Toward trustworthy IoMT healthcare: federated learning and verifiable blockchain for secure edge-driven EHR systems

Abstract The advent of Internet of Medical Things (IoMT) has made it necessary to exchange electronic health records (EHRs) across distributed healthcare entities in a manner that ensures their authentication and preserves patients’ privacy. This paper proposes the Federated and Blockchain-enabled Electronic Health Record (FBEHR) framework whose major contribution is the Improved Universal Designated Verifier Signature Proof (IUDVSP). Different from all other pairing-based UDVSP schemes, the suggested solution uses elliptic curve arithmetic compatible with ECDSA to ensure signer authenticity, designated-verifier privacy, and verification control without employing bilinear pairings. According to our experiments in the × 86 edge-server environment, IUDVSP takes 0.22 ms to generate signatures and 0.28 ms to verify them. While these results are below 2-4 ms that we take as the cost of one pairing operation in the adopted reference architecture, they are not the result of a direct same-platform performance comparison of the suggested algorithm with a complete pairing-based UDVSP protocol. The framework also features blockchain-based EHR registration, validation, access, revocation and federated ANN-LSTM-based malicious-user detection. In an internal evaluation of the suggested system, activation of edge processing results in the mean registration and validation latency reduction from 9.87 to 3.59 s which corresponds to the 62.76% latency improvement relative to the same framework but without edge processing. Performance of the FBEHR relative to existing baselines is similar; thus, the primary benefit of FBEHR is its pairing-free designated-verifier security, not latency improvement over those systems. An estimation of the total cost of ownership for a month-long period of FBEHR deployment gives us USD 1850 for the considered FBEHR architecture and USD 12,500 for the same architecture but centralized. This difference of 85.2% is calculated based on a model and is not the experimental performance cost reduction.