Automated Detection and Location Specification of Large Vessel Occlusion on Computed Tomography Angiography in Acute Ischemic Stroke

Abstract

Background Fast and accurate detection of large vessel occlusions (LVOs) is crucial in selection of patients with acute ischemic stroke for endovascular treatment. We assessed accuracy of an automated LVO detection algorithm with LVO localization feature. Methods Consecutive patients who underwent computed tomography angiography in 2 centers between January 2018 and September 2019 and between June and November 2020 for suspected anterior circulation LVO were retrospectively included. Reference standard for presence and site of an anterior circulation LVO (intracranial internal carotid artery, M1, or M2 segments of the middle cerebral artery) was established by consensus of 2 independent neuroradiologist readings. All computed tomography angiographies were processed by StrokeViewer‐LVO, Nicolab. Accuracy of this algorithm with LVO localization feature was assessed. Results In total, computed tomography angiographies of 364 patients with suspected anterior circulation LVO were analyzed (mean age 67±15 years; 185 male patients). A total of 180 patients (49%) had an LVO (intracranial internal carotid artery [n=49 (27%)], M1 [n=91 (51%)], and M2 [n=40 (22%)]). Sensitivity and specificity for LVO detection were, respectively, 91% (95% CI, 86%–95%) and 87% (95% CI, 81%–91%). NPV and PPV were, respectively, 91% (95% CI, 86%–94%) and 87% (95% CI, 82%–91%). Accuracy of the LVO localization feature was 95%. Median upload‐to‐notification time was 04:31 (interquartile range, 04:21–05:50) minutes. Conclusions The automated LVO detection algorithm evaluated in this study, rapidly and accurately detected anterior circulation LVOs with high accuracy of the LVO localization feature. Therefore, it is a suitable screening tool to support and speed up diagnosis of stroke.

Keywords

• artificial intelligence
• large vessel occlusion
• stroke
• thrombectomy