Explainable Hybrid Deep Learning Framework for Cardiovascular Disease Prediction and Clinical Decision Support

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning for Disease Prediction

Background:

• Cardiovascular disease (CVD) poses a significant global public health challenge.
• Machine learning (ML) models offer potential for early CVD identification and risk prediction.
• Existing models often lack interpretability, hindering clinical adoption.

Purpose of the Study:

• To propose HCVDNet, a hybrid ML-deep learning model for enhanced CVD risk prediction.
• To improve the accuracy and interpretability of AI-driven CVD diagnostics.
• To provide clinicians with clear, actionable explanations for AI predictions.

Main Methods:

• HCVDNet integrates ensemble methods (XGBoost, CatBoost, Random Forest) with deep learning (CNNs, LSTM).
• Post-hoc explainable AI techniques (SHAP, LIME) are incorporated for global and instance-level explanations.
• Experiments utilize cardiovascular datasets from UCI and Kaggle repositories.

Main Results:

• HCVDNet achieved superior performance over baseline models, with 96.94% accuracy and 98.72% AUC.
• Explainable AI methods confirmed clinically relevant feature patterns in model outputs.
• The model demonstrated a low Expected Calibration Error (ECE) of 0.021.

Conclusions:

• Combining ensemble learning and deep temporal modeling significantly boosts predictive accuracy and interpretability.
• Explainable AI enhances physician trust in AI-assisted decision-making for CVD risk.
• HCVDNet offers a high-performance, interpretable, FHIR-compliant framework for real-time clinical application.