Noncontrast CT-based habitat analysis for differentiating hemorrhagic transformation from contrast extravasation after endovascular therapy of acute ischemic stroke.

objectivesHemorrhagic transformation (HT) is a serious and common complication after endovascular therapy (EVT) in patients with acute ischemic stroke (AIS). This study aimed to evaluate the diagnostic value of habitat analysis in differentiating HT from contrast extravasation (CEx) within hyperdense areas (HDAs) after EVT, and to assess whether the preoperative imaging features provide additional diagnostic value. MATERIALS AND

methodsThis retrospective study included 187 AIS patients with HDAs. A stratified three-fold cross-validation was used to split the dataset into training and validation cohorts, and preoperative hyperdense artery sign (HAS) status and clinical characteristics were collected. Habitat analysis was performed using the k-means clustering algorithm applied to postoperative noncontrast CT. The radiomics model, habitat model, and corresponding combined models integrating HAS and clinical characteristics were developed. Diagnostic performance was evaluated by the area under the receiver operating characteristic curve (AUC).

resultsThe optimal k-value was determined to be 5. In differentiating HT from CEx of HDA, the habitat model achieved the highest AUC values of 0.89 and 0.82 in the training and validation cohorts, respectively. Habitat analysis showed improved diagnostic performance compared with the radiomics analysis (AUCs were 0.81 and 0.75 in the training and validation cohorts). The combined model integrating preoperative imaging features, clinical characteristics, and habitat features did not improve performance.

conclusionHabitat analysis outperformed radiomics analysis in differentiating HT from CEx of HDAs after EVT on NCCT. Five‑cluster habitat analysis revealed intralesional heterogeneity within HDAs in AIS patients, supporting its potential as a novel tool for clinical decision‑making. KEY POINTS: Question Differentiation of HT from CEx in post-EVT HDA on non-contrast CT remains a clinical challenge in AIS. Findings Habitat model achieved AUCs up to 0.89 in the training cohort and 0.82 in the validation cohort, outperforming radiomics analysis for differentiating post-EVT HDA and revealed HDA heterogeneity. Relevance Statement Habitat analysis revealed intralesional heterogeneity in the HDA and outperformed radiomics analysis in differential diagnosis on noncontrast CT, which is widely available in most stroke centers. Our findings suggest its potential as a novel clinical decision-making tool.