Outcome-defining Hematoma Expansion in Intracerebral Hemorrhage: A CT-based Risk Stratification Study

Hematoma expansion (HE) strongly influences spontaneous intracerebral hemorrhage, yet targeted therapies have shown limited benefit. We aimed to predict outcome-defining, threshold-crossing expansion and establish a clinically actionable risk stratification framework.

We retrospectively analyzed 407 patients with small-to-moderate baseline hematomas, classified as no expansion, HE, or severe HE (sHE). Three models were constructed and compared for each outcome: radiomics-only (R), radiomics-semantic imaging (RI), and an integrated radiomics-imaging-clinical (RI-CL) model. Model performance was evaluated using area under the receiver operating characteristic curve (AUC). A two-threshold, three-tier risk stratification system was derived from the optimal sHE model, and SHapley Additive exPlanations (SHAP) were used for model interpretation.

The integrated RI-CL model demonstrated the superior performance for both outcomes (test set AUC: HE = 0.768, sHE = 0.723). SHAP analysis identified basal ganglia involvement, irregular hematoma shape, and lymphocyte percentage as the most influential predictors. Risk stratification based on model probabilities revealed a marked gradient in sHE incidence, ranging from 5.2% in the low-risk group to 27.6% in the high-risk group. All expansion events in the high-risk group were threshold-crossing and clinically malignant.

Hematoma expansion in sICH is clinically heterogeneous, and only threshold-crossing expansion drives malignant outcomes. A multimodal CT-based model estimated the risk of threshold-crossing hematoma expansion. Risk stratification showed a marked gradient in sHE incidence, identifying patients at highest risk for outcome-defining expansion.