The Human Brain Image Segmentation Based On Markov Random Field On Ds Evidence Theory

With development of medical systems, medical imaging has become a new subject, and morever, the medical image segmentation in medical image analysis has been playing an increasingly important role, which is widely used in medical research fields.In essence, medical image segmentation is marking all kinds of voxels in different medical images into reasonable classes, it is the segmentation results after labeling the image,medical image segmentation is the basis of organization measurement,3D reconstruction, and image matching, at the same time,it has played a certain role of adjuvant therapy for clinical diagnosis. Research on human brain in medical imaging has become a priority, because of the brain is the central nervous system of human life and activities. However the essence of brain magnetic resonance images is ambiguous because of the complexity and irregular brain structure, as well as the uneven nature of the magnetic field during imaging, etc. so for these features of the brain MRI, we usually use Markov Random Field and Fuzzy Clustering theory as well as the Dempster-Shafer evidence theory to study image segmentation. This article studys the brain MRI segmentation based on these three theories, this article reads as follows:1.Analysis the current situation and development trend of the domestic and foreign medical image segmentation, describing the difficulty and significance of medical image segmentation. Common method used for medical image segmentation was introduced, and then segmented the brain MRI by common segmentation methods so that compared the results.Introducing the basic knowledge of medical imaging, in particular the imaging principles of computed tomography and magnetic resonance imaging technology, and describing the imaging characteristics and clinical application.2.Study the Markov Random Field and Gibbs Random Field on the brain MRI segmentation deeply. Based on the Markov Random Field theory,prior function is equivalent to Gibbs Random Field,creating a Markov posterior energy field in the brain MRI according to the spatial context information of MRI, and then combining Two-Dimensional histogram of Fuzzy Clustering method to study the brain MRI segmentation and extracting the white matter and gray matter. Finally discussing the advantage and deficiency of segmentation results.3.Markov Random Field can take the relationship between the neighborhood pixels into account and uses the spatial characteristics of the images, accordingly,using it for segmentation can obtain better segmentation results.As the Two-Dimensional histogram of Fuzzy Clustering method integrates clustering theory and correlation of pixels, so an ideal segmentation results can be obtained. However, these two ways leads to different classification results while classifying the controversial pixels in images,due to Dempster-Shafer evidence theory can integrate information of different sources images, so extracting the label image and redundant image from the segmentation results of these two methods,and then using the Dempster-Shafer evidence theory fuses the original image and the label field image as well as redundant image in accordance with the integration rules of the Dempster-Shafer evidence theory to get the final segmentation results, it resolved classification problem of the dispute pixels.4.Markov Random Field is used to image segmentation,the most difficulty is how to choose reasonable parameters so that the image modeling and segmentation is more reasonable and accurate,including the potential function form and the selection of the direction functionβas well as optimization of the likelihood function,etc.so this article gave some definition methods of parameters, and proposed "Typing"method to define the potential function and the direction functionβ,it can describe the spatial relation of local details and neighboring pixels and local correlation,and improved the likelihood function by introducing membership degree as fuzzy likelihood function,thus achieving Markov Random Field parameter estimation,it made the segmentation results have a clear boundary between categories,much more in line with the actual situation.