Collaborative Mechanism of Privacy Computing and Blockchain in Multi-Center Medical Data Sharing

The fragmentation of sensitive medical data across institutions and stringent privacy regulations creates isolated data silos, hindering the development of robust AI-driven healthcare solutions. Existing approaches often lack sufficient trust mechanisms, comprehensive auditability, and dynamic access control for multi-center collaboration. Goal: This protocol aims to resolve the tension between collaborative utility and data sovereignty by integrating federated learning (FL) with blockchain technology, providing a secure, auditable, and scalable framework for medical data sharing. The framework employs a tripartite architecture: cryptographic FL with adaptive regularization and Paillier homomorphic encryption for privacy-preserving local training; a multi-channel, tiered-sharding blockchain for decoupled access control, data provenance, and FL operation logging; and a dynamic privacy-computation coordination mechanism that adapts differential privacy and cryptographic protocols based on data sensitivity and hardware capability. The workflow follows a sequential set of steps, including data preprocessing, local encrypted gradient generation, blockchain-backed secure aggregation, and global model updating. Under experimental conditions, the framework achieved 92.5% model accuracy after 20 training rounds and supported up to 30 nodes. No raw data exposure was detected, and tamper detection succeeded in all evaluated test cases. The architecture reduced computational overhead by 41% compared to a centralized baseline, achieved 96% automated regulatory compliance for audited access approvals, and demonstrated linear scalability, with the block generation interval stabilizing at 4.2 s.