Brain MR Image Segmentation Based On Level Set Method

Image segmentation refers to the technology and process of extracting the region of interest in the image taking advantage of artificial and computer-aided means.Image segmentation is essential to image analysis.And the segmentation results are also good for the clinicians to diagnose and analyze the illness in clinical practices.With the advantages of non-ionizing radiation damage,high resolution for soft tissues,multi-parameters and formation of image in any direction,Magnetic Resonance Imaging(MRI)becomes one of the most important detection means of pathological tissues clinically currently.However,due to the universal phenomena such as the noise,artifacts,intensity inhomogeneity of MR brain images,it is very difficult to make an accurate segmentation of MR brain images.Considering the above situation,this thesis completes the segmentation of MR brain images in medical science in two phases.At first,it makes a reduction of the noise to the images,reserving as many edges and details as possible.Then it completes the segmentation of the MR brain images using the gray-level information of the images.The main tasks in the thesis are as follows:(1)Firstly,it explores the principles and characteristics of three image filtering technologies: median filtering,gaussian filtering and anisotropic diffusion filtering,then compares the results and evaluations of the segmentations of MR brain images with the three filtering technologies using the index of peak signal to noise ratio.It turns out anisotropic diffusion filtering is of good performance in filtering out the noise and retaining the edges.MR brain images have abundant of edge details as well as the noise generally.Therefore,the anisotropic diffusion filtering is selected as the filtering technology for MR brain images.Meanwhile,the analysis shows that anisotropic diffusion filtering algorithm fails to filter out the isolated noise spots.With the introduction of the edge detection operators into the diffusion operators in the original algorithm,the improved algorithm in this thesis and original algorithm are then both applied to filtering the MR brain images.The experimental results and Jaccard index all verify that the improved algorithm put forward in this thesis is able to solving the problem of filtering out the isolated noise spots for the original algorithm while reserving the advantages of the original algorithm;(2)Secondly,it introduces the theoretical basis of level set method and analyzes C-V model,local binary fitting(LBF)model and the energy penalizing term in Tang Liming model in great details: C-V model only considers about the internal and external global gray-level information of the images in the evolving curve,failing to handle the images with great intensity inhomogeneity though the segmentation effect is ideal when it is used to segment the images with intensity homogeneity;LBF model only covers the local gray-level information around the evolving curve,so it has advantages in the segmentation of images with severe intensity inhomogeneity,but it can fall into local minimum value easily;Tang model doesn’t need initialize the level set function again,and there is a lower computation complexity,but it fails to segment partial MR brain images.Finally,it gets the Jaccard scores in the segmentation of MR brain images with the three models respectively;(3)Lastly,the thesis comes up with a new active contour model which integrates with the global and local gray-level information to solve the problems of Tang model cannot segment partial MR brain images correctly.This model combines with the merits of Tang model,and builds the global energy term,local energy term and energy penalizing term by means of the global,local gray-level information and the information of level set function.Besides,the adaptive coefficient is introduced in this active contour model to adjust the ratio of the global and local gray-level information in the curve evolution,guiding the curve to stop at the targeted edge.At last,the active contour model is applied to segment the MR brain images,and is used to make a comparison with Tang model.The Jaccard scores prove that our method put forward in this thesis gets a better performance in the segmentation of the MR brain images as Tang model.