A boundary optimization scheme for liver tumors from CT images

Liver CT scan image analysis plays an important role in clinical diagnosis and treatment. Accurate segmentation of liver tumor from CT images is the prerequisite for targeted therapy and liver resection. Existing semi-automatic segmentation based on graph cuts or fully automatic segmentation methods based on deep learning have reached the level close to that of radiologists. To improve tumor segmentation on liver CT images, we propose a simple post-processing scheme to optimize tumor boundaries. This method improves boundary prediction performance by optimizing a sequence of patches extracted along the initial predicted boundary. The proposed boundary refinement segmentation network obtains strong semantic information and precise location information through the information interaction between different branches, to achieve precise segmentation. The Liver Tumor Segmentation (LiTS) dataset is used to evaluate the relative segmentation performance obtaining an average dice score of 0.805 using the new method.