Detection of Large Vessel Occlusion Using AI: Evaluating Performance of RapidAI LVO vs Viz.ai LVO in 1,589...

Accurate and rapid identification of large vessel occlusion (LVO) on CT angiography (CTA) is crucial for optimal management, especially regarding endovascular therapy decisions. Automated tools for LVO detection, including RapidAI and Viz.ai, have been employed in some small studies, but their accuracy has rarely been compared in a large consecutive patient series. The purpose of this study was to evaluate RapidAI and Viz.ai's LVO detection software on CTA in a consecutive series of suspected stroke patients at a comprehensive stroke center. Both software programs were used in parallel for two years to compare the diagnostic accuracy.

CTA data from 1,589 consecutive stroke alerts were retrospectively reviewed. Radiology reports and expert review of CTA and CTP imaging confirmed the LVO diagnosis. LVO was defined as occlusion or high-grade stenosis of the intracranial ICA or MCA M1 segment. Cases were excluded if not sent to the software or if there was poor bolus, metal artifact, or brain hemorrhage.

1,523 cases met the inclusion criteria. Among these, 147 (10%) had LVOs. RapidAI processed 1,521 cases (>99%), and Viz.ai processed 1,430 (90%). RapidAI identified 144 LVO cases (98%) vs. 108 (73%) by Viz.ai (P<0.0001). RapidAI correctly identified 94% of LVO-negative cases vs. 91% by Viz.ai (P=0.004).

RapidAI detected a higher percentage of LVOs compared to Viz.ai (98% vs 73%) and correctly identified more LVO-negative cases (94% vs 91%). Viz.ai did not detect 39 LVOs. The number of LVOs missed by the Viz.ai software (26%) could potentially lead to delays in LVO diagnosis and treatment times.