Strategic Adaptive Federated Learning Framework for Privacy-Preserving Image Data Mining in Highly Heterogeneous Medical Environments

Gailan, Mayyadah Jabbar

Journal La Multiapp
Vol. 7 No. 3 (2026): Journal La Multiapp

The data mining of medical imaging data is being increasingly restricted by stringent data privacy regulations like the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountability Act (HIPAA). Even though FL offers a decentralized framework for model training, it suffers from significant performance degradation in heterogeneous settings characterized by non-IID data. In this work, a novel framework, namely Adaptive Privacy-Preserving Federated Learning, is proposed. This framework combines an adaptive weighting scheme with Differential Privacy to address the issue of divergence caused by statistical heterogeneity. As per the experimental evaluation of the MedMNIST dataset, a classification accuracy of 94.2% is achieved with a privacy budget of ε = 1.0.