BLOCKCHAIN-BASED FEDERATED LEARNING WITH SMPC MODEL VERIFICATION AGAINST POISONING ATTACKS FOR HEALTHCARE SYSTEMS

Abstract

The rapid digitization of healthcare systems has led to an unprecedented growth in sensitive medical data generated from Electronic Health Records (EHR), wearable devices, medical imaging, and Internet of Medical Things (IoMT) platforms. Leveraging this distributed data using Artificial Intelligence (AI) and Machine Learning (ML) can significantly enhance disease prediction, early diagnosis, precision medicine, and healthcare management. However, strict privacy regulations, institutional data silos, and increasing cyber-security threats severely limit crossorganizational data sharing. Traditional centralized machine learning approaches require aggregating patient data into a single repository, creating a high-value target for cyberattacks and raising serious concerns about confidentiality, compliance, and ethical governance. Federated Learning (FL) has emerged as a promising decentralized paradigm that enables collaborative model training across multiple healthcare institutions without sharing raw data. Despite its potential, existing FL systems remain vulnerable to challenges such as malicious model updates, lack of trust among participants, data leakage through gradient sharing, and reliance on centralized aggregation servers.This research proposes a secure, decentralized, and privacy-preserving federated learning framework for healthcare by integrating Blockchain technology and Secure Multi-Party Computation (SMPC). The proposed framework eliminates the need for a trusted central authority by leveraging blockchain’s distributed ledger to ensure transparency, traceability, and tamper-proof recording of model updates. Smart contracts are used to authenticate participants, manage training rounds, and enforce secure collaboration policies. Simultaneously, SMPC protocols enable encrypted aggregation of model parameters, ensuring that no participant or aggregator can access individual updates, thereby preventing inference attacks and preserving data confidentiality.The framework is evaluated through simulated healthcare datasets and performance metrics including model accuracy, training efficiency, communication overhead, and resistance to adversarial attacks such as model poisoning and data reconstruction. Experimental results demonstrate that the proposed system maintains competitive predictive performance while significantly enhancing privacy protection, trust, and security in collaborative healthcare AI environments