A BLOCKCHAIN-INTEGRATED FRAMEWORK FOR SECURE FEDERATED LEARNING OF ENCRYPTED EHR DATA WITH HOMOMORPHIC ENCRYPTION

Abstract

Healthcare data analytics requires collaborative model training across institutions while preserving strict patient data privacy. Traditional centralized machine learning approaches require pooling sensitive Electronic Health Records (EHRs), creating unacceptable security and regulatory risks. This paper presents a Blockchain-Integrated Framework for Secure Federated Learning of Encrypted EHR data using Homomorphic Encryption. Multiple healthcare institutions train local models on locally stored EHRs encrypted with the CKKS homomorphic encryption scheme, enabling gradient computation without decryption. Blockchain technology manages model update authentication, access control, and provides an immutable audit trail. Smart contracts automate secure aggregation of encrypted local model weights into a global model. Experimental results on the kidney disease prediction task demonstrate competitive global model accuracy with strong privacy guarantees and acceptable computational overhead, offering a scalable foundation for privacy-preserving collaborative healthcare analytics.