Understanding and Improving Risk Assessment of Abdominal Aortic Aneurysm Growth Using Statistical Shape Modeling of the Infrarenal Aorta

Abstract

Maximum diameter of the aneurysmal sac remains the primary method for defining surveillance intervals and the timing for surgical intervention. Recent studies have attempted to expand on this one-dimensional metric by incorporating additional descriptors of the aneurysmal sac such as undulation, curvature and tortuosity. However, these metrics are often biased and may overlook certain anatomical abnormalities, limiting predictive potential. This study aimed to identify novel three-dimensional shape features of the aneurysmal flow lumen and outer wall structure (OWS, with or without intraluminal thrombus) to predict aneurysmal growth. A cohort of 192 patients (median age, 72 [range, 62-79]; 99% male) with infrarenal abdominal aortic aneurysms (AAAs) (initial max diameter, 50.3 ± 4.2 mm) and serial computed tomography images obtained during AAA surveillance (median follow-up, 2.0 years [range, 1.0-5.3 years]) were retrospectively considered for this study. Estimated annual AAA growth was the clinical end point and was used to classify patients into slow (first quintile, <2.5 mm/year) and fast (fifth quintile, >5.0 mm/year) cohorts. A previously validated pipeline was implemented to automatically segment computed tomography images. Three-dimensional statistical shape models of the aneurysmal flow lumen and OWS were used to identify three-dimensional patterns involved in AAA growth. Undulation index and radius of curvature of the AAA, previously suggested to be predictive of AAA growth, were also calculated. Model training with derived shape features was performed using iterated 10-fold optimization with shuffling (100 iterations). C-statistics for each feature combination was used to evaluate performance. Integrating the lumen and OWS as unique components within the three-dimensional statistical shape model captured the lumen-thrombus interface. These features proved to be superior to max diameter, undulation index, and radius of curvature in prediction of AAA growth phenotype (P < .001). Corresponding median C-statistics for predicting slow and fast growth are highlighted in the Figure. Integrated models outperformed individual components in prediction of both slow and fast AAA growth (P < .001). Standardized feature extraction from both the aneurysmal flow lumen and the OWS, enabled by a fully automated pipeline, can characterize the evolving lumen-thrombus interface, and improves AAA growth prediction. Future work is focused on validating this statistical shape-based model on larger cohorts that also include smaller AAAs.