Answers are grounded in this paper and cite their sources.
Federated Learning (FL) is a machine learning approach that is decentralized and has been created in order to deal with serious privacy issues in the way healthcare data is managed. It provides a sound way of progressing healthcare AI while at the same time keeping patient confidentiality by allowing collaborative training of models without the need to share the actual patient data. This systematic review brings together the results from 59 empirical studies that were published between 2021 and 2025 and which were selected using a methodology that complies with PRISMA, and looks at the applications of FL in five areas of healthcare: medical imaging, electronic health records (EHRs), genomics, wearables, and research into rare diseases. The review focuses on the main challenges associated with non-IID (non-independent and identically distributed) data, communication overhead, security flaws, and compliance with HIPAA and the GDPR. The privacy-preserving techniques that are examined include differential privacy (DP) together with formal accounting of the privacy budget (ε, δ), secure aggregation, homomorphic encryption (HE), and secure multi-party computation (SMPC). The quantitative synthesis of the model performance reported, the use of FL algorithms, the adoption of privacy mechanisms, and the communication efficiency found in the 59 studies is what sets this review apart from earlier narrative surveys. The future directions considered are the integration of FL with large language models (LLMs), energy-efficient green FL for use in IoT and wearable environments, multimodal FL frameworks, and cross-border governance models. By systematically reviewing the current body of evidence and pointing out the structured research gaps that exist, this review shows the great potential that FL has for enabling secure and cooperative healthcare AI.
No takes yet. Share an insight, caveat, or question.
Paul et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: