Automatic Liver and Tumor Segmentation of CT Based on Cascaded U-Net

Automatic segmentation of liver and tumor plays a crucial role in medical-aided diagnosis. At present, neural networks have been widely used in medical image processing. There are many FCN-based methods used for the automatic segmentation of the liver and the tumor, but results are not precise enough to the details in the images. In this paper, we use cascaded U-Net to segment livers and tumors automatically. The first U-Net is used to segment livers, and the livers are the input of the second U-Net. We perform experiments on the published 3DIRCAD dataset and the dataset provided by medical institutions. Medical institutions provide CT of patients with advanced liver cancer. Compared with FCN, U-Net is more accurate. When the false positive rate is the same, U-Net’s true positive is higher. The accuracy of segmentation of the liver is 91.3 and 89.8%, respectively, and the accuracy of segmentation of the tumor reaches 82.4 and 86.6%.