Explainable-enhanced AI for diagnosing coronary microvascular dysfunction with multimodal imaging

This study develops a clinically applicable, explainability-enhanced AI model, CMVD_MDAS, to improve the diagnosis of coronary microvascular Dysfunction (CMVD) by integrating deep learning and multimodal machine learning. The model was trained and internally validated on 592 myocardial segments and externally validated on 352 clinical patient segments. The architecture includes automated myocardial segmentation, convolutional neural network-based feature extraction, and multimodal diagnostic modeling with SHAP-based feature ranking to enhance explainability. The CMVD_MDAS demonstrated excellent internal performance (AUC: 0.999) and robust external performance (AUC: 0.79), surpassing physician assessments and commercial software. This explainability-enhanced AI solution significantly improves the accuracy and efficiency of CMVD assessment, reducing diagnostic time by approximately 90% and offering a potential diagnostic tool for clinical practice.