Rethinking 2D segmentation network to segment CT scans

Heonjin Ha

Keywords: 2D segmentation, KITS21 dataset

Abstract: This work experiments with how 2D segmentation models work to segment CT scans. To overcome the lack of spatial information, the stacking-slices method is devised. This performs similar to 3D segmentation models like 3D-UNet. Moreover, this work suggests a new post-processing technique, and this can keep the organs in the background from being removed.