Abstract WP109: CHASM: Comparing Hyperacute Advanced Imaging To Guide Interventional Stroke Management

Introduction: “Stroke AI” platforms discriminate potentially salvageable tissue from infarcted core, to determine embolectomy eligibility. Numerous platforms are in clinical use, but studies comparing their results are limited. Our institution has access to both viz.AI and RAPID for CT perfusion analysis; as such, we assessed whether core and penumbra estimates were statistically different between the two software. Methods: We performed a retrospective review of images from 2/18-1/21 of acute stroke patients with symptoms 6-24 hours from onset who had CTP in the setting of large vessel occlusion (IRB#804221). Studies initially processed through RAPID were re-analyzed by Viz.ai; core volume (CBF<30%), penumbra+core (TMax>6s), and mismatch ratio were compared using paired t-test, and for infinite mismatch McNemar statistics. Results: A total of 157 cases were identified. 48 were excluded for processing failure or deleted raw data, and 6 research participants were excluded. Viz.ai had larger TMax>6s volumes relative to RAPID (160.13ml,108ml;p=<0.001), but less infinite mismatch (30.1%,40.8%;p=0.04). No significant difference was found for: core volume (16.7ml,18ml;p=0.47), percentage of infarcts >70ml (6.8%,7.8%;p=1.0), and NIR eligibility based on DEFUSE-3 (71.8%,68.9%;p=0.63), with <70ml core(93.1%,92.2%;p=1.0), ≥15ml penumbra (78.6%,76.7%;p=0.79), and mismatch >1.8 (83.1%,81.9%;p=1.0). Conclusions: Both Viz.AI and RAPID provided similar core volumes and NIR eligibility, which gives assurance that both algorithms provide similar actionable results. Reasons why Viz.ai showed larger TMax>6s volumes but less infinite mismatch are unclear, though the latter implies that viz.ai identifies a potential core that may or may not be there in more cases. Penumbra estimation is multifactorial and variance may be due to algorithm alterations in voxel selection to account for artifact, scanner and AI protocols as both have different standards for contrast, radiation exposure, and slice #. Understanding if Viz.ai overestimates or RAPID underestimates TMax>6s will require further assessments of imaging and functional outcome variables. Further analysis to assess if results are accounted for by the above factors is ongoing.