Medical BCFL: A blockchain-enabled federated learning architecture for secure healthcare data sharing
Qianyu Dong, Jiazheng Liu, Yuchen Jiang · 2024
In response to the growing demand for secure healthcare systems, this paper disaplys a blockchain-enabled federated learning architecture called Medical BCFL. The architecture aims to address privacy concerns and facilitate secure healthcare data sharing by combining the decentralized nature of blockchain with the collaborative learning model of federated learning. First, a blockchain system is built through smart contracts to record model updates and participant reputations for each round of communication. Next, a federated learning system is built to ensure privacy protection during collaborative model training and encourage data sharing through a reward mechanism. Ultimately, a loosely coupled architecture is used to integrate the blockchain with the federated learning system, where model training takes place in the federated learning system and intermediate information is synchronized to the blockchain. The proposed method achieves over 92% accuracy in MNIST dataset in an adversarial setting of 40% malicious clients and 8% accuracy loss in a real-world BloodMNIST dataset with the same proportion of malicious clients.