Computed tomography-based automated measurement of abdominal aortic aneurysm using semantic segmentation with active...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Vascular Surgery

Background:

• Accurate abdominal aortic aneurysm (AAA) measurement is critical for successful endovascular aneurysm repair (EVAR).
• Conventional imaging-based measurements are time-consuming and prone to inaccuracies.
• Improved AAA measurement can prevent complications associated with EVAR.

Purpose of the Study:

• To develop and validate an automated workflow for AAA segmentation and measurement.
• To compare the efficiency and accuracy of the automated workflow against conventional methods.
• To assess the performance of various deep learning models in AAA segmentation.

Main Methods:

• An automated workflow integrating semantic segmentation with active learning (AL) and computer-aided design (CAD) was developed.
• Deep learning models including UNETR, SwinUNETR, and nnU-Net (2D, 3D U-Net, ensemble, cascaded) were trained and evaluated.
• CT scans from 300 patients were used for segmentation, with 7 clinical landmarks automatically measured for 96 patients.

Main Results:

• The 3D U-Net model achieved the highest Dice Similarity Coefficient (DSC) in semantic segmentation at AL stage 5.
• SwinUNETR demonstrated superior performance in 95% Hausdorff distance (HD95).
• Automated segmentation significantly reduced analysis time (e.g., aorta: 9.51 min vs. manual) and provided accurate measurements.

Conclusions:

• The developed automated workflow enhances efficiency and accuracy in AAA measurement compared to traditional methods.
• AI-powered semantic segmentation and measurement offer a promising solution for pre-operative planning in EVAR.
• The study highlights the potential of deep learning for improving AAA assessment and patient outcomes.