Automated CT Radiomics Predicts Brainstem Haemorrhage Outcomes
Automated CT radiomics predicted poor functional outcomes in patients with primary brainstem haemorrhage, according to a study involving 372 patients.
- Automated segmentation-based radiomics models performed comparably to manual segmentation methods across training, internal test, and temporal validation datasets.
- Combined models achieved area under the curve values up to 0.926 for predicting poor functional outcomes, outperforming conventional radiological assessments.
- Predictive performance for in-hospital post-stroke pneumonia remained less consistent, yielding lower area under the curve values across datasets.
Emergency non-contrast computed tomography scans formed the basis of the retrospective study, which evaluated patients divided into training, internal testing, and temporal internal validation groups. Researchers segmented haematomas using a three-dimensional U-Net and compared radiomics models built from automated and manual segmentation.
The combined predictive models—integrating automated radiomics features with conventional radiological predictors—achieved area under the curve values of 0.926 in the training dataset and 0.923 in the internal test dataset. In the temporal internal validation dataset, the combined model reached an area under the curve of 0.858.
These figures represented higher area under the curve values and lower Brier scores than models relying solely on conventional radiological assessment. Using SHapley Additive exPlanations, researchers identified haematoma classification as the single most influential predictor within the prognostic model.
When researchers assessed the model for predicting in-hospital post-stroke pneumonia, the combined model achieved area under the curve values of 0.828, 0.742, and 0.681 across the three respective datasets, showing no consistent improvement in performance.
Automated segmentation-based non-contrast computed tomography radiomics showed feasibility for early prognostic assessment in primary brainstem haemorrhage. However, the post-stroke pneumonia prediction model requires further refinement and validation before clinical adoption.
Disclaimer: The information provided in this article is for educational and scientific communication purposes only and does not constitute medical advice. Always consult with a qualified healthcare provider regarding any medical condition, diagnosis, or treatment plan.