Abdominal artery segmentation method from CT volumes using fully convolutional neural network.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate segmentation of abdominal arteries from CT volumes is crucial for surgical planning and navigation.
• Existing deep learning methods struggle with segmenting small, intricate structures like blood vessels due to inter-patient variations.

Purpose of the Study:

• To develop a fully automated method for abdominal artery segmentation from CT volumes.
• To enhance the segmentation accuracy of small arteries using novel deep learning techniques.

Main Methods:

• A fully convolutional network (FCN) was employed for abdominal artery segmentation.
• A 2D patch-based segmentation with an area imbalance reduced training patch generation (AIRTPG) method was used to improve small artery segmentation.
• A three-plane segmentation approach (axial, coronal, sagittal) was introduced to achieve clear 3D segmentation results.

Main Results:

• The proposed method achieved an average F-measure of 87.1%, precision of 85.8%, and recall of 88.4% on 20 abdominal CT volumes.
• The method demonstrated superior performance compared to previous automated FCN-based segmentation techniques.
• The results indicate competitive performance against existing 3D blood vessel segmentation methods.

Conclusions:

• A novel, fully automated abdominal artery segmentation method using FCN was successfully developed.
• The 2D patch-based and AIRTPG methods effectively improved the segmentation of challenging artery regions.
• The three-plane approach yielded high-quality 3D segmentation results, beneficial for clinical applications.