Segmentation Algorithm Research On Brain MR Image Based On Markov Random Field And Bayesian Theory

With the progress of image processing technology, medical image has become more and more important in clinical medicine, and can provide important information to the doctor. Medical image segmentation is a technology to extract the interesting part of medical images for the doctor. Medical image segmentation is the basis of medical image analysis, and the segmentation results directly affect the accuracy of clinical diagnosis. In the study of medical image segmentation, Because of the structure complexity and role importance, the accurate segmentation of human brain medical images is an important research topic. The Nuclear magnetic resonance imaging technique has incomparable advantages than other imaging equipment in diagnosing the soft tissue especially brain diseases, so it was significant to divide the different segmentation of the brain MR image. In the imaging process, the influence of the equipment and the body organs lead to the inherent fuzziness of medical images. In order to fit this kind of fuzziness, the method based Markov Random Field theory is an effective method, but there are some defect of the method. Traditional MRF potential function does not include the distance between the pixel and the pixel intensity factor, but the results of image segmentation are affected by these factors. Likelihood function of traditional MRF-MAP model is to assume that various types of images are Gaussian distribution, however, in fact they do not completely comply with the Gaussian distribution. Based on the Markov Random Field theory and Bayesian theory, an improved method is presented. At first, introduce of Markov Random Field theory and its application in image processing is presented, then the improved algorithm and experiment are also presented in the paper. The main research work and innovations are as follows:1. In this paper, the current research situation and development trend of the medical image segmentation are analyzed. The basic principles of image segmentation and the basic theory of several common medical image segmentation algorithms (such as threshold method, edge method, regional method, mathematical morphology method, method which combined with particular algorithm theory) are systematically expounded. And the advantages and disadvantages of the several common algorithms are pointed out. The evaluation method and the evaluation criteria of image segmentation are reviewed. And the principles, characteristics and clinical applications of the magnetic resonance imaging are introduced.2. Bayesian theory and its application in medicine are introduced. Meanwhile, this paper expounds the basic theory of Markov random field, the equivalent relation between the Markov random field and the Gibbs random field and the MRF-MAP framework. Several basic MRF model, several common optimization algorithm and several common optimal standards are discussed. At the same time, the paper analyses the disadvantages of the traditional algorithm which based on Markov random field and Bayesian theory.3. In view of the disadvantages of the traditional algorithm which based on Markov random field and Bayesian theory, an improved algorithm which based on Markov random field theory and Bayesian theory is proposed. On the basis of the form of Markov random field’s potential function and Bayesian theory’s likelihood function, the paper proposes a new potential function which contains observation field’s grayscale and the distance factor. Then the main advantages and disadvantages of the Gaussian model and Gaussian mixture model are analysed. This paper uses Gaussian mixture model as Bayesian likelihood function to improve the traditional algorithm.From the synthetic image segmentation experiment, it’s shown that the misclassification rate is greatly reduced compared with the traditional MRF algorithm. From the human brain MR image segmentation experiments, it can be seen that the improved algorithm overcomes the disadvantages of incomplete division of the traditional MRF algorithm. So the segmentation accuracy has been greatly improved, more in line with the actual situation. Experimental results show that the proposed algorithm can get an ideal segmentation.