Explainable Federated Learning for Multi-Class Heart Disease Diagnosis via ECG Fiducial Features.

Background/Objectives: Cardiovascular disease (CVD) remains a leading cause of mortality and disability worldwide, with timely diagnosis critical for preventing long-term functional impairment. Electrocardiograms (ECGs) provide essential biomarkers of cardiac function, but their interpretation is often complex, particularly across multi-institutional datasets. Methods: This study presents an explainable federated learning framework with long short-term memory (FL-LSTM) for multi-class heart disease classification, capable of distinguishing arrhythmia, ischemia, and healthy states while preserving patient privacy. Results: The model was trained and evaluated on three heterogeneous ECG datasets, achieving 92% accuracy, 99% AUC, and 91% F1 score, outperforming existing federated approaches. Model interpretability is provided via SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), highlighting clinically relevant ECG biomarkers such as P-wave height, R-wave height, QRS complex, RR interval, and QT interval. Conclusions: By integrating temporal modeling, federated learning, and interpretable AI, the framework enables secure and collaborative cardiac diagnosis while supporting transparent clinical decision-making in distributed healthcare settings.