Image-Based 3D Characterization of Abdominal Aortic Aneurysm Deformation After Endovascular Aneurysm Repair

Figure 1. Volume renderings of computed tomography angiography scans, with superimposed aneurysm segmentations for some patients in our study. (a,b) Images of patients with unfavorable evolution at t1 (left) and t2 (right). (c,d) Images of patients with favorable evolution at t1 (left) and t2 (right).

Figure 2. Required segmentations that serve as input for the proposed alignment pipeline. (Left) Aortic lumen segmentation and centerline extraction, keeping only the three main branches. (Right) Consistently segmented aneurysm, starting from the lowest renal to the iliac bifurcation.

Figure 3. (Left) Rigid registration of the vertebrae and co-registration of the aortic lumen. A displacement in the aortic lumen between the reference scan (green) and the co-registered lumen (yellow) is observed in the zoomed region. (Right) Lumen and thrombus straightened curved planar reformations from the two scans of a patient.

Figure 4. Displacement fields produced by the proposed pipeline for two unfavorable evolution patients (top and middle rows) and a favorable evolution patient (bottom row). The first column depicts the straightened thrombi that serve as input to the deformable registration step, where the reference case is shown in green and the thrombus at the second time point is shown in yellow. The second column represents the obtained displacement fields, whereas the third column corresponds to a zoomed region. For unfavorable evolution cases (first and second rows) most arrows point toward the outside of the aneurysm, while for favorable evolution cases arrows generally point toward the inside, representing an shrinkage of the aneurysm. The color of the arrows represents the magnitude of the displacement. The vectors are discretized and scaled by their displacement magnitude for visualization purposes.

Figure 5. Tensile strain patterns for some example favorable evolution patients. The first column shows tensile strain fields for three different patients, where the tensile strain vectors are randomly oriented. The second, third, and fourth columns depict the magnitude of the vector components in each direction, projected into the main AAA axis to show the dominant direction of the strain patterns. For favorable evolution cases no dominant direction is observed, represented by the lack of red areas, which coincides with the conclusion that the strain vectors in these cases are randomly oriented. Vectors have been sampled every 100 points for visualization purposes.

Figure 6. Tensile strain patterns for some example unfavorable evolution patients. The first column shows tensile strain fields for three different patients. In unfavorable evolution cases, the tensile vectors are mostly tangential to the AAA surface. The second, third, and fourth columns depict the magnitude of the vector components in each direction, projected into the main AAA axis to show the dominant direction of the strain patterns. For unfavorable evolution cases the dominant directions can be clearly observed, which are represented by reddish areas mostly in the region the AAA is expanding. Vectors have been sampled every 100 points for visualization purposes.

Figure 7. Compressive strain patterns for some example favorable evolution patients. The first column shows compressive fields for three different patients, where the vectors are randomly oriented. The second, third, and fourth columns depict the magnitude of the vector components in each direction, projected into the main AAA axis to show the dominant direction of the strain patterns. For favorable evolution cases no dominant direction is observed, represented by the lack of red areas, which coincides with the conclusion that the strain vectors in these cases are randomly oriented. Vectors have been sampled every 100 points for visualization purposes.

Figure 8. Compressive strain patterns for some example unfavorable evolution patients. The first column shows compressive fields for three different patients, where the vectors are generally oriented normally to the AAA wall forming a hedgehog-like shape. The second, third, and fourth columns depict the magnitude of the vector components in each direction, projected into the main AAA axis to show the dominant direction of the strain patterns. For unfavorable evolution cases the dominant directions can be clearly observed, which are represented by reddish areas mostly in the region the AAA is expanding. Vectors have been sampled every 100 points for visualization purposes.

Figure 9. Mean tensile strain magnitudes vs. mean compressive strain magnitudes for all the patients in our study. A negative correlation between tensile and compressive strains can be observed, but no relation with patient evolution can be inferred only from the mean magnitude values.

Figure 10. Mean ROC-AUC for all four-fold cross-validation steps and global mean for the 10 runs using only principal components from tensile strain fields (top) and compressive strain fields (bottom).

Figure 11. Mean ROC-AUC and precision-recall curves for all four-fold cross-validation steps and global mean for the 10 runs using tensile and compressive principal components together.

Figure 12. Maximum diameter difference per patient. Patients with unfavorable evolution are shown in blue, whereas patients with favorable evolution are shown in orange.