Background: Large vessel occlusion (LVO) accounts for up to 40% of all acute ischemic strokes (AIS) and is associated with poor outcome. Fast and accurate identification of LVO and notification of relevant specialists is critical for maximizing the benefit of reperfusion therapies. Viz LVO (Viz.ai Inc. San-Francisco, CA, USA) is a medical product leveraging a convolutional neural network designed to detect LVOs in CTA scans and to notify a neurointerventional specialist within minutes via a dedicated mobile application. We report our experience with Viz LVO at Rambam Medical Center (RMC).
Materials and Methods: Viz LVO was installed and reviewed from January 2018 to March 2019. During that period, CTA scans were auto forwarded to the system (including non-stroke protocols). Device sensitivity and specificity were evaluated in this non-standard setup in comparison to radiologist report. In addition, qualitative system usability feedback was collected.
Results: During the study period, the system processed 1189 scans. Patients' mean age was 62, and 59% were male. Of those, 411 were stroke protocols, and 84 were positive for distal ICA/M1 occlusions according to radiologist read (the target population of the Viz LVO device). Proximal M2 accounted for another 24 cases. The total accuracy of the system was 92% for distal ICA/M1 occlusion. Regarding stroke protocol population only, sensitivity and specificity were 78% and 91%, respectively. PPV in that group was 63% and NPV 94.6%. Adding the proximal M2 specificity raised to 95%. Median time interval from scan to LVO notification was under 3 minutes.
Conclusion: Automatic LVO detection using artificial intelligence, coupled with notification and preliminary viewing via a mobile application has real potential for early, accurate identification of stroke patients, enabling quick decision-making for reperfusion therapies. Our experience evaluating Viz LVO suggests the system might qualify as a decision support tool.
