Research On Liver Segmentation Methods Based On Improved Snake Model

Liver is the largest internal organ for human beings. It has great diagnostic and research significance because of its complex structure and high morbidity. Image segmentation is of significant importance in medical image processing. Three-dimensional visualization of specific organs, computer-aided surgery and radiological treatment are all based on image segmentation. For example, when we do research on liver identification and three-dimensional liver visualization, the first step we need is exact the accurate liver segmentation. Because the organs and tissues have complex structures and similar density, together with the limitation of imaging devices, in two-dimensional abdomen CT images, a liver is shown with complex structure, low contrast, adhesion to other tissues, and blur edges. And also, a liver appears different in different slices of CT images. These factors increase the difficulty of liver segmentation.Segmentation algorithm based on deformable model has become more and more active in the domain of medical image segmentation. The basic idea of this algorithm is to construct an energy function of the model to make the curve or surface evolve under the inner control force and the outside image force of the model. When the energy function is minimized, the evolving curve or surface reaches the target. The advantage of the segmentation algorithm based on deformable model is that it concludes image data, initial contour and target contour into a uniform mathematical model. While the disadvantage of the algorithm is that the deformable model is sensitive to the initial contour and difficult to detect concave boundaries. In this thesis, we proposed an approach for automatic liver segmentation for sequential abdomen CT images based on Snake model and region growth. First, image format conversion, image filtering and image enhancement are implemented in the part of image preprocessing. Then considering that the liver may appear in not only one region but many regions in CT images, an automatic method combined the liver's location and area information with basic methods of image segmentation is proposed to judge the num of liver regions. After that, for the single liver region, an improved Snake model is used to segment the liver. The improved Snake model can automatically extract the liver's contour by combining threshold method, morphology operation, edge detection and contour tracking and can avoid its sensitivity of initial contour. And for the multi liver regions, a modified region-growing method is used in this paper. Based on the results of liver region judging, an automatic seeds selection approach is implemented. Considering over segmentation may be appeared in the process of region growing, a modification to the results of region growing is taken to solve the over segmentation problem.In this thesis, we evaluate the segmentation performance by compare automatically segmented areas and manually segmented areas by doctors. The application of the approach proposed in this thesis to ten sets of abdomen CT images with blur edges for liver segmentation proved the adaptability and effectiveness of our approach for complex image segmentation.