Diagnostic Value of Artificial Intelligence—Based Software in Detection of Large Vessel Occlusion in Acute Ischemic Stroke

Abstract

Background: Aim of the study was to test the accuracy of AI-based software for detection of large vessel occlusion (LVO) with computed tomography angiography (CTA) in stroke patients using an experienced neuroradiologist’s evaluation as the reference. Methods: Consecutive patients who underwent multimodal brain CT for suspected acute ischemic stroke were retrospectively identified. The presence and site (classified as proximal and distal) of LVO were assessed in CTA by an experienced neuroradiologist as a reference and compared to readings of three medical students and AI-based software, the e-CTA. Results: One-hundred-eight participants with a mean age of 70 years (±12.6); 55 (50.9%) females were included. Neuroradiologist found LVO in 70 (64.8%) cases: 45 (41.7%) proximal, and 25 (23.1%) distal. The overall sensitivity for e-CTA was 0.67 (95%CI 0.55–0.78); 0.84 (95%CI 0.71–0.94) for proximal, and 0.36 (95%CI 0.18–0.57) for distal LVOs. Overall specificity and accuracy for e-CTA were 0.95 (95%CI 0.82–0.99) and 0.77 (95%CI 0.68–0.84), respectively. The student’s performance was similar to e-CTA. Conclusions: The tested software’s performance is acceptable for the detection of proximal LVOs, while it appears to be not accurate enough for distal LVOs.

Keywords

• artificial intelligence
• cerebral angiography
• spiral computed tomography
• acute stroke
• cerebrovascular occlusion