Cerebral Image Processing Based On The Region Of Interest And Its Application

In medical image processing technologies, there is always a very critical issue, which is how to correctly extract the anatomical organization in which we are interested, and that is to say how to precisely segment the medical images. Medical image segmentation is the basic step for the extraction of lesions, quantitative analysis, volume measurement, registration processing, three-dimensional reconstruction and so on in the manipulation of medical images, and it's also the prerequisite for high levels of understanding them. All in all, the accuracy of medical interesting target segmentation will be directly affected the doctor's diagnosis and treatment of the illness.There are extreme similarity and complexity of the different anatomical tissues in medical images. Therefore, the medical image segmentation has always been the bottleneck in various image analysis systems. The segmentation of the region of interest in cerebral images has been one of the hot research problems in medical image processing technology because of its features and very important application value.By analyzing the state-of-the-art research methods, this thesis concludes and generalizes the common medical image segmentation methods and those based on region of interest. According to the characteristics of medical images, we present two fast interactional cerebral image segmentation methods based on the region of interest from the viewpoint of medical application. The main research contents and results of this thesis are as follows.Firstly, this thesis describes the development of medical imaging technologies and applications and its significance in clinical diagnosis, points out the important position of medical image segmentation in the medical image processing area, gives necessity of the segmentation of region of interest in cerebral images, summarizes the current common medical image segmentation methods and the ones based on region of interest.Secondly, a fast segmentation method of region of interest in cerebral image based on isoperimetric algorithm is proposed. Since the image pixels are considered as graph nodes to build a weighted graph in order to implement image segmentation in graph-based image segmentation method, therefore, along with the increase of the image, the algorithm will be more and more time-consuming as well. Generally speaking, it is a NP hard problem. For an image, especially for a medical one, what we often care about is the region of interest and it is only distributed in a small area. While using the isoperimetric algorithm to segment the region of interest in cerebral images, this method reduces the scale of solving the problem effectively, and thus, the time complexity of segmentation was improved and good effect was also gained.Thirdly, a fast segmentation method of interesting space neighborhood in 3D medical dataset is presented. Because many segmentation methods used in 2D image can not be directly used in 3D situation of image segmentation. For all practical purposes of medical applications, with the characteristics of medical images such as rich information, highly complex data, we combined advantages of the threshold method and the regional growth method with 2D slices obtained from 3D medical dataset and its spatial information, and improved them at the same time. As a result, we got the interesting space segmented accurately with high robustness.Finally, combining the practical applications, this thesis designed and carried out a framework of neural surgeon surgery 3D visualization auxiliary application system. By constructing a 3D medical dataset from the standard DICOM files, the system can accurately segment the region of interest and visualize the part. Therefore, it provides an important reference and assistant diagnosis basis for a brain surgeon.