Brain Tissue Segmentation Of MRI And Its 3D Visualization

Magnetic resonance imaging(MRI)technology is suitable for brain imaging,the accurate segmentation of MRI brain has important medical significance for clinical brain disease research and treatment.The Gaussian Mixture Model(GMM)can effectively describe the slowly changing characteristics of brain tissue intensity distribution in MR images.It is the most commonly used statistical model for MRI brain tissue segmentation,but the GMM segmentation model itself does not consider spatial information,and this causes the GMM model to work only on well-defined images with low levels of noise;unfortunately,this is often not the the case for practical applications.Spatial information is usually introduced in the GMM segmentation model;or get rid of the GMM,and directly model the spatial relationship of the voxels to improve the model.The Hidden Markov Model(HMM)considers the spatial correlation of the neighborhood voxels and can simultaneously model the statistical and spatial characteristics of the image.However,most of the current HMM-based segmentation algorithms use global fixed parameters and cannot reasonably select both anti-noise and detail preservation.Therefore,the main research content of this paper is to improve the GMM segmentation model and the HMM segmentation model,introduce the spatial information into the GMM segmentation model,and improve the parameter estimation strategy of the HMM segmentation model.The main contents of this theis are as follows:1、There are two main ways to introduce spatial information in the GMM segmentation model: tissue probability map and spatial neighborhood relationship.This paper first analyzes the advantages and disadvantages of these two methods.The former can avoid the EM algorithm falling into the local optimal solution and retain the image details;however,it does not consider the fact that the large probability of neighborhood voxel belongs to the same tissue lable,so the segmentation result at the edge of the different tissue is not accurate enough.The latter can improve the anti-noise ability of the algorithm,but there is a problem that the segmentation result is too smooth and the details of the edge of the tissue are lost.2、In order to further improve the segmentation accuracy of the algorithm and improve the segmentation result,this paper proposes a TPM-SVGMM algorithm that combines the tissue probability map and the spatial neighborhood relationship.Adding constraints priori of tissue structure in the local spatial neighborhood relationship,giving full play to the advantages of the two spatial information,suppressing noise while preserving the edge detail information and improving the accuracy of tissue segmentation.3、Compared to the GMM model,the HMM model is spatially dependent.At present,most of the HMM segmentation models use global fixed parameters.Excessive parameters will appear to be too smooth locally,which will result in the loss of image detail information.Too small parameters will reduce the anti-noise ability of the algorithm.To solve this problem,this paper proposes a brain tissue segmentation algorithm with local parameter adaptive adjustment: PAHMRF algorithm.Experiments show that the algorithm is suitable for segmenting brain images containing high noise and has the advantage of retaining more detailed image information.4、Finally,this paper analyzes the proposed brain tissue segmentation algorithm by means of three-dimensional visualization.The cerebrovascular information and brain tissue information of the same experimenter obtained by segmentation,and three-dimensionally fused and displayed.