Foundation Models for Glioma MRI: A Systematic Scoping Review of Clinical Applications and Translational Readiness

Background/Objectives: Glioma diagnosis increasingly depends on the integration of magnetic resonance imaging (MRI), molecular characteristics, and clinical information. Foundation models may support AI-assisted diagnosis through transferable representation learning and multimodal integration, but evidence supporting clinical use remains fragmented. This systematic scoping review maps current applications of foundation models in glioma neuroimaging and evaluates their relevance to tumor delineation, molecular characterization, multimodal diagnostic support, and clinical translation. Following PRISMA-ScR principles, six bibliographic sources were searched through 26 July 2026. Eligible records were categorized by model family, pretraining and adaptation strategy, imaging application, multimodal integration, validation design, and clinical relevance. Among 3688 identified records, 2071 remained after deduplication, 525 underwent detailed eligibility assessment, and 125 were included. Evidence was concentrated in promptable segmentation (n = 47), reusable brain MRI representations (n = 26), vision language and language model applications (n = 22), molecular classification and prognosis prediction (n = 16), and translational or governance studies (n = 14). Foundation models showed potential for tumor delineation, reusable representation learning, molecular phenotype prediction, and multimodal diagnostic support. However, independent external validation, robustness to imaging heterogeneity and missing sequences, calibration, and prospective workflow evaluation remain limited. Foundation models offer promising capabilities for AI-assisted glioma diagnosis, particularly in segmentation, reusable MRI representation learning, molecular characterization, and multimodal support. Current evidence is nevertheless insufficient for routine clinical implementation. Multicenter validation, standardized evaluation, uncertainty assessment, and prospective studies of diagnostic and workflow utility should be priorities for future research.