| MRI, better than CT, has a higher soft tissue contrast, which is more conducive to find brain lesions. The positioning of the tumor and surrounding tissue structures show very clearly, and the direct use of MRA can show the structure of brain blood vessels. Basing on the characteristics of magnetic resonance imaging, this article shows the study over the two brain diseases -the brain tumors and cerebrovascular stenosis. While we design a computer-aided diagnosis system to help doctors detect and treat brain diseases. For the analysis of brain tumors, methods include Level Set, Fuzzy C-means clustering and threshold analysis basing on difference transformation, which can get tumor segmentation with different characteristics. After achieving each algorithm, this article analysis and compares the results, and discusses the advantages and disadvantages of each algorithm and improvement methods. At last, we propose two improvement methods. The design of cerebrovascular stenosis analysis system bases on vascular stenosis analysis in the DSA. According to the characteristics of cerebral brain in MRA (Magnetic Resonance Angiography), the system has been adjusted by adding mew methods, which can measure the vascular stenosis rate of selected region accurately. Methods of brain tumor segmentation in this article:(1) Geometric active contour model-Level Set Evolution Without Re-initialization, Mumford-Shah Model, Ameliorative Tumor Segmentation Level Set. Level Set Evolution Without Re-initialization algorithm utilizes that the gradient of level set is a constant to incorporate the evolution with inner energy, which can find out the edge of object by minimize the energy. However, due to various patterns of brain tumors, this algorithm does not have a good applicability, since it will depends on the gradient which is external energy. Simplified M-S model considers the gray-scale features of image and brings the gray-scale difference of segmentation area into the iterative step, while the result will be more satisfactory. Ameliorative Tumor Segmentation Level Set add region grow before Level Set to find original contour and morphologic remedy after Level Set to get the final result. Fuzzy C-means clustering is a clustering algorithm based on the division. Its idea is that it makes objects divided into the same clusters with the greatest similarity, and different clusters with the minimum similarity in order to achieve the tumor segmentation. (3) Threshold analysis basing on difference transformation utilizes the MRI imaging characteristics to make a transform between T1 and T2 weighted image, so signals contrast of different organ have been enhanced in transformed images. (4)Combination FCM with Level Set algorithm use FCM to find the original edge firstly, and then use M-S model to find the accurate object edge. This combination can promote the accuracy and stability, decrease the running time and simplify the manual operation. Cerebral vessel stenosis analysis basing on MRA imaging includes:Gauss filter, enhancing vessel using Hessian matrix, binaryzaiton using threshold, extracting vessel centerline, measuring vessel diameter and emending error, finding correct position to protract diameter and vessel edge. Applying all the method in various kinds of brain tumor images, the combination FCM with Level Set achieve the best effect, which can adapt the capricious features of brain tumor. Meanwhile, the cerebral vessel stenosis analysis system acquires excellent evaluation. |
PDF Full Text Request