Research On Medical Image Segmentation Using Geometric Deformable Models

Medical image segmentation is the essential step of medical image processing, and it plays a crucial role in the virtual surgery system. The common segmentation model cannot satisfy the medical image segmentation due to its complexity. A new segmentation algorithm named deformable model become well known, which is useful in this problem. This paper aims to do some research on deformable model, especially the geometric deformable model.Level set is an effective method for curve or surface evolution in two spatial dimensions or three spatial dimensions. This paper do some research on the principle of level set, and introduce some of its method.There are basically two types of deformable models: parametric deformable model and geometric deformable model. Researching on this two models, and have a comparison, we find the latter is more fit for the medical image segmentation, because of the geometric deformable model, which is based on curve evolution theory and level set, can treat with the change of topology naturally.Using the deformable model in image segmentation, we must choose a segmentation model, in order to form the energy field attracting the evolving curve to the boundary of our desired object. This paper introduces some segmentation models, and then discusses the kernels of level set procedure detailedly. During this process, we improve the fast marching method, and reduce the computing time to O(N).Mumford-Shah model is popular these years. The M-S model overcome the shortage of the classical segmentation model by using the global information of image to make curve stop at the edge of the object, and can detect objects with very smooth boundary or even with discontinuous boundaries. Our model has a level set formulation, interior contour are automatically detected, and the initial curve can be anywhere in the image. More over, a fast segmentation model was introduced, which combines the M-S model and narrow band scheme. The experiments show that the our model can obtain almost the same segmentation result as the original M-S model.At last, a kind of simple smooth scheme is used to improve the stability of our model.