| Image segmantation is a process that partition the image into different non-overlapping regions, which are called region of interest (ROI). The magnetic resonance image (MRI) technique has played more and more important role in the medical diagnosis and treatment. Segmentation for MRI is the basis for organ diagnosis, tissue quantization, pathalogy factor analysis, and disease developement tracking, etc. However, as the result of the instability of radio frequency and the existence of the imprurity, the intensity inhomogeneity and noise is induced during the process of imaging. The intensity inhomogeneity is the phenomenon that the intensity in the same ROI varies spatially. And for the intensity inhomogeneity in MRI, the intensity varies slowly. The segmentation models assuming that the image intensity is homogeneous may obtain the inaccurate results when the images are intensity inhomogeneity. Besides that, there is usually partial volume effects in the medical MR images. The partial volume effects leads to the low-contrast and uncontinuous boundaries in brain and vessel MR images, which are difficult to extract. The components of the medical images including the textures, the noise, the irregular intensity inhomogeneity, and the complex topology structures of the edges brings about difficulties for image segmentation.In this paper, we focus on the research of image segmentation for medical MR images to solve the problems given above. The primary work and the remarks are as follows(1) For the images with multiple objects, intensity inhomogeneity, and noise, a multiphase Chan-Vese model based on an improved fuzzy c-means algorithm is proposed. First, the classes of the intensity are calculated based on the histogram statistics, and the spatial information computed in the neighborhood revise the grade of membership. The improved FCM algorithm applied with the region fitting term of CV model, working as the reliance of evolving the level-set curve. Anisotropic local template is then used to handle the different objects so as to control the split-up of the contour accurately and segment more objects in less time.(2) For the components of the cartoon part and the texture in images, a segmentation model is proposed for both the two kinds of the components. Two kinds of region data terms are designed for detecting cartoon and texture parts respectively. The local statistic information is extracted in the adaptive patch to solve the over-segmentation induced by the intensity inhomogeneities. And the texture feature information calculated in the adaptive patch is utilized to compute the Kullback-Leibler distance for detecting the texture part. The proposed model is solved by the split Bregman method for efficiency. Experiments are carried on both medical and texture images to compare our approach with some competitors, demonstrating the precision and efficiency of the proposed model.(3) A novel active contour model with image structure information and multiple statistical information for image segmentation. The intensity inhomogeneity varies spatially. To solve this problem, this model utilizes an improved region fitting term to partition the regions of interests in images depending on the local statistics regarding the intensity and the magnitude of gradient in the neighborhood of a contour. In particular, integrated with the duality theory and the anisotropic diffusion process based on structure tensor, a new regularization term is defined through the duality formulation to penalize the length of active contour. By this new regularization term, the structure information of images is utilized to improve the ability of capturing the geometric features such as corners and cusps. From a numerical point of view, we minimize the energy function of the proposed model by an efficient dual algorithm, which avoids the instability and the non-differentiability of traditional numerical solutions, e.g. the gradient descent method. Experiments on medical and nature images demonstrate the advantages of the proposed model over other segmentation models in terms of both efficiency and accuracy.(4) For the images with noise, intensity inhomogeneity and the concave structures, a robust patch-statistical active contour model is proposed. The intensity inhomogeneity and noise are both considered as the irregular intensity variation. The patch-statistical region fitting term computes the local statistical information by Nadaraya-Watson operator in each patch as the basis for driving the curve accurately with resist to the intensity inhomogeneity and the weak boundaries. And the regularization term coupling with the gradient information improves the ability of capturing the boundaries with cusps and narrow topology structures. Furthermore, an intensity variation penalization term is proposed to overcome the negative effectiveness of the irregular intensity variation. Experiments on medical and nature images show that the proposed model is more robust than the popular active contour models for images with noise and intensity inhomogeneity. (5) It is a challenge to extract the vessels and the ends of the tissues in the MR images because of the partial volume effects, noise and the intensity inhomogeneity. Consequently, the active contour models based on the structure information of images are proposed. Coupling with the duality theory and a structural gradient vector flow (SGVF) method, we formulate a new regularization term of the level set function via a duality formulation to penalize the length of active contour. By this new regularization term, the structural information of images is utilized to improve the ability of preserving the elongated structures in the MR images. The experiments on brain and vessel MR images demonstrates that this model incorprates the advantages of both edge-based and region-based active contour models and preserves the enlongated structures in MR images. |
