Answers are grounded in this paper and cite their sources.
Scoping review evaluates AI models for dementia diagnosis, highlighting challenges and clinical applications.
Introduction Dementia diagnosis remains challenging due to symptom overlap, label noise, and limited access to specialist expertise. Artificial intelligence (AI) holds promise for inferring diagnostic patterns from multimodal and neuroimaging data. Methods This scoping review, conducted in accordance with the PRISMA-ScR guideline, searched Scopus, PubMed and IEEE Xplore for peer-reviewed papers (2014–2025) evaluating AI models for the diagnosis of dementia. Fifteen studies were included and synthesised narratively. Results AI tasks clustered into binary AD versus cognitively normal classification (AUCs 0.88–0.98), multi-etiology differential diagnosis (AUCs 0.82–0.96), and MCI-to-AD prognosis (AUCs 0.74–0.84). Deep-learning models dominated, with pathology-supervised multi-label learning, confound-aware vision transformers and elastic-net–based proteomic classifiers showing the strongest clinical promise. External generalisation, label noise and fairness gaps persisted across studies. Discussion Credible translation requires multi-centre validation, pathology-aligned ground truth and minimal multimodal configurations integrating anatomical, functional and molecular signals.
No takes yet. Share an insight, caveat, or question.
Genni et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: