Classification of Myocardial Infarction and COVID-19 Related Cardiac Injury from ECG Signals Using Advanced Deep Learning Architecture

Abstract

Cardiovascular diseases, particularly Myocardial Infarction (MI), are still a primary cause of international mortality. The last COVID-19 pandemic has further complicated the cardiac health landscape, with the virus known to induce cardiac injuries such as Advanced heart block (AHB) and Myocardial Injury (HMI). The Electrocardiogram (ECG) is a primary, non-invasive diagnostic tool for these conditions. This paper presents a comprehensive comparative analysis of three state-of-the-art deep learning architectures—ResNet-50, YOLOv8, and Meta CLIP—for the automated classification of cardiac conditions from ECG signals. We collected a dataset comprising ECG traces from several sources with MI, AHB, HMI, COVID-19, and normal rhythm. Each model was trained and validated on this dataset. Our experimental results demonstrate exceptional performance across all architectures. ResNet-50 and YOLOv8 achieved a training accuracy of 0.99 and 0.97, respectively, with training losses of 0.0982 and 0.181. whereas the Meta CLIP model achieved a training loss of 0.106 and an accuracy of 0.99, with commensurate high validation accuracy. The findings suggest that deep learning models, originally designed for computer vision, can be effectively adapted for robust and accurate ECG analysis, paving the way for enhanced clinical decision-support systems