Adv Exp Med Biol. 2026;1489:335-345. doi: 10.1007/978-3-032-03394-9_33.
ABSTRACT
Abdominal aortic aneurysm (AAA) is one of the leading causes of death worldwide. Considering that most AAAs remain asymptomatic until they are close to rupture, an efficient workflow for the accurate detection and delineation is crucial for the clinical outcome. In this study, we conduct a comparative analysis of two different AAA segmentation algorithms using X-ray Computed Tomography (CT) images from 18 patients diagnosed with AAA, which have not been used in similar studies before. The methodologies employed include an in-house segmentation algorithm based on conventional image analysis techniques, and a deep learning approach based on the nnU-Net -framework called TotalSegmentor. The CT dataset, which contained baseline studies, was processed, and the manual annotations by clinicians were used as ground truth. Results demonstrated a high degree of accuracy and robustness, with TotalSegmentor achieving an average Sorensen-Dice coefficient of 0.89 and Jaccard index of 0.81 across the dataset, compared to the proposed unsupervised method’s scores of 0.85 and 0.77, respectively. These findings highlight the potential of deep learning-based models to enhance clinical workflows, ultimately improving early detection and monitoring of AAA.
PMID:41252020 | DOI:10.1007/978-3-032-03394-9_33
