Secure and Efficient Federated Statistical Tests for Medical Data based on TEE and Secure Resharing Protocols

The quality of empirical statistical analysis is closely related to the quality and quantity of available data. However, due to privacy concerns, it is often difficult to collect data from different data owners, thus limiting the usefulness of the collected data. Existing solutions mostly focus on evaluating federated statistical analysis using homomorphic encryption (HE) or secure multiparty computation (SMPC). Our solution incorporates trusted execution environment (TEE) for federated statistical tests. Initially, we propose federated statistical tests framework based on single-party TEE. Considering the security of the proposed framework, we then extend single-party TEE into a distributed TEE cluster and combine it with secure resharing protocols to realize model security and confidential computation while preserving data privacy in federated setting. We provide a comprehensive security analysis for the proposed federated statistical tests framework and evaluate the common Welch's t-test on diabetic dataset. Security analysis and experimental evaluation prove that our framework is end-to-end secure and efficient.