Advances in artificial intelligence for brain imaging and computer-aided diagnosis of neurological diseases: a systematic literature review
TL;DR
The evidence suggests movement from task-specific pattern recognition toward multimodal and context-rich decision support in selected settings, but it does not establish uniform methodological superiority or clinical maturity.
Abstract
Brain imaging is central to the structural, functional, metabolic, and neurophysiological characterization of neurological disease, yet interpretation remains constrained by heterogeneity in acquisition, annotation, expertise, and clinical context. This systematic literature review evaluates recent artificial intelligence approaches for brain imaging and computer-aided diagnosis, with emphasis on methods, modalities, neurological diseases, diagnostic tasks, datasets, validation practices, and translational limitations. The review followed PRISMA 2020 and covered peer-reviewed English-language studies published from January 2020 through June 2026. Eight information sources were searched using database-specific terms for artificial intelligence, neuroimaging, neurological disease, and diagnostic analysis. The search identified 1,078 records; 236 duplicates were removed, 842 unique records were screened, 258 full texts were assessed, and 123 studies were included after a record-level eligibility re-audit. Heterogeneous evidence was synthesized narratively and by descriptive mapping rather than meta-analysis. MRI was frequently represented in the included studies and auxiliary mapping data; CT was commonly used in acute stroke and intracranial hemorrhage; PET supported selected molecular and differential-diagnosis studies; fMRI and diffusion imaging were used for network and white-matter analysis; and EEG extended the scope to neurophysiological diagnosis. CNN and three-dimensional networks were commonly reported, while attention, Transformer, graph, self-supervised, generative, multimodal, and foundation-model approaches appeared in selected recent studies. Stroke, dementia, and brain tumors were frequently represented, but heterogeneous populations, outcomes, and incomplete validation extraction precluded ranking diseases, modalities, architectures, or validation strategies. The evidence suggests movement from task-specific pattern recognition toward multimodal and context-rich decision support in selected settings, but it does not establish uniform methodological superiority or clinical maturity. Dependable translation requires patient-level data governance, transparent reporting, uncertainty estimation, independent multicenter testing, prospective workflow evaluation, and clinically meaningful endpoints.
