Research And Implementation On Liver Segmentation Methods Based On Statistical Shape Model

Medical image segmentation is basic work in the field of life science research and clinical application. In computer aided diagnosis (CAD) field, accurate and stable liver tissue segmentation of liver is prerequisites of disease diagnosis, research and surgical planning. However, the complex human anatomy abdomen makes the liver segmentation more difficult. In this thesis, we construct statistical shape model (SSM) of livers to be applied to the level set segmentation algorithm for the early diagnosis and help doctor make right decision.In this thesis, a statistical shape model of livers is build referenced on large of classical image segmentation algorithms. In shape alignment, an adaptive sampling method is proposed to improve the original MDL approach to solve the uneven sampling problem. The approach combines original MDL with particle method, making feature points evenly distribute on the surface. We also proposed a level-set method based on SSMs for image segmentation. The algorithm created mean liver shape scaled certain times placed within the image to be segmented, as the initial boundary of target area which is initial zero level-set curve, according to the improved level-set evolution to get the precise boundary of liver. First constructed statistical shape model of liver by using a large number of training samples, using implicit function represents a priori which is the model changes in the shape, and then introduce the energy function as a priori shape of the shape model, minimize the distance between the shape of the model and the evolution curve. The level-set segmentation method guided by a shape priori not only has the implicit surface constraint consistent with the overall shape but also maintains the level-set local deformation characteristics.Through the experiment of training and testing data sets and verification, evaluation and analysis of the results proved that the proposed algorithm improves the level set function of the initial contour position-sensitive features, and the final boundary of the segmented liver is smoother, effectively avoiding omitting segmentation and over segmentation phenomenon. Meanwhile, the improved level-set method faster 3-4 times than the classical evolutionary, improves segmentation efficiency.