The Application Of Deformable Model In Medical Images Segmentation

Deformable model is playing a more and more important role in image segmentation owing to its unique advantages. This paper aims to do some research on medical image segmentation with deformable model.GVF (Gradient Vector Flow) model has resolved two key problems of the traditional deformable model. However, it still requires both the initial contour being close to the target and a large amount of computation. Wavelet transform can analyze signals in multi-scale and the local maxima of wavelet transform represent the edge of the image. The quadratic B-spline wavelet performs superior than the other methods in edge detection, and the lifting scheme has been the key for designing the second-generation wavelet, so the multi-scale GVF model based on the B-spline lifting wavelet is proposed here to solve the problem of GVF model. This model can segment complex structures by taking the advantages of GVF model and it reduces the sensitivity of GVF model. The lifting scheme can also simplify the computation at the same time.Although the geometric deformable model based on level set method has a distinguished property of dealing with topology change, it still apt to be interfered by noise. Traditional Gaussian filter blurs the details of the image when it smoothes the image. The anisotropic diffusion not only can keep the edge of the image but also can enhance the edge. The symmetric exponential filter is used to improve the anisotropic diffusion and an adaptive estimating method of K is introduced, so the geometric deformable model based on this improved anisotropic diffusion is proposed here, it can make the geometric deformable model based on level set method more robust to noise.Mumford-Shah method proposed by Mumford and Shah can deal with the leak problem of level set method. Chan and Vese proposed a simplified level set solution of Mumford-Shah method, further they proposed multi-phase Mumford-Shah method based on multi-level sets to segment the complex image structure, but this method costs a lot of computation and the initialization is difficult. The hierarchical multi-phase C-V method combined with a narrow band here is composed of several steps, every step needs just one level set, next segmentation step is to be executed afterthe current step has been finished, and in every segmentation step only the points of narrow band need to be treated. This method can segment the complex structure, and the computation has been greatly reduced compared with the multi-phase C-V method.