FEDNEUROX: PRIVACY-PRESERVING AND EXPLAINABLE FEDERATED LEARNING FOR EARLY DETECTION OF ALZHEIMER DISEASE THROUGH MULTI-MODAL ANALYSIS

Alzheimer’s Disease (AD) is a progressive neurological disorder that significantly impairs cognitive function, making early detection essential for effective intervention and disease management. Despite extensive research, identifying reproducible and interpretable biomarkers for early AD detection remains a persistent challenge. Deep learning (DL) has shown enormous potential in neuroimage analysis and brain disease diagnosis due to recent developments. Training high-quality DL models, on the other hand, sometimes calls for access to vast, varied data sets, which are hard to centralize given privacy issues, logistical challenges, and the sensitive character of medical data. By allowing cooperative model training across several institutions without the need for centralized data, Federated Learning (FL) has become a fascinating alternative. However, most such FL methods in brain disease diagnosis face two main issues: first, the curse of dimensionality, where brain networks’ high-dimensional features far exceed available sample sizes, resulting in a suboptimal model; and second, substantial data distribution disparities across sites, making communication inefficient, as well as affecting model generalization overall. To overcome these challenges, we introduce FedNeuroX, a novel multi-modal FL infrastructure specifically designed to promote early and differential diagnosis of neurodegenerative brain disorders, such as AD. FedNeuroX exploits a decentralized structure that unites multi-modal data sources like magnetic resonance imaging (MRI), Positron Emission Tomography (PET), and Electronic Health Records (EHRs), to ensure both strong model training as well as top-notch data privacy. Our system builds medical confidence in AI- based diagnostic tools by means of explainable artificial intelligence (XAI) techniques. Experimental evaluation through the Open Access Series of Imaging Studies (OASIS-3) dataset proves that FedNeuroX offers superior diagnosis accuracy, provides enhanced data protection, and preserves strong generalizability across various clinical environments overall.