Interpretable Alzheimer's Diagnosis via Multimodal Fusion of Regional Brain Experts

The article introduces MREF-AD, a Multimodal Regional Expert Fusion model designed for the early and accurate diagnosis of Alzheimer's disease by integrating multimodal neuroimaging data. This Mixture-of-Experts (MoE) framework adapts the contributions of amyloid PET and MRI using a gating network to learn subject-specific fusion weights, achieving competitive performance against traditional models while offering interpretable insights into the contributions of different imaging modalities. The model's implementation is available on GitHub, making it a valuable tool for practitioners seeking to enhance diagnostic accuracy in Alzheimer's research.