The blockchain-based privacy-preserving searchable attribute-based encryption scheme for federated learning model in IoMT

Abstract

Federated learning enables training healthcare diagnostic models across multiple decentralized devices containing local private health data samples, without transferring data to a central server, providing privacy-preserving services for healthcare professionals. However, for a model of a specific field, some medical data from non-target participants may be included in model training, compromising model accuracy. Moreover, diagnostic queries for healthcare models stored in cloud servers may result in the leakage of the privacy of healthcare participants and the parameters of models. Furthermore, the records of model searching and usage could be tracked causing privacy disclosure risk. To address these issues, we propose a blockchain-based privacy-preserving searchable attribute-based encryption scheme for the diagnostic model federated learning in the Internet of Medical Things (BSAEM-FL). We first adopt fine-grained model trainer participation policies for federated learning, using the attribute-based encryption (ABE) mechanism, to realize model accuracy and local data privacy. Then, We employ searchable encryption technology for model training and usage to protect the security of models stored in the cloud server. Blockchain is utilized to implement distributed healthcare models' keyword-based search and model users' attribute-based authentication. Lastly, we transfer most of the computational overhead of user terminals in model searching and decryption to edge nodes, achieving lightweight computation of IoMT terminals. The security analysis proves the security of the proposed healthcare scheme. The performance evaluation indicates our scheme is of better feasibility, efficiency, and decentralization.