Breast MR Image Segmentation Algorithm Level Set Theory

Medical image segmentation is one of the importante task in the field of medical image processing and analysis. It also takes much attention of researchers in recent years. The purpose of medical image segmentation is to separate the different special regions, and makes segmentation results close to anatomical structures, so as to provide reliable basis for clinical diagnosis and pathological studies. With the development of the medical imaging, a variety of medical imaging technology, such as X-ray mammography, UltraSound Imaging, CT Computerized Tomography, MRI Magnetic Resonance Imaging, etc., are widely used in clinical. Among them, MRI has a higher resolution and spatial resolution of the soft tissue characteristics, it is a non-destructive imaging technology and can be imaged on a variety of levels, a variety of parameters, multiple sequence and can provide more information. So, it plays an important role in early detection and treatment of breast cancer. Therefore, breast MR image processing and analysis becoming one of the important contents in medical image processing and analysis. As one of the key steps of breast image segmentation has thus becoming a hot topic.There are some difficulties in the existing segmentation methods for the segmentation of breast MR image, for the presence of large amounts of information, organization’s complex, low-contrast, weak borders and intensity inhomogeneities. Among image segmentation methods, active contour model has been widely concerned. This model completes segmentation of image through solving the minimization of it’s energy, and it is produced through a procedure of curve evolution. Active contour model includs parametric active contour model and geometric active contour model (referred to as the level set method). Parametric active contour model, which represents curves and surfaces explicitly in their parametric forms. This kind of model is simple, but can’t adapt topology changing during evolution. The level set method can represent curves and surfaces implicitly with zero level set,which is obtained by updating the level set function, and then complete the segmentation of image. The main advantages of level set method are:1) adapt topology naturally during evolution.2) level set method segments image, which is essentially to solve a partial different equation and can be easily extended to random higher dimensional case.According to the current situation of domestic and foreign research, there are still many problems of level set theory and its applications need to be solved urgently. Because of the extensive invasion of breast disease and for the assist of clinical diagnosis, the research of breast MR image segmentation is particularly important. Then, this paper will research level set theory and it’s application on breast MR image segmentation.The main work are as follows:(1) Analyzing and discussing the level set theory including curve evolution theory, summarize the research situation of medical image segmentation with the level set method.(2) According to the characteristics that the traditional level set algorithm’s sensitivity to noise, re-initialization of the level set function, and breast MR images’s low SNR, we researched continuous level set algorithm. In the algorithm, Linear combination of B-spline function represents level set function, shifted Heaviside functions with adaptive parameter is introduced to the troditional CV model, and α-CV model is constructed to control the evolution of the level set. The convolution filter of b-spline basis function will smooth noise, it’s coefficient can simplifie the calculation process. And it will effectively avoid the re-initialization of the level set function. The shifted Heaviside functions will prevent local minimum. The algorithm makes full use of the regional information of image, and it can better deal with the weak boundary breast MR image segmentaion.(3) In the presence of intensity inhomogeneities, we researched a multi-resolution level set segmentation algorithm based on the theory of wavelet transform. The main idea of the algorithm is as fallows. Firstly, getting the rough dimension image by using wavelet multi-scale decomposition to analyse the image on multi-scale space. Secondly, segmenting the result with improved CV(W-CV) model. Finally, we introduce Kernel function, produce CV(KW-CV) model to deal with the effect of bias field on the segmentation of breast MR image. The algorithm avoids the dependence on initial contour, accelerates the segmentation speed and guarantees the accuracy of segmentation.