Coronary Microvascular Disease Detection Using Ensemble Learning and Advanced ECG Signal Processing

Abstract

Coronary Microvascular Disease (CMD) is a type of ischemic heart disease which affects the very small blood vessels of the heart. It causes chest pain and reduces blood flow. As the large arteries look normal in standard test, CMD is hard to detect. Heart diseases are usually checked using electrocardiograms (ECG), but small changes related to CMD are often too minimal to notice it. This research presents if machine learning can automatically identify CMD from 12 lead ECG signals taken during an Adenosine stress test. The study used a dataset of 97 subjects, including 19 participants with CMD and 78 healthy individuals, derived from ECG data. The ECG was recorded into three stages, Rest, Stress, and Recovery. During the stress phase, a lot of noise was present because of heavy breathing. To solve this, Dual Path preprocessing pipeline was used to clean the noise and measure the features correctly. From the ECG signals, the study extracted 612 features related to heart rate and Morphological features. Using mutual information and correlation filtering, it was reduced to the top 12 most critical biomarkers. An ensemble model, combination of three models Support Vector Machine, Random Forest, and XGBoost was trained and validated using a Leave-One-Subject-Out (LOSO) cross-validation strategy