Commercial AI Model Diagnostic Accuracy for Intracranial Large- and Medium-Vessel Occlusion in Emergency CT Angiography

The diagnostic accuracy of AIDOC-VO, the first commercial artificial intelligence tool for intracranial large-and medium-vessel occlusion (LVO/MeVO) detection on head-and-neck CT angiography (CTA), was evaluated in a multicenter emergency setting. A prospective diagnostic-accuracy study of 3,031 adult CT angiograms (mean age, 67.3 years ± 16.4 [SD]; 1,549 females) acquired March-July 2024 across a ten-hospital region was performed. The AI model was compared with clinical radiology reporting. Examinations flagged positive or doubt by either the AI model or report underwent blinded rereading for reference-standard establishment. Of 3,031 CT angiograms, valid AI model output was yielded for 2,804 (92.5%), of which 224/2,804 (8.0%) had vessel occlusion (VO) on referencestandard reading. For VO detection within intended use (218/224), sensitivity was 81.7% (178/218) (clinical report: 81.2% [177/218]; P =.91), and specificity was 99.6% (2,569/2,580) (clinical report: 99.3% [2,561/2,580]; P =.12). LVO sensitivity was 92.8% (64/69) (clinical report: 87.0% [60/69]; P =.42) and MeVO sensitivity was 76.1% (121/159) (clinical report: 79.2% [126/159]; P =.55). The AI model identified VOs missed by radiologists in 42 examinations, for an enhanced detection rate of 18.8% (42/224; 15 per 1,000 CT angiograms), and generated 11 false alerts (3.9 per 1,000 CT angiograms). Performance did not differ significantly from clinical radiology reporting.