Study On Segmentation Of Liver On CT Images

Recently, with the incidence of liver cancer rising, early diagnosis and accurate surgical planning is important for the treatment of liver cancer. Accurate liver segmentation is a crucial step for computer-aided liver disease diagnosis and surgical planning. The results of liver segmentation can be used in quantitative analysis of liver tissues, image registration and three-dimensional liver modeling which has significant academic research and applied value.In this paper, a new coarse-to-fine method is proposed to segment liver on abdominal CT images. This new segmentation method makes full use of the space continuity of CT slices and achieves the liver segmentation of a full sequence of images. The main contents of this paper contain the following three aspects:(1) Pre-processing methods of CT images and post-processing of segmentation results. Pre-processing includes format conversion from DICOM to BMP and the de-nosing method by median filter. Post-processing includes image smoothing by morphological operations, filling holes and extracting the largest connected region on binary image.(2) The rough segmentation is implemented based on a kernel fuzzy C-means algorithm with spatial information algorithm(SKFCM) which is firstly used to segment liver. Because of introducing kernel function and spatial information into fuzzy C-means clustering algorithm(FCM), both the clustering ability and noise immunity of SKFCM algorithm is stronger than FCM’s. In abdominal CT images, the shape of liver between adjacent slices has a little change. With this feature, the rough segmentation of liver can be achieved continuously. About half of results of rough segmentation do not need refined segmentation which can be used to generate seed template in the refined stage.(3) Refined segmentation on results of rough segmentation is performed based on the proposed improved GrowCut algorithm(IGC). Traditional GrowCut algorithm is an interactive segmentation algorithm, segmentation results depend on the choice of seeds. We propose an improved GrowCut algorithm which can generate seed labels automatically. The improved GrowCut algorithm improves the efficiency and accuracy of segmentation.The proposed coarse-to-fine segmentation strategy is tested on three sets of abdominal CT images. The qualitative analysis and quantitative evaluation of segmentation results show that the proposed segmentation method of liver is accurate and efficient.