Liver segmentation based on region growing and level set active contour model with new signed pressure force function

Abstract

Segmentation of the whole liver in computed tomography (CT) images is important in liver disease diagnosis. Although the commonly used level set methods (LSMs) and active contour extraction method, are used to segment liver, the segmentation results are not very good. A semiautomatic method consisting of three stages to overcome the leakage and over-segmentation problems is proposed in this paper. Firstly, the original input images are preprocessed by a series of methods to obtain binary images. Then some seed points are set on the binary image for the region-growing to get the initial contour of the liver. The new level set active contour model is built finally to refine the initial liver segmentation. In the first step, the gradient magnitude of original images is smoothed by Gaussian function to suppress noise, and enhance the edges. In addition, a binarization method is put forth on the basis of gradient information of images to adaptively select the optimal threshold without artificially setting parameters. In the third step, a new signed pressure function (SPF) is presented. It can integrate both local information and global information, automatically allocate the ratio of local information and global information based on gradient information of images. The two public datasets SLIVER07, 3Dircadb were used to verify the validity of the method by five measurements. Experimental results show that the proposed liver segmentation is better than other semiautomatic methods based on the SLIVER07 and 3Dircadb dataset and required less interaction.