A Federated Learning Approach to Privacy-Preserving Medical Image Classification Across Distributed Healthcare Systems

Abstract

The surge in medical imaging data and the expansion of distributed healthcare systems have emphasized the need for privacy-preserving machine learning solutions. Traditional centralized approaches to training deep learning models pose risks related to data leakage and non-compliance with health data privacy regulations. Federated Learning (FL) has emerged as a powerful paradigm enabling collaborative model training without raw data sharing. This paper presents a federated deep learning architecture for privacy-preserving classification of medical images, particularly across hospital systems with heterogeneous imaging modalities.We propose a federated convolutional neural network (CNN) framework using federated averaging (FedAvg) and incorporate differential privacy techniques to enhance data protection. Experimental results on benchmark datasets (e.g., BraTS, ChestXray14) demonstrate that FL achieves near-centralized accuracy while maintaining data locality. The study also explores challenges such as data heterogeneity, communication overhead, and defense against adversarial attacks in federated settings.