BCFTL: Blockchain-Enabled Multimodal Federated Transfer Learning for Decentralized Alzheimer’s Diagnosis

Abstract

This article introduces the blockchain-enabled multimodal federated transfer learning (BCFTL) framework designed to improve Alzheimer’s disease (AD) diagnosis by effectively integrating federated learning (FL), transfer learning (TL), and blockchain (BC) technologies. The BCFTL framework facilitates collaborative training of diagnostic models across decentralized institutions, combining clinical and MRI data without compromising patient data privacy or security. TL enhances the generalizability of the model through pretrained VGG16 architectures for robust feature extraction. BC integration ensures data integrity, transparency, and accountability by providing an immutable and verifiable record of data exchanges and model updates across the network. Experimental evaluations demonstrate that BCFTL achieves an impressive diagnostic accuracy of 97% and a low error rate of 2.6%, with privacy safeguards implemented through encryption-based protection mechanisms for shared model features and cryptographically secure aggregation methods. The scalability and accessibility of the framework underscore its practicality for deployment in resource-constrained environments, highlighting its significant potential for broader applications in various medical domains, including the diagnosis and management of neurodegenerative diseases and other conditions that require secure multimodal data integration.