Artificial intelligence and radiomics in dementia assessment: Systematic literature review and cross-sectional survey of imaging professionals

Abstract

Dementia is challenging to identify early because structural brain changes may be subtle and symptoms insufficiently specific. This two-part study examined AI-supported radiomics in dementia assessment, particularly Alzheimer’s disease, and explored imaging professionals’ knowledge and perceptions. A systematic literature review searched PubMed and citation sources, with searches completed on 10 August 2026. From 41 unique records or reports, 18 studies published between 2019 and 2025 were included and narratively synthesised. All investigated MRI-derived radiomics combined with artificial intelligence or machine learning for disease classification or progression prediction. Five used longitudinal data, four directly modelled progression or conversion, and two performed external validation. In parallel, a bilingual online questionnaire was completed by healthcare professionals. After excluding one pre-launch test response, 48 responses were analysed descriptively. Radiomics awareness was limited: 62.5% had never heard of radiomics and 66.7% reported no knowledge. Although all respondents knew of AI applications in healthcare and 66.7% used AI-based tools at least occasionally, only 25.0% had received specific training; 89.6% expressed interest in further training. Overall, AI-supported radiomics shows potential as a clinical decision-support approach, but heterogeneity in methods and samples, limited external validation and concerns about observation independence constrain clinical translation. More robust validation, methodological standardisation, suitable infrastructure and targeted professional training are required before routine implementation.