Research On Brain MR Image Segmentation Method Based On Skew Finite Mixture Model

Accurate analysis and judgment of the distribution of brain tissue is the basic guarantee for doctors to formulate effective treatment plans,and brain image segmentation is the key step in quantitative analysis of brain images.Finite mixture model is one of the most widely used models in image segmentation.However,affected by some factors,there will be intensity inhomogeneity artifact and noise in brain MR images,which leads to the histogram of brain MR images may follow a heavy tailed distribution or asymmetric distribution.Therefore,the traditional finite mixture model,such as Gaussian mixture model,is difficult to obtain accurate segmentation results when segmenting such images.To solve these problems,based on the finite mixture model,this paper proposes two novel image segmentation models.The specific contents are as follows:(1).The article proposes a spatially constrained asymmetric Gaussian mixture model for image segmentation.Firstly,the model uses the structural tensor of the image to construct anisotropic spatial information,which can not only reduce the effect of image noise,but also preserve the detailed information of boundary,corner and so on.Secondly,the model couples the anisotropic spatial information to the mixture skew Gaussian distribution,so that the model can better fit the asymmetric data.Finally,the improved EM algorithm is used to estimate the parameters of the coupled model.(2).In this paper,A spatially constrained Skew Student’s-t Mixture model for brain MR image segmentation and bias field correction is proposed.Firstly,in order to reduce the influence of noise,the model proposes anisotropic two-level spatial information,which combines the nonlocal information of priori probability and posteriori probability to enhance the ability to preserve the edges and details.Secondly,the model couples this spatial information to the mixture skew student’s-t distribution,so that the model can better fit the asymmetric data and heavy tailed data.The improved EM algorithm is used to estimate the parameters of the model.Each method is compared with the related finite mixture model methods on simulated brain MR images and real brain MR images.The model proposed in this paper achieves better results in boundary,corners and slim structures.