Detection of Myocardial Infarction in ECG Base Leads using Deep Convolutional Neural Networks

Abstract

Myocardial infarction (MI), commonly known as a heart attack, occurs when blood flow decreases or stops to a part of the heart, causing irreversible damage to the heart muscle. It is a leading cause of mortality around the world according to the WHO reports and, therefore, it is critical to estimate the location & extent of the damaged tissue. Similarly, localization of MI is also significantly important to correctly manage the patient medically and/or surgically. In this paper we propose & implement a system in which the signals from 6 leads (I, II, III, aVR, aVL, aVF) of the ECG are used to detect the cases with MI in the lateral &Inferior walls of the heart. The use of Convolutional Neural Networks (CNN) & a novel voting scheme provides acceptably accurate estimates of MI. The proposed algorithm has been validated on MI & Normal Healthy Controls from the Physio Net dataset. This approach is robust & can be used in the clinical & research settings.