A Deep Learning Model for Automatic Segmentation of Intraparenchymal and Intraventricular Hemorrhage for Catheter...

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Intracerebral hemorrhage (ICH) carries a high mortality rate, particularly with secondary intraventricular hemorrhage (IVH).
• Optimal surgical management for ICH remains a significant neurosurgical challenge.
• Accurate segmentation of hematomas is crucial for effective treatment planning.

Purpose of the Study:

• To develop a deep learning model for automated segmentation of intraparenchymal hemorrhage (IPH) and IVH.
• To enable precise catheter puncture path planning for ICH treatment.
• To improve the accuracy and efficiency of hematoma volume measurement and centroid estimation.

Main Methods:

• Development of a 3D U-Net model incorporating a multi-scale boundary aware module and consistency loss.
• Segmentation of IPH and IVH from computed tomography (CT) images.
• Clinical validation of the model for puncture path planning, including volume measurement and centroid deviation estimation.

Main Results:

• The model achieved 96% accuracy for intraparenchymal hemorrhage path planning.
• Superior segmentation efficiency and centroid prediction for intraventricular hematomas compared to existing models.
• Demonstrated potential for clinical application through experimental results and clinical validation with 351 cases (103 in test set).

Conclusions:

• The proposed deep learning model offers an efficient and accurate solution for segmenting brain hemorrhages.
• The model facilitates precise catheter puncture path planning, addressing a critical need in neurosurgery.
• The method shows strong generalization ability and potential for widespread clinical adoption.