Abstract 047: Diagnostic Performance and Clinical Impact of AI Algorithms in Large Vessel Occlusion Detection: A Systematic Review

Introduction/Purpose Rapid detection of large vessel occlusion (LVO) is critical for timely endovascular therapy in acute ischemic stroke. While artificial intelligence (AI) algorithms show promise in automating LVO detection, their real‐world diagnostic performance and clinical impact remain unclear. This systematic review evaluates the accuracy, workflow efficiency, and clinical outcomes associated with AI‐based LVO detection tools. Materials and Methods We systematically reviewed 48 studies (2019‐2025) from PubMed, Ovid Medline, and EMBASE that evaluated AI algorithms for LVO detection using CT angiography (CTA), non‐contrast CT (NCCT), or multiphase imaging. Included studies reported on diagnostic accuracy (sensitivity, specificity, AUC), workflow metrics (door‐to‐notification time, transfer times), or clinical outcomes. Two independent reviewers extracted data using a standardized form and assessed study quality with QUADAS‐2. Results AI algorithms demonstrated excellent performance for proximal LVOs (ICA/M1 segments), with sensitivity of 85‐98% and specificity of 85‐98%. However, sensitivity dropped significantly for distal M2 occlusions (13‐58%). Deep learning models achieved AUCs of 0.86‐0.97 for anterior circulation LVOs. Automated ASPECTS and collateral scoring correlated strongly with infarct volume (r=0.67‐0.92) and functional outcomes (p<0.001). Clinically, AI implementation reduced door‐to‐notification time by 22.5 minutes (p=0.047) and transfer times by 25 minutes (p=0.01). While AI triage increased thrombectomy rates (OR 1.4, 95% CI 1.1‐1.8), its impact on 90‐day mRS scores was inconsistent across studies. Conclusion AI algorithms show high accuracy for proximal LVO detection and can significantly improve workflow efficiency. However, limitations in detecting distal occlusions and variable real‐world performance highlight the need for further optimization and standardized implementation. Future research should focus on multicenter validation and outcome‐based studies to establish the clinical value of AI‐assisted LVO detection.