Blockchain-Enabled Federated Learning with Hybrid Deduplication for Privacy-Preserving Biomedical AI: A Multi-Modal Validation Study

Abstract

Federated learning (FL) has emerged as a promising paradigm for privacy-preserving machine learning in healthcare, enabling collaborative model training without centralizing sensitive patient data. However, existing FL approaches face significant challenges in storage efficiency, communication overhead, and trust management in biomedical applications. This paper presents a novel blockchain-enabled federated learning framework with hybrid deduplication mechanisms specifically designed for biomedical AI applications. Our framework integrates three complementary deduplication strategies: Content Based Deduplication (CBD), Exponential Growth Deduplication (EXGD), and Semantic Similarity-Aware Deduplication (SSAD), combined with a lightweight blockchain middleware for decentralized governance and audit trails. We validate our approach through comprehensive experiments on two distinct biomedical tasks: ECG arrhythmia detection and human activity recognition using wearable sensors. Results demonstrate remarkable consistency across domains, achieving 70% storage reduction, 56% communication latency improvement (50ms → 22ms), and maintaining clinical-grade accuracy (72% for ECG, 98.9% for wearable sensors) while providing immutable audit trails through blockchain integration. The framework’s universal performance characteristics across biomedical modalities establish it as a foundational technology for next-generation healthcare AI infrastructure.