Advancing cardiovascular disease diagnosis with an interpretable and responsible AI framework

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Cardiovascular disease (CVD) is a leading cause of global mortality.
• Current diagnostic methods are often reactive, missing early detection opportunities.
• Accessible and accurate CVD risk stratification is crucial for preventive care.

Purpose of the Study:

• To develop a machine learning (ML) ecosystem for enhanced cardiovascular disease (CVD) diagnosis.
• To create an early warning system using non-clinical data for accessible risk stratification.
• To build specialized diagnostic models integrating both clinical and non-clinical data for improved accuracy.

Main Methods:

• Utilized advanced ML techniques including TabNet, TabPFN, XGBoost, and Random Forest.
• Employed Shapley Additive Explanations (SHAP) for feature importance and interpretability.
• Incorporated counterfactual explanations, fairness mitigation (FairLearn), and uncertainty quantification.

Main Results:

• ECG-related features, specifically ST-segment slope and ST depression, were identified as dominant CVD risk predictors.
• Non-clinical model counterfactuals indicated that reducing angina and chest pain severity can improve CVD risk predictions.
• A hybrid ensemble model achieved 89% accuracy, informed by causal inference and dimensionality reduction.

Conclusions:

• The developed ML ecosystem offers a scalable solution for early CVD detection and equitable diagnosis.
• The system integrates interpretability, fairness, and trustworthiness aligned with regulatory guidelines.
• Actionable insights from the models support personalized lifestyle interventions for CVD prevention.