A Blockchain-Based Federated Learning Approach with Secure Third-Party Computation System for Securing Electronic Health Records

The secure, efficient, and privacy-preserving management of Electronic Health Records (EHRs) remains a critical challenge as healthcare systems increasingly depend on digital infrastructures. There is a centralized EHR storage model, which is susceptible to data breaches, unauthorized access, and single points of failure. In order to overcome these shortcomings, this research paper presents a blockchain-based federated learning architecture with a secure third-party computation (BFL-STPC) system to support decentralized, tamper-proof, and privacy-enhancing management of EHRs. The model uses Federated Learning (FL) so that patient information is held at separate healthcare facilities, but encrypted updates to models are jointly utilized to train a universal model. Blockchain technology offers a logical audit trail, open access control, and authentication based on smart contracts without assisting central authorities. The STPC module also has a stronger level of security, as it allows aggregation encrypted by homomorphic encryption (HE), secure multiparty computation (SMPC), and differential privacy (DP). In order to assess real-world deployability, the system was tested with simulated multi-hospital conditions with distributed AWS EC2 nodes, which allows assessment of generalization of the system with a variety of cloud-based institutional settings. The experimental analysis based on a synthetic healthcare dataset proves the effectiveness of the proposed system in comparison with the benchmark strategies, including PPFLB, FEACS, FLBM-IoT, and EJSS. The highest accuracy of 97.38 of the model outweighs that of FEACS (93.18) and PPFLB (91.14). It has the best throughput (1178.32 kbps), minimum authentication (72.59 ms), minimum execution (36.78 ms) and near perfect interruption detection (96.79) that points to remarkable improvements in the system responsiveness and computer efficiency. Moreover, the AUC of the model is 1.00, that is why it displays a good classifying ability and safe decision reliability. The ablation experiments support the fact that each of the elements federated learning, blockchain, and STPC is significant to the performance and security of the whole system. In conclusion, the proposed BFL-STPC model is a regulation-compliant, scalable, and enhanced security model of the existing EHR management. It gives a good ground to the future of the healthcare systems, which contains credible data dissemination, data confidentiality, and credible teamwork intelligence within the various clinical environment.