Incorporating both shape and topological priors into a unified latent representation improves the accuracy and anatomical consistency of automated liver vessel segmentation in medical images.
