qxAI: Quantifiable xAI for Cardiac Diseases

Deep Learning (DL) performs well in Cardiovascular Disease (CVD) classification using 12-lead Electrocardiogram (ECG). However, explainable artificial intelligence (xAI) in CVD classification, still remains largely qualitative. In this paper, we introduce a Region of Interest (ROI) based quantifiable xAI (qxAI), to compare different xAI techniques. Then, we add CVD specific post-processing steps, to increase the explanation performance. Furthermore, proposed qxAI enables selection of an optimal DL model, within the performance space defined by classification, explanation and time-complexity. Finally, we present two distinct use-cases, having ST Segment Elevation (STE) and First-degree Atrioventricular Block (IAVB). The proposed qxAI pipeline increases the accuracy and F1-score of post-processing explainability in STE from 50.7% to 70.5% and 26.1% to 35.8% respectively. Accuracy and F1-score of explainability in IAVB are increased from 39.2% to 40.6% and 49.8% to 71.8% respectively. To the best of our knowledge, this is the first effort to quantify xAI, on CVD classification.