Privacy-Preserving Authenticated Federated Learning Scheme for Smart Healthcare System

With the rapid advancement of artificial intelligence and network technology, smart healthcare system provides patients with satisfactory medical experience and clinical diagnosis, thus alleviating the imbalance between limited medical resources and a large patient population. However, the patient privacy security in smart healthcare system is still facing severe challenges. Additionally, due to the signal interruption in the federated learning mechanism, the model parameters can not be transmitted normally between local user and central server. In response to these issues, we propose a privacy-preserving authenticated federated learning scheme for smart healthcare system. Specifically, there is a hybrid federated learning framework composed of peer-to-peer and server-client architecture in the proposed scheme. In the proposed scheme, data owners can interact directly with each other for federated training to overcome the data silos issue. In addition, we leverage a homomorphic cryptosystem and the Schnorr signature algorithm to ensure the security and integrity of local model parameters. Security analysis and experimental results show that the proposed scheme can not only protect the sensitive information of data owners, but also has high efficiency.