Our dataset includes 331 contrast-enhanced abdominal clinical CTimages in the portal venous phase used for pre-operative planning ingastric surgery. Each CT volume consists of 460–1177 slices of512 × 512 pixels. The voxel dimensions are [0.59–0.98, 0.59–0.98,0.5–1.0] mm. A random split of 281/50 patients is used for training andvalidating the network, i.e., determining when to stop training to avoidoverfitting. In order to generate plausible deformations during training,we sample from a normal distribution with a standard derivation of 4and a grid spacing of 32 voxels, and apply random rotations between−5° and +5° to the training images. No deformations were appliedduring testing. We trained 200,000 iterations in the first stage and115,000 in the second. Table 1 summarizes the Dice similarity scoresfor each organ labeled in the 50 validation cases. On average, weachieved a 7.5% improvement in Dice scores per organ. Small, thinorgans such as arteries especially benefit from our two-stage cascadedapproach. For example, the mean Dice score for arteries improved from59.0 to 79.6% and from 54.8 to 63.1% for the pancreas. The effect isless pronounced for large organs, like the liver, the spleen, and thestomach. Fig. 7 shows an example result from the validation set andillustrates the tiling approach. The 3D U-Net separates the foregroundorgans well from the background tissue of the images.