Analysis And Visualization Of Brain MR Image Based On Multi-Atlas

The human brain is a highly developed organ. It controls and regulates all the physiological activities, including thinking, awareness and other senior activities. There are multiple structures in human brain. It has important clinical significance to study the structures of the human brain and many scholars attach great importance to brain science research. Magnetic resonance image has the advantage of non-invasive and high contrast of soft tissue and it plays a more and more important role in clinical diagnosis. Centre on the brain MR image, this paper uses multi-atlas registration method to study three aspects of content:The human brain is a very complex soft tissue and its individual differs much from each other. Besides, the magnetic resonance technology produces magnetic field intensity inhomogeneity, random noises and cerebrospinal fluid pulsation. Because of these, it is more difficult to segment human brain MR image. The atlas registration method uses the prior knowledge of image. It registers the segmented template image to the target image for the transformation parameters between them. Then, it uses the transformation parameters to map the structures of interest to the target image to achieve the aim of the segmentation of target image. Sometimes, a single atlas differs much from target image and it is difficult to get ideal segmentation results. This paper registers multiple atlases to the target image and fuses the results of each atlas segmentation. At last it can get the final segmentation result. The key of multi-atlas registration method is fusion strategy. The paper proposes a new fusion method. The method regards the similarity between atlas and target image as the weight of fusion strategy.The method of segmenting brain structures based on atlas belongs to binary judgment. The segmentation result on boundary surface based on the method is not accurate. Volume rendering technology can display the boundary surface of structure obscurely. The design of the transfer function has been the difficulty in volume rendering technology. It remains a challenging problem to design an appropriate transfer function for a practical 3D image containing multiple different structures such as human brain. Based on the theory of multi-atlas registration, this paper proposes a new method to design transfer function for human brain MR image. The method regards the probability values of the fusion after multi-atlas registration as the opacity to design transfer function. Then, this paper uses fuzzy c means algorithm to improve the design of transfer function which is based on multi-atlas registration. The experimental results show that the proposed method can effectively classify the interest in human brain.In clinical diagnosis, it can provide more useful information for the doctors for further diagnosis and treatment if we can judge whether the interested brain structures belong to lesion areas and determine the ownership structure of the brain lesions. This paper uses atlas registration method to analyze textural feature difference between multi-atlas and target image. This method can analyze the possibility of lesions in interested brain structures. Then this paper uses atlas registration method to map the lesion area which is determined by artificial delineation to the template image. According to the corresponding relation between the labels in template image and brain structures, the method can determine brain structures which the lesions belong to. At last, this paper studies the registration and fusion technology between ASL(Arterial Spin Labeling) and ECT(Emission Computed Tomography). The fusion image of these two kinds of images is helpful in clinical diagnosis of regional cerebral blood flow and can provide a good basis for the diagnosis of ischemic vascular disease.