Privacy-Preserving Computation for Multi‑Center Medical Research: Challenges, Techniques, and Yidu Cloud Solutions

The talk, presented by Jiang Jinpeng, Chief Architect of Yidu Cloud, introduces the emerging field of privacy‑preserving computation, its principles, and why it is essential for medical research that requires secure, compliant data sharing across multiple hospitals.

Background and Challenges : Modern evidence‑based medicine needs large, diverse datasets that single hospitals cannot provide; data heterogeneity, privacy regulations (e.g., GDPR, China’s Data Security Law and Personal Information Protection Law), and cumbersome manual data integration create bottlenecks.

Privacy‑Preserving Computation : Described as “data usable but invisible”, it includes homomorphic encryption, secret sharing, differential privacy, trusted execution environments, and federated learning. Historical milestones from Rivest’s homomorphic encryption (1978) to recent advances in federated learning and secure MPC are outlined.

Medical Multi‑Center Requirements : Strict data privacy, complex statistical analyses (survival analysis, hypothesis testing), large‑scale heterogeneous data, and uneven computing resources demand a solution that combines local processing with secure cross‑hospital computation.

Yidu Cloud Solutions : Three deployment modes—centralized NCRC platform, pure multi‑party secure computation platform, and SaaS collaborative platform—are described. The architecture separates hospital‑side data governance, privacy‑computing nodes, and a cloud‑side orchestration layer with blockchain‑based audit trails.

The platform uses advanced secret‑sharing (Shamir with verifiable shares), optimizes MPC communication to constant‑time complexity, supports multi‑channel parallel tasks, and allows plug‑in switching between cryptographic techniques and trusted execution environments.

Case Studies : (1) Predictive modeling of relapse risk for acute leukemia patients across three hospitals using federated learning, achieving AUC comparable to centralized XGBoost. (2) Multi‑center retrospective study on early prostate cancer diagnosis involving seven hospitals, demonstrating consistent results between secure multi‑party computation and traditional centralized analysis.

The presentation concludes that privacy‑preserving computation is a fast‑growing technology, essential for future multi‑center medical research, and invites collaboration with academic institutions such as Tsinghua University.