Automated Diagnosis of Coronary Artery Disease and Myocardial Infarction Using Optimal Anti-symmetric Wavelet-Based Features
Authors :- Telangore, H., Sharma, M., Gadre, V.
Publication :- Paradigm Shifts in Communication, Embedded Systems, Machine Learning, and Signal Processing. PCEMS 2024. Communications in Computer and Information Science, vol 2490. Springer, 2025.
Coronary artery disease (CAD) contributes largely to fatal deaths world-wide. CAD is the most fatal of all Cardiovascular diseases (CVD). The formation of plaque in the inner wall of the arteries of the heart causes CAD. CAD progresses rapidly, so if left undiagnosed and untreated in its early stages, it may progress to an irreversible condition of heart failure known as Myocardial Infarction (MI), commonly referred to as a heart attack. Cardiac condition is recorded using electrocardiogram (ECG) signals. However, manual ECG signal detection is tedious and prone to error. Thus, automatic diagnostic systems have been developed to address these drawbacks. In this study, the detection of CAD and MI is performed using an optimal time-frequency localized anti-symmetric biorthogonal wavelet filter bank. We have formulated a multi-class classification task (CT) to separate normal (N) ECG signals from CAD and MI signals. In particular, we consider a 3-class classification task that classifies N versus (vs) CAD vs MI ECG signals. We also consider the following three relevant binary CTs: (i) N vs CAD, (ii) N vs MI, and (iii) CAD vs MI ECG signals. Multiple supervised machine learning techniques were explored for the classification task. The best results are attained by the K-nearest neighbor (KNN) classifier. For 3-class CT (N vs. CAD vs. MI), the average accuracy (AVAC) obtained in this study is 99.5%. Our model produced 100% AVAC for N vs CAD, 99.9% for CAD vs MI and 99.5% for N vs MI ECG signal classification. Our study’s results surpass most of the existing methods deployed for the automated identification of CAD and MI. Further, this study also considers the separation of CAD and MI ECG signals. This study can help cardiologists detect CAD and MI with minimal error and without tedious manual work.