ABSTRACT NUMBER: ESOC2026A206 ARTIFICIAL INTELLIGENCE–ASSISTED DETECTION OF LARGE VESSEL OCCLUSION IN ACUTE STROKE: A SYSTEMATIC REVIEW AND META-ANALYSIS OF DIAGNOSTIC ACCURACY

Abstract Background and aims Rapid identification of large vessel occlusion (LVO) is critical for enabling timely endovascular treatment. Artificial intelligence (AI)–based imaging tools have emerged as potential accelerators of hyperacute stroke workflows, yet their diagnostic performance compared with expert human assessment remains uncertain. This systematic review and meta-analysis evaluated the accuracy and clinical utility of AI-assisted LVO detection on computed tomography angiography (CTA). Methods A systematic search of PubMed, Scopus, Web of Science, and Cochrane Library was conducted from inception to December 2024 for studies assessing AI-based tools for LVO identification on CTA. Eligible studies included prospective or retrospective cohorts reporting sensitivity, specificity, or area under the curve (AUC). Pooled diagnostic estimates were calculated using a random-effects model. Heterogeneity was assessed with the I2 statistic. Risk of bias was evaluated using QUADAS-2. Results Eighteen studies involving 12,946 patients met inclusion criteria. AI systems demonstrated a pooled sensitivity of 0.89 (95% CI: 0.85–0.93; I2 = 71%) and specificity of 0.92 (95% CI: 0.87–0.95; I2 = 64%) for LVO detection. The pooled AUC was 0.94 (95% CI: 0.92–0.96). Subgroup analysis showed higher accuracy for proximal occlusions (sensitivity 0.94; specificity 0.93) compared with distal occlusions. Workflow studies (n = 5) reported reductions in door-to-notification times ranging from 6–22 minutes. No significant increase in false-positive downstream activations was observed (OR: 1.08; 95% CI: 0.84–1.39). Conclusions AI-assisted LVO detection demonstrates high diagnostic accuracy and measurable workflow benefits, supporting its integration into hyperacute stroke pathways. Conflict of interest all authors have has nothing to disclose