CT Radiomics Predicts Outcomes After Brainstem Haemorrhage

AUTOMATED CT radiomics showed promise for predicting poor functional outcomes in patients with primary brainstem haemorrhage, according to a study involving 372 patients. The combined model outperformed conventional radiological assessment, while its ability to predict post-stroke pneumonia was less consistent.

Developing an Automated CT Radiomics Model

Researchers developed an automated segmentation-based radiomics framework using emergency NCCT scans. The retrospective study included 372 patients with PBSH, divided into training (n=182), internal testing (n=79), and temporal internal validation (n=111) datasets.

Haematomas were segmented using a three-dimensional U-Net. Researchers compared radiomics models based on automated and manual segmentation before combining automated radiomics features with conventional radiological predictors. These combined models were developed to predict poor functional outcomes at discharge and in-hospital post-stroke pneumonia (PSP).

Combined Model Predicted Functional Outcomes

The automated segmentation-based radiomics model performed comparably to the manual segmentation-based model across all three datasets.

For predicting poor functional outcomes, the combined model achieved area under the curve (AUC) values of 0.926 in the training dataset, 0.923 in the internal test dataset, and 0.858 in the temporal internal validation dataset. These results represented higher AUCs and lower Brier scores than the conventional radiological model, indicating improved predictive performance.

Researchers also assessed the model’s ability to predict PSP. However, the combined PSP model achieved AUCs of 0.828, 0.742, and 0.681 across the three datasets, with no consistent improvement in performance.

Using SHapley Additive exPlanations (SHAP), researchers identified haematoma classification as the most influential predictor in the prognostic model.

Potential for Early Risk Stratification

The findings suggest that automated segmentation-based NCCT radiomics could support early risk stratification in patients with PBSH. Combining radiomics features with conventional radiological predictors showed potential to provide additional prognostic information beyond standard CT assessment.

However, the PSP prediction model requires further refinement and validation.R esearchers concluded that automated CT radiomics is feasible for early prognostic assessment, with further work needed to establish its predictive value.

Reference

Li Y et al. Automated segmentation-based emergency CT radiomics for predicting functional outcome at discharge and in-hospital post-stroke pneumonia in primary brainstem hemorrhage. Eur J Radiol.2026; DOI: 10.1016/j.ejrad.2026.113290.