Multicenter Validation of an AI-Based CTA Tool for Anterior Circulation Large Vessel Occlusion Detection.

BACKGROUND: Rapid identification of anterior circulation large vessel occlusions (LVOs) is critical for timely mechanical thrombectomy in acute ischemic stroke. Computed tomography angiography (CTA) interpretation can be challenging, particularly in settings without continuous subspecialty expertise. Artificial intelligence (AI) based decision support tools may improve workflow efficiency and diagnostic consistency. The purpose of this study is to evaluate the diagnostic performance, processing time, and generalizability of the Methinks CTA-LVO software for automated detection of anterior circulation LVOs. METHODS: This retrospective multicenter study included consecutive CTA scans from four external institutions in the United States and Europe. After quality assessment, 379 patients were analyzed (142 LVO, 237 non-LVO). Ground truth was established by review with adjudication by independent expert neuroradiologists. Primary endpoints were sensitivity and specificity for LVO detection. Secondary analyses included performance by occlusion subtype (ICA, M1, M2), time-to-notification, false positive/negative characterization, and institution-level generalization. RESULTS: The algorithm achieved a sensitivity of 95.8% (95% CI: 91.0-98.4%) and specificity of 88.6% (95% CI: 84.0-92.4%), with an AUC of 97.6%. Sensitivity by subtype was 98.5% for M1, 89.7% for M2, and 97.1% for ICA occlusions. The mean time-to-notification was 3.30 minutes. Error analysis showed that several apparent false positives and negatives reflected ground-truth ambiguity, severe stenosis, or occlusions outside the study definition. Performance remained robust across institutions and CT vendors. CONCLUSION: The Methinks CTA-LVO software demonstrated high accuracy, rapid notification, and good generalizability for anterior circulation LVO detection, supporting its use as a triage tool to assist timely stroke care, with final interpretation remaining under expert supervision.