A blockchain-based Verifiable Aggregation for Federated Learning and Secure Sharing in Healthcare
Shubhasri Roy, Debasish Bera · 2023
Increasing demand for secure and efficient health-care system, prompts a transformative shift. Conventional centralized disease prediction systems encounter data access and privacy problems. Federated learning (FL) offers a secure alternative for accurate machine learning. However, existing FL approaches are susceptible to attacks. Moreover, secure storage and sharing of prediction data among medical professionals pose challenges, as reliance on third-party cloud services exposes vulnerabilities. In this paper, a blockchain-based FL scheme is proposed for secure disease prediction. Moreover, a modified certificate-based proxy re-encryption method(PRE) is suggested for sharing the prediction with medical practitioners, which prevents data leakage and addresses the key-escrow issue. Further, data access is regulated through chaincode. Our scheme offers authentication and verification for disease prediction and sharing. Elliptic Curve Digital Signature Algorithm (ECDSA) scheme has been incorporated into the existing Hyperledger framework for achieving the same. Performance analysis on Hyperledger Caliper shows a nominal increase of 14.6% in performance overhead due to the ECDSA layer which is negligible compared to its benefits. The comparative study shows the benefits of data request authentication, elimination of key-escrow problem, user Identity management, data integrity and verifiable global model for FL, which makes our proposed scheme a more secure and trustworthy healthcare system.