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

Researchers introduce ProMUSE, an AI system that intelligently decides when to use expensive medical imaging for Alzheimer's diagnosis by first analyzing low-cost clinical data and progressively incorporating MRI or PET scans only when uncertainty warrants it. The approach maintains diagnostic accuracy while reducing imaging costs by 50-90%, demonstrating practical efficiency gains for real-world clinical deployment.

ProMUSE addresses a critical inefficiency in modern healthcare: the reliance on expensive, resource-intensive imaging for conditions where early detection significantly improves outcomes. By implementing staged, uncertainty-guided acquisition, the system mirrors how experienced clinicians approach diagnosis—starting with basic assessments and escalating only when necessary. This represents a meaningful shift from one-size-fits-all diagnostic protocols toward personalized, cost-aware pathways.

The technical foundation combines evidential reasoning through Dirichlet-based subjective logic with Dempster-Shafer theory for multimodal fusion, enabling the system to quantify and act on diagnostic confidence. Early-stage Alzheimer's treatment effectiveness creates strong incentives for accessible screening, yet MRI and PET infrastructure remains concentrated in wealthy regions. ProMUSE's ability to reduce expensive imaging by 50-90% while maintaining accuracy directly addresses healthcare equity and resource allocation challenges.

From a healthcare economics perspective, the implications extend beyond Alzheimer's diagnosis to any multimodal diagnostic workflow where imaging costs constrain access. Healthcare systems face persistent budget pressures, and reducing unnecessary expensive tests while preserving diagnostic sensitivity provides measurable operational value. Validation across three independent datasets (ADNI, AIBL, OASIS) and multiple classification tasks strengthens confidence in generalizability.

The framework's uncertainty quantification opens pathways for clinical decision support systems that explicitly communicate confidence to practitioners. Future work may extend this adaptive acquisition strategy to other neurodegenerative diseases or complex diagnostic scenarios. Healthcare AI adoption increasingly depends on solutions that balance accuracy with practical deployment constraints, making resource-efficient diagnostic systems critical for translating research into clinical impact.

• →ProMUSE reduces MRI/PET imaging costs by 50-90% while maintaining competitive diagnostic accuracy for Alzheimer's disease classification.
• →The system uses uncertainty quantification from clinical data to intelligently trigger expensive imaging only when necessary, mimicking expert clinical judgment.
• →Validation across three independent datasets demonstrates generalizability and practical applicability in real-world clinical workflows.
• →Progressive multimodal fusion via Dempster-Shafer theory and Dirichlet-based evidential reasoning provides calibrated, confidence-aware diagnostic predictions.
• →The staged acquisition approach addresses healthcare equity by reducing reliance on expensive, geographically concentrated imaging infrastructure.