Automated Detection and Quantification of Hemorrhagic Transformation After Endovascular Thrombectomy

BackgroundHemorrhagic transformation (HT) after endovascular thrombectomy (EVT) is a principal determinant of clinical outcome. Artificial intelligence (AI) algorithms for spontaneous hemorrhage detection exist, but none has been validated for post-procedural HT across multiple imaging modalities. MethodsWe conducted a multicenter diagnostic accuracy study within the Clinical Research Collaboration for Stroke in Korea registry (18 centers, 2022-2023). Patients who underwent EVT and received follow-up NCCT, GRE, or SWI within 168 hours were included. AI-derived hemorrhage volumes were compared against expert-determined ECASS classification. Three-month modified Rankin Scale (mRS) scores were evaluated for volume-outcome association. ResultsAmong 1,490 patients (median age 73; 57.4% male), HT was present in 41.4% and parenchymal hemorrhage (PH) in 11.1%. PH detection sensitivity exceeded 94% across all modalities (NCCT 95.4%, GRE 94.4%, SWI 98.3%), with AUCs of 0.900, 0.943, and 0.953, respectively. AI-derived volume correlated with 3-month mRS (Spearman {rho} = 0.353, P < 0.001); good outcome (mRS 0-2) declined from 61.8% to 6.7% across increasing volume categories. Among ECASS 0 cases, AI-positive patients had significantly worse outcomes than true-negatives (good outcome 48.2% vs 67.2%, mortality 10.7% vs 4.6%, P < 0.001). ConclusionsAI-based hemorrhage quantification provides high detection of clinically significant PH after EVT and demonstrates a dose-response association with functional outcome. AI-derived volume may serve as a continuous prognostic biomarker that identifies at-risk subgroups beyond categorical ECASS grading.