An Explainable Federated Vision Transformer Framework for Privacy-Preserving Multi-Modal Alzheimer's Disease Diagnosis Using Distributed Healthcare Data

Abstract

Alzheimer's Disease (AD) is one of the most prevalent neurodegenerative disorders affecting millions of elderly individuals worldwide, causing progressive cognitive decline, memory impairment, behavioral abnormalities, and irreversible neurological damage. Early diagnosis of Alzheimer's Disease is critical for initiating timely therapeutic interventions, slowing disease progression, improving patient quality of life, and reducing healthcare costs. Recent advances in Artificial Intelligence (AI), particularly Deep Learning, have significantly improved the automated diagnosis of Alzheimer's Disease using neuroimaging modalities such as Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), Computed Tomography (CT), and clinical biomarkers. However, developing highly accurate diagnostic models requires access to large-scale, diverse, and representative healthcare datasets distributed across multiple hospitals. Due to stringent healthcare privacy regulations, including HIPAA and GDPR, direct sharing of sensitive patient data among healthcare institutions is often prohibited, limiting the development of generalized diagnostic models. This paper proposes an Explainable Federated Vision Transformer Framework for privacy-preserving multi-modal Alzheimer's Disease diagnosis using distributed healthcare data. The proposed framework integrates Federated Learning, Vision Transformers, multi-modal feature fusion, explainable artificial intelligence using attention visualization and Grad-CAM, differential privacy, and secure model aggregation to establish a collaborative diagnostic architecture that preserves patient privacy while maintaining high diagnostic accuracy. Experimental evaluation conducted on benchmark Alzheimer's Disease datasets demonstrates that the proposed framework significantly improves classification accuracy, model interpretability, federated learning efficiency, privacy preservation, and generalization capability compared with conventional centralized deep learning approaches. The proposed architecture provides a trustworthy and scalable solution for next-generation intelligent healthcare systems supporting collaborative neurological disease diagnosis across distributed medical institutions.