Segmentation using nnU-Net, registration and meshing enable the elasticity of the ascending aorta to be measured on a zone-by-zone basis; this elasticity decreases with age and the diameter of the vessel.
A team proposes a method to estimate, point by point, the deformation of the ascending aorta during the cardiac cycle from 4D CT angiographies. The processing chain combines automatic segmentation by the nnU-Net network, optimized surface remeshing, deformable registration and deformation calculation on the mesh. To validate it, the authors generated synthetic CT sequences from finite element simulations, whose actual deformation is known.
Reproducible measurement
The median deformation obtained deviates very little from the reference, whether the contours are drawn by an observer or by nnU-Net, and the two sources of segmentation agree closely. Extreme values (95th percentile) remain more sensitive, especially between two human observers. A spectral analysis shows that by retaining only the low-frequency components, segmentation-related disagreements are reduced by 52 to 56%.
Clinically, the highest deformity decreases as the maximum aortic diameter, aortic height index and age increase. Three patients with severe aortic leakage had heterogeneous regional profiles. The authors see this as a non-invasive tool to assess the mechanics of each aorta.
Read the full article on the original source
Go to the source →