Abstract 054: Mapping the Landscape of AI Applications in Acute Stroke Management: A Scoping Review

Le résumé fourni par la source

Introduction Acute stroke is a time‐critical condition where delays in diagnosis and treatment significantly affect outcomes. Artificial intelligence (AI) has shown promise across multiple domains of stroke care, yet its clinical integration remains inconsistent. A comprehensive mapping of current AI applications is needed to understand the scope, trends, gaps, and translational challenges. This scoping review synthesizes peer‐reviewed evidence on AI use in acute stroke care, focusing on imaging, triage, prognostication, rehabilitation, and decision support. Methods Following PRISMA‐ScR guidelines, we systematically searched MEDLINE, Embase, IEEE Xplore, and Cochrane CENTRAL from inception to March 2025. Inclusion criteria comprised original studies involving AI tools for acute stroke management across any clinical setting. Data extraction focused on publication year, study design, AI type, clinical task, validation status, and model performance metrics. Using R and Python, we performed descriptive analytics and visualized distribution by clinical domain. A custom ontology was developed to categorize AI use into six domains: imaging/diagnosis, prognostication, triage/workflow, rehabilitation, clinical decision support (CDS), and implementation/usability. Results Out of 2176 records screened, 133 studies met inclusion criteria. Imaging and diagnosis dominated the landscape (42 studies, 31.6%), particularly in ischemic stroke detection via non‐contrast CT and diffusion‐weighted MRI, with average AUCs exceeding 0.90. Prognostication followed (27 studies, 20.3%), featuring deep learning models predicting 90‐day mRS scores and hemorrhagic transformation with accuracies up to 88%. Workflow and triage AI systems (19 studies) demonstrated strong potential to reduce door‐to‐needle times, yet only 5 were validated in real‐world emergency settings. Rehabilitation‐focused studies (14) applied AI to robotics, motion tracking, and tele‐rehabilitation, but showed wide variability in outcome measures. Clinical decision support tools (22) included integrated CDS in telestroke platforms, though only 8 achieved clinical integration. Implementation and usability studies (9) highlighted concerns around algorithmic bias, regulatory hurdles, and clinician trust. Only 12% of studies were externally validated and just 1.5% were part of interventional trials. Conclusion The AI research landscape in acute stroke management is rapidly expanding, with imaging and prognostication being the most mature domains. However, critical translational gaps exist in clinical validation, regulatory clearance, and human‐centric usability. Rehabilitation and CDS applications remain underexplored and inconsistently evaluated. To transition from innovation to impact, future research must prioritize multi‐center trials, harmonized reporting standards, and ethical integration frameworks. This review offers a structured roadmap for researchers, clinicians, and policymakers to navigate and strengthen AI's role in acute stroke care. image

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.