ProMUSE: Progressive Multi-modal Uncertainty-guided Staged Evidential Alzheimer Disease Classification

ProMUSE, a Progressive Multi-modal Uncertainty Guided Staged Evidential Network, has been proposed for Alzheimer’s disease classification, leveraging low-cost clinical data to improve diagnostic efficiency. The model utilizes a Dirichlet-based subjective logic framework to assess uncertainty and only incorporates costly MRI or PET imaging when necessary, achieving a 50-90% reduction in imaging usage while maintaining competitive accuracy against full-modality benchmarks. This approach is significant for practitioners as it enhances the feasibility of early AD diagnosis in resource-constrained environments, optimizing both cost and diagnostic reliability.