A User-Centered Medical Data Sharing Scheme for Privacy-Preserving Machine Learning

With the rapid development and application of artificial intelligence technology, medical data play an increasingly important role in the medical field. However, there are privacy protection and data ownership issues in the process of data sharing, which brings difficulties to machine learning and data mining. On the one hand, for fear that they may risk being held accountable by users or even breaking the law due to these issues, healthcare providers are reluctant to share medical data. On the other hand, users are also reluctant to share medical data due to the possibility of privacy disclosure in the data sharing process. To improve the security and privacy of shared medical data, we propose a user-centered medical data sharing scheme for privacy-preserving machine learning. Our solution combines blockchain and a trusted execution environment to ensure that adversaries cannot steal the ownership and control of user data during sharing. A blockchain-based noninteractive key sharing scheme is proposed that allows only the users and the TEE to decrypt the shared data. At the same time, we design an auditing mechanism to facilitate users to audit the sharing process. The security analysis shows that the scheme ensures the privacy and security of user data during storage and sharing. We have completed simulation experiments to demonstrate the effectiveness and efficiency of our scheme.