutional Neural Networks Based Diagnosis of Myocardial Infarction in Electrocardiograms

Abstract

Myocardial infarction also called heart attack, is the most dangerous Coronary heart disease for humans beings. Portable Electrocardiogram(ECG) device is useful for the identification and control of ECG signals for myocardial infarction. These ECG signals record heart electrical activity and reflect the unusual movement of the heart. Visually, it is difficult to identify a variation in ECG due to its small amplitude and period. Therefore in this paper, we implemented a convolutional neural network (CNN) made of two layers of convolution-pooling, two dense layers and one output layer for the diagnosis of myocardial infarction using ECG. For batter performance, this network uses Leaky ReLU neurons with categorical cross-entropy loss function and the ADAM optimizer algorithm. To avoid the problem of overfitting, we used L2 regularisation method for regularization of the dense layer of CNN. For experimentation, we use the Physikalisch-Technische Bundesanstalt (PTB) diagnostic database. In this database, we obtained results of sensitivity, specificity, and accuracy of 100 %, 99.65%, and 99.82%, respectively, for data taken from the training set. And sensitivity, specificity, and accuracy of 99.88 %, 99.65%, and 99.82%, respectively, on patients, it hasn’t seen before which indicating that the model can achieve excellent classification performance.