Figure 1.

Comparison between manual and automatic segmentation results of the validation set. The 3D reconstructed structures show liver parenchyma (A and B), tumor mass (C and D), hepatic vein (E and F), portal vein (G and H), and bile duct (I and J). The manual segmentations are denoted as ground truth (A, C, E, G, I), while the automatic segmentations by the 3D residual U-Net model are denoted as inferred images (B, D, F, H, J). The first column of each row on the y-axis represents the number of patient cases. The images in this figure were created using 3D Slicer v.5.02 (http://www.slicer.org).