Research And Implementation For Image Segmentation Techniques On Brain Mr Images

Magenetic resonance(MR) imaging plays an important role in clinical medicine, and has become an important assistant method for clinical diagnosis of various diseases such as brain pathologies, etc. However, the bias field is produced during the process of MR imaging, which leads to the intensity inhomogeneity. Hence, it is difficult to obtain the accurate segmentation results of MR image using intensity-based methods. The work in this thesis is shown as follows:Against the fact that bias field is the main reason of intensity inhomogeneity of the brain MR images, a set of legendre basis functions is constructed to fit the smoothness bias field to overcome the intensity inhomogeneity; then the information of the bias field is added to the Gaussian density function; and according to the advantages of the mutual information, the energy function for the brain MR digital image segmentation and bias field correction is defined. In addition, the energy function is incorporated into a three-phase level set framework to form a unified variational model for the optimizing of brain MR digital image and bias field correction. Simulation results tell that the proposed scheme can chieve a good segmentation result. What’s more, based on the advantages of graph theory and region growing, a image segmentation model for brain MR digital image combining graph theory with region growing is presented, where, initialize the subregions of brain MR image using OSTU segmentation algorithm, introduce the concept of fuzzy connected graph and obtain target area of each region in MR image by utilizing region growing segmentation algorithm. The experimental results show that the segmentation algorithm can segment the organization of brain MR image accurately.Based on the interactive operation of GUI in Matab, a system for brain MR digitial image segmentation is constructed. There are friendly interface and simple operation on the system. The testing results demonstrate that the proposed brain MR digital image segmentataion method can obtain accurate segmentation results.