Federated Learning in Biomedical and Health Informatics: A Systematic Review and Future Directions

The growing digitization of healthcare data has enabled rapid advances in biomedical and health informatics, but strict privacy regulations, institutional data silos, and data heterogeneity continue to impede centralized machine learning. Federated learning (FL) has emerged as a pragmatic solution, allowing institutions to jointly train models without sharing raw patient data. This systematic review provides a comprehensive survey of FL in biomedical and health informatics, spanning electronic health records (EHRs), medical imaging, genomics, and wearables. We examine core FL methodologies and system architectures, with particular attention to challenges prevalent in clinical environments, including non-independent and identically distributed (non-IID) data, inconsistent coding standards, and communication constraints. Methodological advances such as personalized federated models, communication-efficient optimization, and privacyenhancing technologies, including differential privacy and secure aggregation, are discussed in depth. We review key applications and assess practical considerations for realworld deployment, including interoperability with hospital systems, regulatory compliance, and model interpretability. Finally, we identify open challenges and future research directions, outlining a roadmap for developing scalable, trustworthy, and privacy-preserving FL systems for clinical practice