Medical Image Segmentation Based On Modified Geometric Deformable Model

Medical image segmentation plays a key role in medical diagnosis, planning and clinical applications. At present, PDE (partial differential equation) based deformable model for image segmentation has been widely used by researchers, and this technique is under development as well. Traditional deformable model based segmentation is a method merely based on edge information, which does not make full use of image information. This thesis investigates and then modifies the existing geometric deformable model by integrating area and edge information and satisfactory result has been successfully reached.In this thesis, the author introduces and analyzes the mathematic basis and application of deformable models. Then a new method merging area information and edge information, geometric deformable model with color and intensity priors for medical image segmentation, is proposed. The prior knowledge used here is firstly represented as thresholds in different color spaces searched by genetic algorithm. Then the prior knowledge is merged into speed function in deformable model with its contour evolution by the level set technique.Experiments are taken on clinical marrow images and mammograms and the segmentation results are evaluated by qualitative and quantitative analysis. These evaluations have successfully demonstrated the superiority of the proposed algorithm over the existing deformable models, which deal with image edge information only, both in accuracy and efficiency.