Deep Learning to Automatically Segment and Analyze Abdominal Aortic Aneurysm from Computed Tomography Angiography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• Manual segmentation of abdominal aortic aneurysm (AAA) thrombus in contrast-enhanced Computed Tomography Angiography (CTA) is time-consuming and operator-dependent.
• Accurate thrombus segmentation is critical for endovascular treatment planning and outcome monitoring.

Purpose of the Study:

• To develop a fully automatic deep learning pipeline for segmenting intraluminal thrombus in AAA from CTA images.
• To subsequently analyze AAA geometry using the segmented thrombus and lumen.

Main Methods:

• A deep learning pipeline utilizing multi-view U-Nets was developed to localize and segment thrombus from CTA scans.
• Polygonal models of thrombus and lumen were generated, with lumen centerline extraction for diameter computation.
• The pipeline was trained on 63 CTA scans and validated on 14 CTA scans.

Main Results:

• The multi-view integration approach improved thrombus segmentation accuracy, achieving a Dice Similarity Coefficient (DSC) of 0.89 ± 0.04.
• AAA geometry analysis showed high correlation with manual measurements, with an Intraclass Correlation Coefficient (ICC) of 0.92 for total diameter.
• The model demonstrated effective thrombus segmentation and accurate diameter extraction on the test set.

Conclusions:

• The developed deep learning models effectively segment thrombus in patients with AAA.
• Automated AAA geometry analysis, particularly diameter measurements, shows high correlation with expert manual measurements.