Automatic segmentation and registration of abdominal aortic aneurysms using 3D ultrasound

Abdominal aortic aneurysms (AAAs) can lead to a fatal haemorrhage when ruptured. To predict the rupture risk of an AAA, Computed Tomography (CT) based wall stress analysis has been proposed and showed its merit in rupture risk assessment. However, CT has some drawbacks, e.g., ionising radiation exposure. As an alternative, 3D ultrasound (US) has shown its feasibility to determine AAA geometry. The major limitations were the lack of automated segmentation and the suboptimal field-of-view. Therefore, this study aims to assess 3D AAA geometry based on multiple 3D datasets using automatic segmentation and registration. Thirteen patients were included prospectively. For each patient CT data were available. For 8 patients one 3D US acquisition was insufficient to capture the complete AAA geometry. Therefore, a proximal and distal sub-volume were acquired and registered. Subsequently, an active contour model was applied to segment the 3D AAA geometry. Results reveal a good agreement with the CT geometries (Similarity indices = 0.78 - 0.93). The Hausdorff distance (HD) values were higher (median = 7.0 mm) at the proximal and distal sides compared to the middle (median = 5.4 mm). Moreover, the median HD decreased by 23% when registration and re-segmentation were applied. This study shows that automatic segmentation and registration of multiple 3D US volumes of AAAs is feasible to determine AAA geometry using 3D US.