AI Science Tracker
EN

MEMOIR-VLM—a multimodal vision-language model for Alzheimer's disease classification and question answering

Department of Neurology, University of California, San Francisco, San Francisco, California, USA · 1 Oct 2026 · medicine

Discovery

MEMOIR-VLM—a multimodal vision-language model for Alzheimer's disease classification and question answering

Source check: Source checked. Not independent scientific confirmation.

Finding: On held-out ADNI subjects, the corrected encoder reported 91.3% balanced accuracy for cognitively normal versus dementia classification and 68.2% for three-class diagnosis. Zero-shot OASIS-3 evaluation reported 78.7% balanced accuracy for cognitively normal versus impaired classification (AUC 0.889) with the DTI branch masked. Subject-level held-out ADNI test set (474 subjects) and separate zero-shot OASIS-3 cohort (1,048 subjects); retrospective dataset evaluations only. The paper notes that one initial configuration leaked CDR-SB as input and reports corrected evaluation excluding it. No prospective clinical-use or outcome validation is reported.

Technology: A multimodal learned encoder combines MRI, diffusion imaging and clinical variables for multi-task diagnosis and regression; a retrieval-augmented language layer supports case-based question answering.

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
A multimodal learned encoder combines MRI, diffusion imaging and clinical variables for multi-task diagnosis and regression; a retrieval-augmented language layer supports case-based question answering.
Test and result
On held-out ADNI subjects, the corrected encoder reported 91.3% balanced accuracy for cognitively normal versus dementia classification and 68.2% for three-class diagnosis. Zero-shot OASIS-3 evaluation reported 78.7% balanced accuracy for cognitively normal versus impaired classification (AUC 0.889) with the DTI branch masked.
Limitation / still unknown
Subject-level held-out ADNI test set (474 subjects) and separate zero-shot OASIS-3 cohort (1,048 subjects); retrospective dataset evaluations only. The paper notes that one initial configuration leaked CDR-SB as input and reports corrected evaluation excluding it. No prospective clinical-use or outcome validation is reported.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2026.1902258/full

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 22 Nov 2026. Review scope: Primary source full text Methods and Results; version/date checked against bibliographic record..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse medicine discoveries.