How This AI Slashes Alzheimer's Imaging Costs by 90% Without Cutting Corners

Alzheimer’s disease (AD) diagnosis increasingly relies on multimodal data like clinical assessments, structural MRI, and PET imaging. However, MRI and PET are costly and not universally accessible, making full-modality inference impractical in real-world clinics. To address this, researchers from multiple institutions propose ProMUSE (Progressive Multi-modal Uncertainty Guided Staged Evidential Network). The system first performs evidential classification using low-cost clinical data alone and quantifies uncertainty via a Dirichlet-based subjective logic model. When uncertainty exceeds a learned threshold, ProMUSE progressively incorporates MRI or PET features, fusing modality-wise belief and uncertainty through Dempster-Shafer theory to produce a calibrated multimodal prediction. This staged acquisition strategy allows the model to achieve competitive or superior accuracy compared to full-modality baselines while dramatically reducing the need for expensive imaging.

ProMUSE was evaluated on three public datasets (ADNI, AIBL, OASIS) across multiple classification tasks: Cognitively Normal vs. Alzheimer’s, Cognitively Normal vs. Mild Cognitive Impairment, and MCI vs. AD. Results show that ProMUSE reduces MRI/PET usage by 50-90% without sacrificing diagnostic performance, yielding substantial cost savings. The method is uncertainty-aware and resource-efficient, offering a practical path for large-scale AD screening where imaging resources are limited. The authors argue that such adaptive, cost-sensitive AI systems can accelerate early diagnosis and expand access to timely interventions.