PE-Net: a parallel framework for 3D inferior mesenteric artery segmentation.

Area of Science:

• Medical Imaging
• Computational Biology
• Oncology

Background:

• Mesenteric artery vessel morphology is vital for colorectal cancer diagnosis and treatment.
• Automated vessel segmentation is challenging due to limitations in current methods like convolution and transformers.
• Existing methods struggle with long-range dependencies, large dataset requirements, and issues like over-segmentation and discontinuity.

Purpose of the Study:

• To develop an advanced automated method for mesenteric artery vessel segmentation.
• To overcome the limitations of existing convolution-based and transformer-based models.
• To improve the accuracy and robustness of vessel segmentation, particularly for datasets with limited samples.

Main Methods:

• Proposed a parallel encoding architecture combining transformers and convolutions.
• Integrated a vessel edge capture module to enhance continuity and topology.
• Developed a model robust to position deviations and effective on small-scale datasets.

Main Results:

• Achieved a Dice Similarity Coefficient of 81.64%.
• Obtained an Average Hausdorff Distance of 7.7428.
• Demonstrated improved performance in vessel segmentation accuracy and continuity.

Conclusions:

• The proposed parallel encoding architecture effectively segments mesenteric artery vessels.
• The novel approach enhances robustness for small-scale datasets and improves vessel continuity.
• This method offers a promising solution for automated vessel segmentation in colorectal cancer diagnosis.