Clinical Impact of Postrecanalization Hemorrhagic Transformation and Its Prediction Using Baseline Noncontrast CT

BACKGROUND: Hemorrhagic transformation (HT) after recanalization therapy remains a critical concern in acute ischemic stroke management. While severe hemorrhages clearly worsen outcomes, the prognostic impact of mild HT and its optimal prediction methods remain uncertain. In this study, we aimed to evaluate the clinical significance of all HT subtypes and develop automated HT prediction models based on noncontrast computed tomography (CT) images. METHODS: We analyzed 2211 patients receiving intravenous thrombolysis, endovascular thrombectomy, or both from the multicenter Clinical Research Collaboration for Stroke in Korea Imaging repository (2022–2024). HT was classified on follow-up imaging of magnetic resonance or CT. Baseline ischemic lesion volume and Alberta Stroke Program Early CT Score–based net water uptake were quantified on baseline noncontrast CT. Multivariable regression was used to assess the association of HT and 90-day modified Rankin Scale scores. The performance of HT prediction models was compared, the automated imaging model versus established scores (hemorrhage after thrombolysis score and SEDAN), using the area under the curve. RESULTS: HT occurred in 41.2% of patients (hemorrhagic infarction [HI] 1: 13.8% and HI2: 16.8%, parenchymal hematoma [PH] 1: 6.5% and PH2: 4.1%). All HT subtypes independently predicted worse functional outcomes with stepwise increasing odds ratios: HI1 (1.77 [95% CI, 1.40–2.22]), HI2 (2.83 [95% CI, 2.27–3.53]), PH1 (4.65 [95% CI, 3.41–6.36]), and PH2 (14.76 [95% CI, 9.61–22.90]). This association persisted across treatment modalities and vascular territories. For PH prediction, the automated imaging model (noncontrast CT imaging markers combined with clinical variables) achieved superior performance (area under the curve, 0.77 [95% CI, 0.73–0.80]) compared with hemorrhage after thrombolysis score (0.71 [95% CI, 0.68–0.75]) and SEDAN (0.72 [95% CI, 0.69–0.76]) scores (both P <0.01 for area under the curve comparison). CONCLUSIONS: Even mild HI was independently associated with poor functional outcomes after reperfusion therapy. Automated noncontrast CT–derived biomarkers provide superior HT risk prediction compared with conventional scores, offering a practical tool for individualized stroke management in the reperfusion era.