Graph-based federated learning for smart healthcare: A comprehensive survey
S. M. Asiful Huda, Marzia Canzaniello, Daniela Annunziata, Francesco Piccialli · Future Generation Computer Systems · 2026
Graph-based Federated Learning (GFL) is a decentralized machine learning paradigm that enables graph neural networks to be trained across multiple client devices without sharing raw data. By combining federated optimization with graph representation learning, GFL is particularly suitable for healthcare scenarios where information is both distributed across sites and intrinsically relational (e.g., patient–disease–treatment interactions, biomedical networks, and temporal clinical trajectories). As GFL attracts increasing interest in applications such as disease prediction, medical imaging, and patient monitoring, it becomes essential to develop a comprehensive understanding of the emerging method families and their practical trade-offs. Yet, the literature still lacks a comprehensive, healthcare-centric synthesis that compares GFL approaches and reflects system-level constraints relevant to next-generation computing. To address this gap, in this paper we present a comprehensive survey of GFL in healthcare. We use a PRISMA-guided search and screening protocol to support a rigorous and reproducible literature collection process. We organize the field through a structured taxonomy spanning privacy/security, personalization under heterogeneity, scalability/graph management, and clinical performance. In addition, we summarize commonly used datasets, evaluation metrics, and open-source implementations through reproducibility signals, and we highlight open challenges and promising directions for building secure, scalable, and clinically deployable GFL systems. This survey aims to support researchers in developing more robust and reproducible methods, and to help practitioners understand key assumptions and select suitable GFL designs for real multi-institution healthcare settings.