Liver Tumor Segmentation Based On FCM And Level Set

Image segmentation is based on the characteristics of image, to extract the target of interest in the image. The segmentation of the medical image field, which is meaningful and promising, draws a lot attention of many researchers. This paper, we study liver tumor segmentation based on CT images. Accurate tumor segmentation is playing an important in preoperative evaluation and surgical planning. But how to use algorithm to accurately segment liver tumors, and require less interaction? In this paper, we proposed two segmentation methods of liver tumors based on fuzzy c-means clustering and level sets.Firstly, we introduce and analyze the classic image segmentation methods and some researches specific to liver tumors segmentation. CT images of liver tumors exists many noise, besides liver tumor’s gray level is very close to the liver and the tumor has fuzzy boundaries. So it is very hard to segment the tumor. Traditional graph cut and level set method requires more manual initialization.Secondly, we proposed a new three-dimensional region of interest selection in medical image visualization. We introduce the principle of computer graphics, transform from the object’s local coordinate system to screen coordinates, and its inverse transform. By painting a closed polygon in computer screen, we get the region of interest. The algorithm can reduce the pixels, so that we can speed up the follow-up fuzzy clustering and level set algorithm, and remove unrelated organ.Thirdly, we proposed one liver tumor segmentation based on fuzzy local information C-Means(FLICM) and B-Spline level-set. Fuzzy clustering methods use fuzzy membership to describe tumor’s fuzziness. The traditional fuzzy c-means clustering(FCM) is easily affected by noise. We introduce some FCM methods incorporate with spatial information. Among them we choose the fuzzy local information C-Means(FLICM), which is fully free of the empirically adjusted parameters. Benefited from the three-dimensional ROI selection, we can determine the number of categories into three clusters for tumor images. Because FLICM has serrated edge, we use B-Spline level-set to smooth the edge.Finally, we proposed the other liver tumor segmentation based on fuzzy local information C-Means(FLICM) and Adaptive Distance Regularized Level Set Evolution(Adaptive DRLSE). Use FLICM clustering result as initial level set, utilize tumor’s edge gradient information to evolution the level set. However, due to the initial outline restrictions of DRLSE, the initial contour must be entirely in the external or internal of target. We use adaptive evolution parameter v(I) instead of DRLSE former constant evolution parameter v, well incorporate with fuzzy clustering.Both algorithms take advantage of fuzzy clustering and level set algorithm. FLICM just need the number of clustering categories, can avoid the initialization operation of traditional contour level set. Reduce the interaction, and the experimental result turns out we get a robust and fine segmentation of liver tumor, which has certain practical value.