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

Area of Science:

• Cardiology
• Biomedical Engineering
• Medical Imaging

Background:

• Coronary artery disease (CAD) diagnosis relies on accurate assessment of severity and location.
• Magnetocardiography (MCG) offers a non-invasive method for cardiac assessment, but requires advanced feature extraction.
• Deep-mining techniques can potentially enhance the diagnostic capabilities of MCG.

Purpose of the Study:

• To develop an automated and accurate method for assessing CAD severity and location.
• To utilize deep-mining-based magnetocardiography (MCG) pattern features for CAD assessment.
• To create machine learning (ML) models for classifying CAD severity and localizing stenotic regions.

Main Methods:

• Deep-mining of MCG pattern information, extracting features from multiple perspectives including curl, gradient, and divergence fields.
• Application of singular value decomposition, main field, and image class features.
• Introduction of fine granularity and compound statistical parameters for feature analysis.
• Development of ML models to classify CAD severity (none, mild, moderate, severe) and localize stenotic locations (LAD, LCX, RCA).

Main Results:

• The model achieved 85.1% accuracy for CAD severity assessment, with specific metrics including 77.8% precision, 75.9% sensitivity, 95.6% specificity, and an AUC of 0.853.
• Localization models demonstrated high accuracy: 97.6% for LAD, 81.2% for LCX, and 85.9% for RCA.
• Deep-mined MCG features effectively reflected both the severity and location of CAD.

Conclusions:

• Deep-mining-based MCG features provide effective indicators for CAD severity and location.
• The proposed ML-based method offers an automated and accurate diagnostic tool for clinicians.
• This technology can enhance the interpretation and clinical application of MCG for CAD diagnosis.