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

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Coronary microvascular Dysfunction (CMVD) diagnosis remains challenging.
• Current diagnostic methods can be time-consuming and lack comprehensive analysis.
• There is a need for accurate and efficient tools to assess CMVD.

Purpose of the Study:

• To develop and validate a clinically applicable, explainability-enhanced AI model for CMVD diagnosis.
• To integrate deep learning and multimodal machine learning for improved diagnostic accuracy.
• To enhance the explainability of AI models in cardiovascular disease assessment.

Main Methods:

• Development of the CMVD_MDAS model integrating deep learning and multimodal machine learning.
• Automated myocardial segmentation and convolutional neural network-based feature extraction.
• SHAP-based feature ranking for enhanced model explainability, validated on clinical patient data.

Main Results:

• The CMVD_MDAS model achieved excellent internal performance (AUC: 0.999) and robust external validation (AUC: 0.79).
• The AI model surpassed the diagnostic performance of physicians and commercial software.
• Diagnostic time was reduced by approximately 90% compared to traditional methods.

Conclusions:

• The explainability-enhanced AI solution significantly improves CMVD assessment accuracy and efficiency.
• CMVD_MDAS offers a potential new diagnostic tool for clinical practice.
• AI-driven explainable models can enhance cardiovascular disease diagnosis and management.