AI Scan Model Predicts Brainstem Bleed Outcomes
An automated CT radiomics framework showed promise in forecasting poor functional recovery for patients with primary brainstem haemorrhage.
Updated on Nov. 13, 2026 in Stroke

Researchers developed an automated segmentation-based radiomics model using emergency non-contrast CT scans to predict poor functional outcomes in primary brainstem haemorrhage patients. The combined model outperformed a conventional radiological model, though its pneumonia prediction results were less consistent.
Why it matters
Brainstem haemorrhage is a severe form of stroke, and earlier risk stratification could help clinicians identify patients who need closer monitoring. The findings suggest automated imaging analysis may complement traditional radiological assessment in emergency settings.
The combined model achieved AUCs of 0.926, 0.923 and 0.858 across training, internal test and temporal validation sets of 372 patients, exceeding the conventional radiological model. Whether these performance levels hold in prospective, multicentre clinical use has not been established by the retrospective design.
The details
Researchers used a three-dimensional U-Net to segment haematomas on emergency non-contrast CT scans, then combined the resulting radiomics features with conventional radiological predictors. SHapley Additive exPlanations analysis identified haematoma classification as the most influential predictor in the prognostic model.
Timeline
Study findings were published on November 13, 2026.
Big Picture
The study extends the use of three-dimensional U-Net segmentation, a deep learning architecture widely used for medical image delineation, into automated radiomics for primary brainstem haemorrhage. It shows that radiomics derived from routine emergency CT can be paired with conventional predictors to refine prognosis for this stroke subtype.
If validated in larger prospective studies, automated CT analysis could help clinicians identify brainstem haemorrhage patients at higher risk of poor recovery earlier in their care. For patients and families, this remains a research finding and not yet a change to standard assessment or treatment.
The takeaway
Automated analysis of routine emergency CT scans may add prognostic value for brainstem haemorrhage, but the models still need prospective validation before they inform clinical decisions.
Further reading
For related context on stroke types and treatment, see the site's Stroke section.
Source note: This article includes information reported by European Medical Journal.



