Expert-Guided Transformers: Advancing CAD Diagnosis

Employing Artificial Intelligence (AI) for 12-lead Exercise Stress ECG (ESE) analysis holds promise in refining coronary artery disease (CAD) risk estimation. Despite a 68% sensitivity in traditional clinician analysis, AI's potential in CAD risk remains uncertain. Our method combines transformers [1] with clinical expertise to improve ESE analysis beyond clinical factors. It aims to unveil attention nuances in ECG signals, focusing on gender differences and identifying critical segments for CAD detection. Through meticulous analysis, we achieve superior performance metrics, upholding ethical standards and fostering trust among clinicians. Our research utilizes the Mayo Integrated Stress Center (MISC) database, correlating ESE cases with invasive coronary angiography (ICA) data. Rigorous validation yields robust metrics: Positive Predictive Value (PPV) of 91.2% (±3.3), Negative Predictive Value (NPV) of 93% (±1.5), and Area Under the Curve of Receiver Operating Characteristics (AUCROC) of 92.2% (±1.1). Our findings reveal enhanced detection precision, notably a 39% improvement in females compared to conventional methods.