A Novel Approach Using Voting from ECG Leads to Detect Myocardial Infarction

Abstract

Myocardial infarction (MI) is the irreversible necrosis of the heart muscle due to starvation of oxygen supply to the heart tissue itself. It is a leading cause of mortality globally as reported by WHO. Therefore, it is of critical importance that the occurrence of MI should be accurately and efficiently detected from the Electrocardiogram (ECG) to timely manage the patient medically and/or surgically. In this paper, we have proposed a novel approach using Convolutional Neural Networks (CNN) and consensus of their results representing the 12 ECG leads’ data (I, II, II, aVR, aVL, aVF, V1, V2, V3, V4, V5 and V6). The results show that the proposed approach achieves accuracy ${>}90\%$, sensitivity ${>}94\%$, specificity ${>}86\%$, positive predictive value ${>}92\%$ and negative predictive value ${>}89\%$ when at least 8 of the 12 leads agree on their prediction. The proposed algorithm has been cross-validated on MI & Normal Healthy Controls using cross sectional data from the PhysioNet dataset. The results show that the approach is robust and can be applied to assist clinician and researchers.