CT-Based Automated Segmentation and Recurrence Prediction in Chronic Subdural Hematoma: A Dual-Label Multicenter...

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Chronic subdural hematoma (CSDH) recurrence after burr-hole surgery is common.
• Previous studies overlooked biomechanical effects of brain compression and relied on subjective methods.
• Objective and reproducible prediction of CSDH recurrence is needed for clinical translation.

Purpose of the Study:

• To develop an automated framework (CSDH-Net) for simultaneous segmentation of CSDH and compressed brain tissue.
• To characterize interactions between CSDH and brain tissue using imaging features.
• To provide objective and interpretable prediction of CSDH recurrence risk.

Main Methods:

• A multicenter dataset of 897 CSDH patients was used.
• CSDH-Net, an automated dual-label segmentation framework using nnU-Net, was developed.
• Radiomic, volumetric, topological, and intensity features were extracted; recurrence prediction used LightGBM with SHAP analysis.

Main Results:

• CSDH-Net achieved high Dice scores for segmentation (0.953 internal, 0.875/0.980 external).
• The recurrence prediction model showed AUCs of 0.830 (training) and 0.741 (external validation).
• Key predictors included gray-level dependence nonuniformity, hematoma surface area, compressed brain tissue curvature, and hematoma thickness.

Conclusions:

• CSDH-Net enables accurate, automated segmentation and interpretable recurrence prediction for CSDH.
• The framework offers objective, reproducible risk stratification across multiple centers.
• Findings support integration into neurosurgical workflows for personalized patient management.