Privacy-Preserving Federated Learning for Multi-Center Rural Health Diagnostics

Abstract Rural healthcare systems often face challenges in diagnostic accuracy due to limited data availability, inconsistent record-keeping, and privacy regulations that restrict data sharing across centers. Federated Learning (FL) enables collaborative model training across multiple health centers without centralizing sensitive patient data. This paper proposes an enhanced Privacy-Preserving Federated Learning (PPFL) framework for multi-center rural health diagnostics, integrating differential privacy, secure aggregation, and lightweight blockchain-based authentication mechanisms. The proposed architecture facilitates multi-center participation, allowing decentralized model updates and improving diagnostic accuracy while ensuring strong privacy guarantees. Simulation results demonstrate that the PPFL approach improves model convergence by 5%, reduces data leakage risk by 97%, and enhances scalability for rural healthcare infrastructures.