Deformation Model Of Technology And Its Application In Medical Image Segmentation

The deformable-model-based image segmentation methods, which combine low-level image information with high-level a priori knowledge, have shown unique advantage and comprehensive applicability in the segmentation area of medical images with various styles and complicated structures. How to perfect its theory and how to spread its application in clinics is the main task of this dissertation.A new taxonomy was proposed after the survey of these deformable-model-based image segmentation methods, systematical classifying and combing over the knowledge were done with a new point of view. Fundamentals and implementations of two kinds of basic deformable models, active contour model and active shape model, were discussed in detail. New methods based on these two models which aimed at particular application of medical image segmentation were implemented and satisfying results were acquired. The most innovative work of this dissertation is as follows:1) Systematical classifying and combing over the knowledge of deformable-model -based methods. Though the researches on deformable models were prosperous, the consistent taxonomy was not proposed until now. After a complete survey of these techniques and referring to the others' ideas, a new taxonomy was proposed: deformable models were classified to free-form and constrained-form models. Systematical classifying and combing over the knowledge were done with a new point of view. Major work and its application in medical image segmentation were also reviewed. Fundamentals, main work and typical application of each model were expounded in this frame. Helpful references were provided to understand these techniques.2) Application of active contour model in the segmentation of digital human datasets. Digital human datasets are very large while the adjacent slices have similarity in shape and position. Traditional segmentation methods often focus on single slice and neglect this similarity. A new method was proposed utilizing this similarity, first, locating rods were found through the improved auto-threshold method of color-space and morphology operation, and the origin datasets were transformed...