ProMUSE Reduces Cost of Alzheimer's Diagnosis with Staged AI

Diagnosing Alzheimer's disease (AD) often relies on a combination of multimodal data, including clinical assessments, structural MRI, and PET imaging. However, the acquisition of MRI and PET scans is both expensive and not universally accessible, making comprehensive multimodal inference impractical in many real-world clinical settings. This research introduces ProMUSE, a Progressive Multi-modal Uncertainty-guided Staged Evidential Network designed to address these challenges. ProMUSE initiates the diagnostic process by performing evidential classification using readily available, low-cost clinical data. It then quantifies the uncertainty of this initial assessment using a Dirichlet-based subjective logic model. If the uncertainty exceeds a predefined threshold, ProMUSE adaptively incorporates additional modalities, such as MRI or PET features, fusing the modality-wise beliefs and uncertainties through Dempster-Shafer theory to arrive at a calibrated multimodal prediction. This staged acquisition strategy allows for accurate diagnosis while substantially minimizing reliance on costly imaging. Experiments on various AD datasets demonstrate that ProMUSE achieves accuracy comparable to or superior to full-modality baselines, while reducing MRI/PET usage by 50-90%, leading to significant cost savings and making early AD screening more accessible.