ABSTRACT NUMBER: ESOC2026A1775 PERFORMANCE OF AI-BASED LARGE VESSEL OCCLUSION DETECTION ON NON-CONTRAST HEAD CT IN A REAL-WORLD CLINICAL SETTING

Abstract Background and aims Heuron StroCare SuiteTM (SCS) is a regulatory agency approved AI-based pre-screening tool for large vessel occlusion (LVO) detection on non-contrast CT. SCS predicts LVO in internal carotid artery (ICA) or middle cerebral artery (MCA) by analyzing the probability of eyeball deviation, dense MCA sign, and early ischemic change. We sought to investigate the performance of SCS in acute ischemic stroke patients who received endovascular thrombectomy (EVT). Methods SCS was applied for stroke suspects who perform CT scan in emergency department (ED). Among patients who received EVT, the performance of SCS were reviewed. Factors related with false negative prediction were analyzed. The ED arrival to EVT time was compared between the pre-SCS and post-SCS periods. Results Fifteen patients with anterior circulation stroke received EVT after application of SCS. LVO locations were ICA in 5, M1 in 8, and M2 in 2 patients. The SCS predicted LVO in 13 out of 15 patients. False negatives were more likely to have non-cardioembolic stroke etiology (100% vs. 15.4%, p=0.010) and tended to have minimal early ischemic change (100% vs. 15.4%, p=0.057) than true positives. Age, gender, NIHSS, onset to CT time, LVO location, eyeball deviation, dense MCA sign were not related with the performance. Door-to-puncture time were not different between periods before and after application of SCS (110 (92-125) minutes (median, IQR) vs. 119 (109-139) minutes, p=0.206). Conclusions The AI-based tool predicted 86.7% of LVO on non-contrast CT in patients who received EVT. False negatives tend to occur in non-cardioembolic etiology and minimal early ischemic change cases. Conflict of interest Hyun-Wook Nah: nothing to disclose