A Multiphase Segmentation Method For Medical Images Based On Level Set Theory

In the architecture of computer vision,image segmentation is a higher level image processing task based on image filtering,image enhancement and image transformation,whose basic goal is to extract region of interest(ROI)for higher level image information analysis.In essence,image segmentation is the process of assigning a limited number of labels to each pixel in an image,and the set of pixels with the same label are called a sub-region.The generation of these tags and their allocation process mainly depend on the different attribute features selected according to different image segmentation tasks,such as texture,brightness,shape,color and so on.In the final segmentation results,each sub-region has some similar characteristics.Image segmentation is widely used,including target recognition,image reconstruction,target tracking and so on.With the improvement of social production level and technology level,medical imaging means have been constantly enriched,such as positron emission computed tomography(PET),single photon emission computed tomography(SPECT),computed tomography(CT),magnetic resonance imaging.Magnetic Resonance Angiography(MRA),Magnetic Resonance Imaging(MR),etc.As a result of the different imaging devices' working principle,the images obtained by different imaging methods have different characteristics.In medical image processing,we need to focus on the selection of processing methods according to the corresponding advantages and disadvantages.Medical image segmentation based on computer vision is developed on the basis of manual segmentation.Most of the original segmentation methods depend heavily on the gray level distribution of the image,but for the reasons mentioned above,medical images often have uneven brightness,which results in overlapping regions in the gray level distribution of the images,although the structures of different organizations do not overlap in fact,which often cause the consequence that the segmentation results of these methods are erroneous when the multi-phase segmentation puts forward higher requirements for the methods.Therefore,more image information should be taken into account in medical image segmentation.In view of these shortcomings,this paper makes a careful and extensive study on the multi-phase segmentation of medical images,and tries to propose a new region-based level set algorithm to deal with the multi-phase segmentation of non-uniform brightness images.Firstly,a local clustering criterion function is defined according to the brightness of the image in the neighborhood of each point to describe the local brightness clustering feature of the image.By integrating the clustering criterion function with neighborhood center,the global criterion of multi-phase image segmentation is obtained.In the level set formula,the definition of the energy function includes two parts: the level set function of image segmentation and the bias field to measure the non-uniformity of image brightness.By solving the problem of minimizing the energy function,the multiphase image segmentation is performed on the basis of estimating the bias field to correct the non-uniformity of image brightness.The paper is divided into six chapters.The structure is as follows: Chapter 1mainly summarizes the current research situation in the field of medical image segmentation,expounds the purpose and significance of the topic selection;Chapter 2mainly expounds the classical image segmentation algorithms that have appeared so far,roughly classifies them according to the ideas and essence proposed by the algorithms,briefly discusses the basic theory and basic algorithm flow of each kind of algorithms,and sums up the advantages and possible disadvantages of each kind of algorithms in the process of improvement;Chapter 3 expounds the main idea,working principle and modeling process of curve evolution theory,introduces the main ideas and contents of parametric contour model and active contour model,expounds the process of curve evolution theory transforming image segmentation problem into image energy functional minimum problem,and introduces in detail two common solving methods for minimum problem-gradient descent;In chapter 4,a new energy functional framework based on level set algorithm is proposed.The local clustering criterion function is defined according to the image brightness in the neighborhood of each point,and to describe the local brightness clustering feature of the image,based on which the global standard of multi-phase image segmentation is integrated.Then the advantage of the level set method is used to solve the complex topological boundary,and the problem of multiphase segmentation is transformed into the level set function to construct membership.Finally,on the basis of local brightness clustering and level set region segmentation,the image energy functional is defined,and the energy functional minimization problem is solved by finite difference method.The optimal region segmentation and bias field estimation are obtained,thus the problems of non-uniform brightness correction and multi-phase image segmentation are solved simultaneously;In the fifth chapter,the validity of the proposed method is verified by the heart CT images.At the same time,experiments are carried out under different initial conditions to verify the robustness of the proposed method to the initial conditions.Brain X-ray images are used to verify the validity of the proposed method for the multi-phase segmentation of non-uniform brightness images.Finally,the proposed method and PS model method,which is classical,are presented and compared,when the brain CT images and vascular CT images are taken from image samples.The comparison proves that the experimental effect of this method is much better than that of the classical PS model;The last chapter summarizes the overall discussion of the paper,the shortcomings of the proposed method in some aspects,and points out the possible improvement direction.