Figure 4
From: Deep-learning-based extraction of circle of Willis topology with anatomical priors
Comparison between path localization using our proposed framework using \(\mathcal {C}_{\text {orient}}\)(middle column) and skeletonization on a 3D U-Net segmentation (right column). The 3D U-Net does not find continuous trajectories for hypoplastic arteries, while our method finds a continuous path by design. Our method sometimes traverses the wrong artery segment. Both errors are indicated with a white arrow.
