Healthcare federated learning: a survey of applications and frameworks

Federated learning has emerged as a transformative paradigm in healthcare, addressing the critical challenge of leveraging distributed medical data while ensuring privacy preservation. This comprehensive survey examines the evolution and current state of federated learning in healthcare, focussing on cross-silo implementations that enable secure collaboration across medical institutions. Our analysis reveals significant advances in three key areas: privacy-preserving techniques specifically optimized for healthcare data, communication-efficient protocols for medical implementations, and novel frameworks for cross-institutional collaboration. Healthcare applications were categorized into seven domains: disease diagnosis, medical imaging analysis, patient outcome prediction, remote monitoring, drug discovery, resource management, and personalized medicine. The study provides insights into technical and practical considerations for healthcare institutions adopting federated learning through a detailed examination of implementation frameworks, privacy mechanisms, and regulatory compliance. Our survey uniquely contributes by presenting (1) a comprehensive taxonomy of healthcare applications with detailed case studies, (2) A comparative analysis of emerging frameworks including NVIDIA Clara, FedML, and IBM Federated Learning, (3) The in-depth evaluation of privacy-preserving techniques specific to medical data, and (4) practical implementation guidelines addressing healthcare-specific challenges. The study is concluded by identifying critical research gaps and future directions, particularly in model interpretability, regulatory compliance, and resource optimization for healthcare settings.