Interpretable Algorithm for Diagnosing Myocardial Infarction from Electrocardiography Data

Myocardial infarction (MI), a fatal cardiovascular disease, occurs when the heart muscle is damaged due to lack of oxygen. Early diagnosis is of critical importance, yet interpreting electrocardiogram(ECG) signals requires expertise and time. Therefore, the development of algorithms capable of diagnosing MI using ECG data is studied well in the fields of health and artificial intelligence. In this study, three different deep learning models were explored: Convolutional Neural Networks (CNNs), CNNs with Multi-Head Attention, and Transformer-based models. These models were trained and tested on the publicly available PTB-XL dataset. The performances of the models were evaluated using metrics such as accuracy, sensitivity, specificity, and F1 score. On the other hand, the study focused on the transparency of the models, where Grad-CAM based methods and a perturbation algorithm inspired by adversarial attacks were used to disrupt correlated derivations and analyze the behavior of the models.