Personalized Federated Learning for Distributed Health Systems

Abstract

Information and communication technologies (ICT) have become essential for delivering healthcare services in a faster, more accessible, and data-driven manner. At the same time, the growing scale of digital health data has raised concerns about privacy, security, and the effective use of distributed data sources. In this chapter, federated learning is presented as a privacy-preserving and scalable approach for collaborative model training in healthcare. A skin cancer classification case study was conducted. Centralized learning achieved the highest accuracy of 87.27%, while FedBN-based personalized federated learning reached an average accuracy of about 86% in a two-client setting, showing performance close to centralized learning and clearly outperforming client-based local learning. Overall, the findings indicate that personalized federated learning is a promising solution for privacy-sensitive healthcare applications by combining strong predictive performance with data protection and scalability advantages.