SECURE BLOCKCHAIN-INTEGRATED FEDERATED LEARNING FRAMEWORK WITH SMPC-BASED MODEL VERIFICATION TO MITIGATE DATA POISONING IN HEALTHCARE SYSTEMS

Abstract

The integration of Artificial Intelligence (AI) into healthcare has revolutionized diagnostics and data-driven decision-making but also introduced significant data privacy and security challenges. Federated Learning (FL) has emerged as a solution by enabling decentralized model training without sharing raw patient data. However, FL systems remain vulnerable to data poisoning and model manipulation attacks that compromise global model integrity. This study proposes a secure blockchain-integrated federated learning framework reinforced with Secure Multi-Party Computation (SMPC) for model verification. Blockchain technology ensures transparency, traceability, and immutability in model updates, while SMPC enables privacy-preserving validation of participating nodes. A hybrid verification mechanism detects and filters poisoned gradients before aggregation, significantly enhancing system resilience. Experimental results on medical imaging datasets demonstrate improved model accuracy, reduced attack success rate, and robust protection against malicious participants. This approach establishes a trustworthy and privacy-aware foundation for AI deployment in healthcare systems.