Zero-Knowledge Proofs for Secure and Privacy-Preserving Electronic Health Record Sharing in Decentralized Federated Learning Systems

In modern healthcare systems, collaborative machine learning on Electronic Health Records (EHRs) is often hindered by stringent privacy regulations such as HIPAA and GDPR. While Federated Learning (FL) mitigates direct data exposure by keeping data localized, centralized FL remains vulnerable to single points of failure, malicious aggregators, and membership inference attacks. This paper proposes a novel, decentralized, blockchain-based federated learning framework that integrates Zero-Knowledge Proofs (ZKPs) to guarantee secure and privacy-preserving EHR sharing. By leveraging zk-SNARKs, our architecture enables medical institutions to verify the validity of local model updates without disclosing the underlying clinical data or the model parameters themselves to unauthorized parties. The decentralized ledger acts as an immutable, trustless orchestrator, replacing the central server with a smart contract-based consensus mechanism. We evaluate our framework using the MIMIC-III dataset, simulating a multi-hospital collaborative diagnostic scenario. The results demonstrate that our approach maintains high model accuracy (AUC of 0.89) comparable to traditional centralized FL while achieving robust defense against poisoning attacks and inference vectors, with an acceptable computational overhead for zk-SNARK generation. This research highlights the viability of combining cryptographic proofs with distributed ledgers to foster secure, collaborative medical research.