Trustworthy Blockchain-based Federated Learning for Electronic Health Records: Securing Participant Identity with Decentralized Identifiers and Verifiable Credentials

TL;DR

This work presents TBFL, a security-first framework for federated learning on electronic health records that anchors trust in cryptographic institutional identity using Decentralized Identifiers (DIDs) and Verifiable Credentials (VCs). By gating participation with SSI-based authentication and coupling it to blockchain-managed access control, TBFL prevents Sybil and poisoning attacks while maintaining strong clinical utility demonstrated on the MIMIC-IV mortality prediction task (AUC $=0.954$, Recall $=0.890$). The architecture stores only lightweight on-chain proofs (IPFS CIDs and credential hashes), achieving near-zero on-chain overhead, transparent auditability, and an economically feasible operation (approximately $18 for 100 rounds across multiple institutions). These results indicate that identity-first security can enable scalable, compliant, inter-institutional health data collaboration without sacrificing performance or budget, and the framework offers a reusable blueprint for secure collaborative learning in other privacy-sensitive domains.

Abstract

The digitization of healthcare has generated massive volumes of Electronic Health Records (EHRs), offering unprecedented opportunities for training Artificial Intelligence (AI) models. However, stringent privacy regulations such as GDPR and HIPAA have created data silos that prevent centralized training. Federated Learning (FL) has emerged as a promising solution that enables collaborative model training without sharing raw patient data. Despite its potential, FL remains vulnerable to poisoning and Sybil attacks, in which malicious participants corrupt the global model or infiltrate the network using fake identities. While recent approaches integrate Blockchain technology for auditability, they predominantly rely on probabilistic reputation systems rather than robust cryptographic identity verification. This paper proposes a Trustworthy Blockchain-based Federated Learning (TBFL) framework integrating Self-Sovereign Identity (SSI) standards. By leveraging Decentralized Identifiers (DIDs) and Verifiable Credentials (VCs), our architecture ensures only authenticated healthcare entities contribute to the global model. Through comprehensive evaluation using the MIMIC-IV dataset, we demonstrate that anchoring trust in cryptographic identity verification rather than behavioral patterns significantly mitigates security risks while maintaining clinical utility. Our results show the framework successfully neutralizes 100% of Sybil attacks, achieves robust predictive performance (AUC = 0.954, Recall = 0.890), and introduces negligible computational overhead (<0.12%). The approach provides a secure, scalable, and economically viable ecosystem for inter-institutional health data collaboration, with total operational costs of approximately $18 for 100 training rounds across multiple institutions.

Figures (10)

• Figure 1: Detailed Experimental Pipeline. Note the explicit separation of the Test Set before applying SMOTETomek to prevent data leakage. The Blockchain layer acts as a gatekeeper between local training and global aggregation.
• Figure 2: High-level architecture of the proposed TBFL framework, illustrating the integration of the Identity Layer (SSI with DIDs and VCs), Coordination Layer (Blockchain-based smart contracts), and Learning Layer (Federated Learning with FedProx).
• Figure 3: Trustworthy Federated Learning (TBFL) Workflow. The timeline illustrates the four sequential security phases: (1) Onboarding and Credential Issuance via a Trusted Issuer; (2) Authentication and Task Selection managed by the Smart Contract; (3) Authenticated Model Update Submission leveraging IPFS and DID signatures; and (4) Secure Aggregation, where only verified updates are incorporated into the global model.
• Figure 4: Long-term convergence analysis over 100 federated learning rounds. Left panel: Global accuracy rapidly stabilizes around Round 4, achieving approximately 88.1% and maintaining stability throughout subsequent rounds. Right panel: Global loss exhibits consistent minimization, converging to $\mathcal{L} \approx 0.272$, demonstrating robust optimization free from adversarial perturbations.
• Figure 5: Impact of data heterogeneity on individual client and global model performance. Despite significant variations in local data distributions, as illustrated by the divergent performance trajectories of Hospital A (serving a predominantly elderly population) and Hospital C (specializing in trauma cases), the aggregated global model (black line) maintains stable, high performance, demonstrating effective knowledge synthesis across heterogeneous institutional datasets.