Abstract: Aiming at the challenges, such as fewer labeled samples and expensive manual annotation in medical images, a network of liver CT(Computed Tomography) images segmentation model based on semi-supervised ladder is presented. First, the input data is reduced by super-pixel segmentation. Next, the patches are
extracted around the center of pixels, and the patches are used to train a semi-supervised model. Finally, the trained model is used to achieve liver segmentation. Experiment results show that a small number of labeled pictures are able to obtain similar results with supervised learning.

Key words: ladder network, medical image segmentation, semi supervised learning, super-pixel