Federated Learning-Enabled Collaborative Access Control for Medical Data with Multiple Authorities
Yifei Li, Zhichao Yuan, Wei Liu, Hongyan Di · 2025
With the rapid development of smart healthcare, ensuring efficient sharing and intelligent utilization of medical data while safeguarding patient privacy has become a pressing challenge. To address this, we propose federated learning-enabled collaborative access control for medical data with multiple authorities. In the proposed scheme, collaborative decryption is enabled by keys generated through secure communication among multiple authorities, allowing authorized users within the same group to decrypt data jointly. Secret parameters are co-generated by the untrusted authorities and a system administrator, which reduces the computational overhead for individual institutions and enhances overall system security. Furthermore, medical data are classified, encrypted, and stored in the tamper-resistant InterPlanetary File System (IPFS), ensuring data confidentiality and integrity. Meanwhile, clients train a global model using local plaintext medical data via the FedAvg algorithm. The hash values of training data and local model updates are then recorded on a blockchain, enabling traceability and auditability of training processes. Security analysis demonstrates that the proposed scheme is both secure and practical.