Blockchain-Based Quantum-Secure Federated Learning with Hybrid Qkd-Pqc Encryption and Multi-Party Secure Aggregation for Healthcare Systems

Abstract

Predictive modelling of brain and critical care conditions requires accuracy, speed, and privacy, yet sensitive clinical data and fragmented healthcare systems hinder AI implementation. Existing Federated Learning frameworks struggle to balance performance, scalability, security, and privacy in real-world medical environments and deployment. This work proposes a fully encrypted federated learning architecture integrating locally trained models, enhanced by Adaptive Differential Privacy (ADP), with MIMIC-III/IV clinical datasets, ensuring hospital-level data locality while maintaining model accuracy. Post-Quantum Secure Communication is enabled through a Hybrid Obfuscated Quantum Key Distribution–Post-Quantum Cryptography (QKD-PQC) system, providing secure key exchange, dynamic obfuscation, and quantum-resistant encryption. AI-driven smart contracts on a Permissioned Blockchain verify PQC signatures, monitor node reputation, coordinate aggregation, and reward reliable participants. During global model synthesis, a Secure Aggregation Layer using Multi-Party Computation (MPC) and Shamir’s Secret Sharing protects encrypted model updates, ensuring privacy and immutability.