Application of artificial intelligence on magnetocardiology

Abstract

Early diagnosis and localization of the ischemic region are critical for effective treatment of ischemic heart disease (IHD), a condition which leads to a dysfunctional microcirculation in the heart. To detect the presence and extent of myocardial ischemia, functional imaging tests such as stress Positron Emission Tomography/Single Photon Emission CT Myocardial Perfusion Imaging and cardiovascular magnetic resonance are commonly used. However, they expose patients to radiation and reproducibility risks. We implement Support Vector Machines, XGBoost, Multiple Layer Perceptron Networks, and Random Forest algorithms on Magnetocardiography (MCG) data acquired from patients already confirmed to have IHD. We achieve prediction accuracies of 0.80, 0.72, and 0.70 for ischemia in the left anterior descending artery, ramus circumflexus coronary artery, and right coronary artery respectively. We further identify which MCG parameters are more useful than others and determine the best feature subset for each artery. Results show a valuable tool for early diagnosis and effective treatment of IHD.