Coronary artery disease severity and location detection using deep-mining-based magnetocardiography pattern features

The objective of this study was to develop an automated, accurate method of assessing coronary artery disease (CAD), including its severity and location, using deep-mining-based magnetocardiography (MCG) pattern features.

The pattern information of MCG was mined deeply, and features were extracted from multiple perspectives. The curl, gradient, and divergence fields were extended based on the current field to visualize hidden pattern information before the singular value decomposition, main field, and image class features were proposed. Finally, the statistical parameters of fine granularity and compound features were introduced. To estimate the CAD severity, stenosis was classified as none, mild, moderate, or severe, and a suitable subset of features for machine learning (ML) modeling was presented. To localize CAD, it was categorized according to the stenosis location, including the left anterior descending (LAD), left circumflex artery (LCX), and right coronary artery (RCA), and the selected subsets of features appropriate for each localization ML model.

The test set exhibited an accuracy, precision, sensitivity, specificity, F1 score, and area under the receiver operating characteristic curve of 85.1 %, 77.8 %, 75.9 %, 95.6 %, 75.9 %, and 0.853, respectively, for CAD severity. The model test sets to detect the LAD, LCX, and RCA locations exhibited accuracies of 97.6 %, 81.2 %, and 85.9 %, respectively.

Deep-mining-based MCG features can effectively reflect the severity and location of CAD. The proposed ML method can serve as an automated, accurate diagnostic tool for clinicians to improve the interpretation and application of MCG technology in clinical settings.