From Time to Tissue: The AI Revolution in Wake up Stroke

Abstract

Wake-up stroke (WUS), representing about 25% of ischemic strokes, presents a major therapeutic challenge due to the unknown time of symptom onset. The shift towards a tissue-based treatment paradigm, reliant on advanced neuroimaging, creates a pressing need for rapid and accurate image analysis. This review systematically synthesizes and appraises the current evidence on artificial intelligence (AI) for the diagnosis, triage, and management of WUS. Following PRISMA guidelines, a comprehensive literature search was conducted across PubMed, Scopus, and Web of Science from inception through March 2024 for studies developing or validating AI models in WUS or late-window stroke. Analysis of 42 included studies demonstrates that AI models, particularly deep learning applied to non-contrast CT (NCCT) and MRI, achieve high diagnostic accuracy (AUCs often 0.89–0.95) in detecting early ischemia, quantifying infarct core and penumbra, and identifying the DWI-FLAIR mismatch. Commercial AI tools now automate these assessments, providing objective, trial-aligned eligibility criteria. Emerging multimodal models show promise for predicting functional outcomes and complications like hemorrhagic transformation. In conclusion, AI integration enables a data-driven, tissue-based approach that can expand treatment access in WUS. For routine clinical adoption, challenges regarding generalizability, interpretability, and the need for robust prospective validation to prove patient outcome benefits must be addressed.

Ähnliche Arbeiten