Automatic segmentation of abdominal aortic aneurysms from CT angiography using a context-aware cascaded U-Net

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Abdominal aortic aneurysms (AAAs) require accurate delineation for clinical management.
• Intraluminal thrombosis (ILT) within AAAs presents a complex segmentation challenge due to low contrast and data imbalance.
• Current segmentation methods may lack complete automation and struggle with intricate structures like ILT.

Purpose of the Study:

• To develop and validate a fully automated image segmentation tool for abdominal aortic aneurysms (AAAs) and intraluminal thrombosis (ILT).
• To improve the accuracy and efficiency of segmenting AAA and ILT from contrast-enhanced computed tomography angiography (CTA) data.
• To assess the clinical potential of automated segmentation for AAA management.

Main Methods:

• A novel context-aware cascaded U-Net architecture was employed for automated segmentation.
• Techniques including auto-context, dilated convolutions, anisotropic context modules, hierarchical supervision, and a multi-class loss function were utilized.
• The method was applied to contrast-enhanced computed tomography angiography (CTA) data from 70 patients.

Main Results:

• The automated segmentation achieved high accuracy, with DICE scores of 0.945 for lumen and 0.804 for ILT.
• Quantitative morphological metrics derived from automated segmentation strongly correlated with those from manual segmentation by trained users.
• The system demonstrated comparable performance to trained human experts in segmenting complex AAA structures.

Conclusions:

• The proposed automated multi-class image segmentation tool effectively delineates AAAs and ILT.
• The tool shows high correlation with manual measurements, indicating reliability.
• This automated approach holds significant potential for translation into clinical software to enhance AAA management.