HeartGuard-AI A Fair, Explainable, and Multimodal Deep Learning Framework for Heart Disease Risk Prediction

HeartGuard-AI is a comprehensive fair, explainable, and multimodal deep learning framework for accurate early detection of heart disease. The system integrates three novel models: FairHeart-Former (a fairness-aware Transformer architecture), TrustHeart-XAI (an uncertainty-aware explainable model with Monte Carlo dropout), and MedFusion-HeartNet (a multimodal fusion network combining clinical, ECG, and lifestyle features). These models, along with their ensemble, were trained and evaluated on a synthetic clinical dataset, achieving superior performance with ensemble accuracy of 95.33%, F1-score of 96.09%, and AUC of 0.938.

The framework incorporates advanced feature engineering, fairness analysis, SHAP-based interpretability, and uncertainty quantification to ensure trustworthy clinical predictions. A user-friendly Flask-based web application enables real-time risk assessment with categorized outputs (Low/Moderate/High Risk), confidence scores, and model transparency. Secure user authentication via MySQL further supports practical deployment. This solution addresses critical challenges in clinical AI including bias mitigation, explainability, and multimodal integration, offering a reliable decision-support tool for cardiovascular healthcare.

Keywords: Heart Disease Prediction, Fairness-Aware AI, Explainable AI (XAI), Multimodal Deep Learning, Transformer, Ensemble Model, SHAP, Uncertainty Quantification, Clinical Decision Support System.