Secure Federated Learning for Medical Image Classification
Abstract: Federated learning (FL) enables decentralized model training without sharing raw data. In this work, we review and synthesize recent FL approaches for privacy-preserving medical image classification, with emphasis on COVID-19 chest imaging. We present its benefits over centralized learning. We detail privacy-enhancing techniques (differential privacy, secure aggregation, homomorphic encryption) with key formulations (e.g. FedAvg layer-wise aggregation, Gaussian DP mechanism) and cite examples of their use. We then summarize empirical findings from prior studies: several FL models achieved performance on par with or exceeding centralized training (e.g. Darzi et al. found FL matched centralized accuracy for COVID-19 detection; Peng et al. reported FedAvg improved generalization on multi-hospital X-rays), while DP-enabled FL maintained accuracy with privacy. Finally, we identify open challenges: limited work on chest MRI, handling extreme non-IID data, communication efficiency, and formal privacy guarantees. This comprehensive review underscores current advancements and gaps in privacy-preserving FL for medical imaging, guiding future research.
Keywords: FL, FEDAVG, COVID-19, DP
How to Cite?: Ramanpreet Kaur, Dr. Amandeep Kaur Virk, "Secure Federated Learning for Medical Image Classification", Volume 15 Issue 7, July 2026, International Journal of Science and Research (IJSR), Pages: 830-832, https://www.ijsr.net/getabstract.php?paperid=SR26710183947, DOI: https://dx.doi.org/10.21275/SR26710183947