Abstract 130: Real World Experience With Viz.AI Automated Large Vessel Occlusion Detection

Abstract

Introduction: Viz.AI LVO is an artificial-intelligence platform which analyzes brain computed tomography angiography (CTA) images in patients with suspected acute ischemic stroke (AIS) and sends an automated alert for suspected large vessel occlusions (LVO). The FDA cleared Viz.AI for LVO detection and Viz.AI reports high sensitivity and specificity (96.3% and 93.8%). We report our experience with Viz.AI LVO detection at our academic comprehensive stroke center (CSC). Methods: We performed a retrospective review of suspected stroke patients who had CTA ordered as a stroke code from September 2020 to December 2020. Data was collected on sex, age, Viz.AI LVO alert, and official Radiology review of CTA. Radiology read of CTA was considered gold standard. True negative was defined as Viz negative, Radiology negative. False positive was defined as Viz positive, Radiology negative. False negative was defined as Viz negative, Radiology positive. True positive was defined as Viz positive, Radiology positive. LVO was defined as occlusion of the intracranial carotid (ICAT) or MCA (M1 or M2 segment). Data was collected on performance of Viz.AI LVO alert, including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results: Among 282 consecutive suspected AIS patients with CTA analyzed by Viz.AI, the mean age was 61.9 years (range 19 to 96) and 152 (53.9%) were female. Viz LVO autodetection alerted for 60 patients (21.3%). Radiologist review reported 32 patients with LVO as follows: 12 (37.5%) M2, 16 (50%) M1, and 4 (12.5%) ICAT. The following were adjudicated: 219 true negatives, 31 false positive, 3 false negative, and 29 true positives. Sensitive was 90.6% and Specificity was 87.6%. PPV was 0.483 and NPV was 0.986. The 3 false negative patients all had M2 occlusions. The median time to alert was 10 min (range 7 to 656). Conclusions: In our series of AIS patients evaluated with CTA at an academic CSC, Viz.AI automated LVO detection performed well with a sensitivity of 90.6% and specificity of 87.5%, with a median time to alert of 10 minutes.