Abstract Number: Esoc2026a203 Artificial Intelligence-Assisted Ct Perfusion in Hyperacute Stroke: A Systematic Review and Meta-Analysis of Diagnostic Performance

Abstract Background and aims Artificial intelligence-assisted perfusion analysis is increasingly used to support rapid stroke diagnosis, yet its comparative accuracy versus expert interpretation remains unclear. Aim To evaluate the diagnostic performance of AI-based CT perfusion tools for identifying salvageable penumbra and core infarction in hyperacute stroke. Methods Studies comparing AI-generated CT perfusion outputs with expert radiological assessment or MRI-DWI were systematically reviewed. Primary outcomes were sensitivity and specificity for detecting infarct core and penumbra. Secondary outcomes included processing time, inter-observer variability, and decision-making impact. Random-effects bivariate modelling generated pooled diagnostic accuracy statistics. Results Fourteen studies involving 4,902 patients were included. For infarct core detection, AI tools demonstrated pooled sensitivity of 0.88 (95% CI 0.83–0.92) and specificity of 0.84 (95% CI 0.79–0.89; I2 = 37%). Penumbra identification yielded sensitivity 0.86 (95% CI 0.80–0.90) and specificity 0.82 (95% CI 0.77–0.87; I2 = 42%). AI systems reduced processing time by a mean of 3.4 minutes (95% CI 2.1–4.7; p 0.001). Decision-support analyses from five studies showed increased treatment eligibility with AI integration (OR 1.21, 95% CI 1.03–1.42; p = 0.02). No significant increase in inappropriate thrombolysis was observed. Conclusions AI-assisted CT perfusion provides high diagnostic accuracy with faster processing and improved workflow efficiency, supporting its use in hyperacute stroke assessment. Conflict of interest all authors have has nothing to disclose