Health-FedNet: A privacy-preserving federated learning framework for scalable and secure healthcare analytics
Ali Asghar, Václav Snåšel, Jan Platoš
Abstract
The growing demand for privacy-preserving healthcare analytics necessitates solutions that comply with global regulatory standards such as HIPAA and GDPR while maintaining high diagnostic accuracy. This paper introduces Health-FedNet, a federated learning (FL) framework designed for secure, decentralized model training across multiple healthcare institutions without transferring raw patient data. Health-FedNet integrates Differential Privacy (DP), Homomorphic Encryption (HE), and an Adaptive Node Weighting Mechanism to enhance privacy, scalability, and robustness against heterogeneous data distributions. Evaluated on the MIMIC-III clinical database, Health-FedNet achieved a 12 % increase in diagnostic accuracy compared to centralized models, with statistical significance confirmed through a Wilcoxon signed-rank test ( p < 0.01). The adaptive weighting mechanism prioritizes high-quality data sources, ensuring robust learning and model convergence. Furthermore, Health-FedNet supports real-time streaming updates, optimizing communication overhead and latency, making it suitable for time-sensitive clinical scenarios. Encryption mechanisms are designed to maintain privacy during transmission, aligning with international data protection standards. Future work will explore cross-border scalability, multi-institutional real-time synchronization, and edge-based streaming analytics to further enhance global healthcare collaborations. Health-FedNet represents a scalable, privacy-focused solution for modern healthcare analytics, advancing secure and efficient federated learning in clinical settings.