A comprehensive survey on federated learning for privacy preservation in digital healthcare applications

Abstract

Internet of Medical Things (IoMT) is a relatively new service that has the possible to revolutionize healthcare by connecting previously analog technologies digitally. Consequently, numerous healthcare applications based on IoMT are utilized in the course of daily life. Despite the abundance of machine learning (ML) techniques aimed at improving healthcare data management, none of them have been able to guarantee the data's complete privacy and security. The precise nature of the clinical data makes ML application difficult and yields unsatisfactory results. A new paradigm in ML called federated learning (FL) has arisen as a means to discover untapped potential in digital healthcare uses that protect patients' and clients' privacy without compromising their data. This survey comprehensively reviews over 105 peer-reviewed publications (2018–2025) sourced from IEEE, Elsevier, Springer, ACM, and MDPI digital libraries. It classifies existing FL approaches for digital healthcare based on architecture, communication efficiency, privacy preservation, and application domain. Survey highlights key findings, comparative analyses with conventional FL-based systems, and lessons learned from prior studies. Finally, open challenges such as scalability, energy efficiency, model heterogeneity, and secure aggregation are deliberated, along with future research directions to enable trustworthy FL-based IoMT healthcare ecosystems.