Segmentation Method of Cerebral Aneurysms Based on Entropy Selection Strategy.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurosurgery

Background:

• Cerebral aneurysm segmentation is difficult due to similar imaging features and foreground-background imbalance.
• Existing 2D methods neglect 3D information and global context, limiting performance.

Purpose of the Study:

• To develop an automatic and accurate method for cerebral aneurysm segmentation.
• To leverage 3D information and global features for improved segmentation accuracy.
• To mitigate the impact of class imbalance in medical image segmentation.

Main Methods:

• A 2D U-Net backbone enhanced with a Transformer block for capturing remote dependencies.
• An entropy selection strategy to focus on challenging regions and reduce class imbalance effects.
• Inputting continuous 3D patches and employing a recombination strategy for inference.

Main Results:

• Achieved a Dice coefficient (DSC) of 0.944 on the CADA dataset.
• Obtained an IOU score of 0.941 and a recall of 0.946.
• Demonstrated strong performance with an F2 score of 0.942 and a Hausdorff distance of 3.12 mm.

Conclusions:

• The proposed method effectively segments cerebral aneurysms by integrating 3D context and addressing class imbalance.
• The enhanced U-Net with Transformer blocks shows significant improvements in accuracy and robustness.
• This approach offers a promising solution for automated cerebral aneurysm detection in clinical practice.