【作者】 姜红；

【导师】 张兆臣；

【作者基本信息】 泰山医学院 ， 影像医学与核医学， 2009， 硕士

【摘要】 目的利用ITK对颅脑MR图像进行半自动及自动分割,并对分割所得的脑组织结构(白质、灰质、脑室)图像进行观察、研究,分析优缺点,为颅脑MR图像脑组织结构的三维显示及手术导航做准备。材料与方法利用ITK编程读取DICOM格式的MR脑图像,半自动分割采用连续阈值区域生长算法、自动分割采用K-Means聚类算法对41幅图像进行分割,得到白质、灰质、脑室等脑组织结构。结果连续阈值区域生长算法及K-Means聚类算法都很好的分割出了各部分脑组织结构。连续阈值区域生长算法综合分割速度慢,受不同层上不同脑组织像素间连通性的影响,将MR脑图像上的各脑组织结构分次单幅显示出来了,较K-Means对不包含脑室的图像进行4分类时的分割精度高、细节多。K-Means聚类算法分割的综合速度比区域生长分割方法分割的快,受初始聚类中心、聚类准则函数、相似度度量方法的影响,可以一次性将白质、灰质及脑室等结构在一幅图像上分割后显示出来,对包含脑室结构的图像进行五分类时的分割效果明显优于连续阈值区域生长的分割效果,而且这样一次性将各脑组织结构在一幅图像上分割出来的快速、高效的分割思想及模式是今后图像分割的方向。结论不同分割方法的优缺点、侧重点不一样,加上医学图像的各异性、复杂性、多模态性,以及图像分割的目的及要求的多样性、具体性、特殊性,使得具体问题需具体研究、分析。连续阈值区域生长算法及K-Means聚类算法的分割各有特点,都很好的分割出了各部分脑组织结构,可以据需要将它们应用于颅脑MR图像脑组织结构的三维显示及手术导航中。更多还原

【Abstract】 ObjectiveTo segment brain tissue from MR Images semi-automaticly and automaticly respectively with ITK. Then to observe and study the segmented brain tissue(whitematter, graymatter, ventricle)images so as to analyze the advantages and disadvantages of those segmented brain tissue images that will be made preparation for the 3D visualization and surgery navigation of MR brain tissue.Material and MethodsTo segment 41 sheets of MR brain image with DICOM format in the ITK and receive corresponding brain tissue(whitematter, graymatter, ventricle) images.Semi-automatic segmentation adopts connected threshold region growing algorithm, while automatic segmentation adopts K-Means clustering algorithm.ResultsBecause of the diversity of connectivity among pixels of different brain tissue in the different slices, connected threshold region growing algorithm segments specific MR Images with 3 sheets of images including leftwhitematter, rightwhitematter and graymatter or 3 sheets of images including whitematter, graymatter and ventricle or 4 sheets of images including leftventricle, rightventricle, whitematter and graymatter. Meanwhile, under the influence of initial means, clustering criterion functions and semblance metric methods, K-Means clustering algorithm segments whitematter, graymatter and ventricle on one sheet of image. Both algorithms obtain high quality of corresponding brain tissue image. On the aspect of overall segmentation speed, region growing algorithm is slower than K-Means clustering algorithm: region growing algorithm receives one sheet of one brain tissue image after one segmentation, while K-Means clustering algorithm receives one sheet of all brain tissue image after one segmentation. Meanwhile, on the aspect of segmentation precision and details, region growing algorithm is better than K-Means 4 clustering while K-Means 5 clustering is better than region growing algorithm. The fast effective segmentation mode of K-Means clustering algorithm which receives one sheet of all brain tissue image after one segmentation is the orientation of image segmentation in the future.ConclusionBecause of the different strong points, advantages and disadvantages of different segmentation algorithms, otherness, complexity and multimodality of medical images and diversity, materiality and particularity of the aims and requests of medical image segmentation, concrete issue should be analyzed concretely. Region growing algorithm and K-Means clustering algorithm segment MR brain images and receive different high quality brain tissue images that can be applied to the 3D visualization and surgery navigation of MR brain tissue.更多还原