人体脑图像分割技术研究

【作者】 薛耿剑；

【导师】 郝重阳；

【作者基本信息】 西北工业大学 ， 信号与信息处理， 2006， 硕士

【摘要】 医学图像分割是医学图像处理和分析领域的基础性经典难题,其中脑部医学图像分割因其重要的应用价值一直是医学图像分割的研究热点。本论文的主要工作是对可视人脑部图像和MRI脑部图像进行分割技术研究,以便提取出特定的组织和结构。论文的创新点包括: 1.提出了一种基于最大熵和数学形态学的可视人脑部彩色图像分割方法。该方法是将原本只用于灰度图像分割的最大熵方法和数学形态学方法有机结合而形成的一种彩色脑图像分割方法,它不仅能对可视人脑部图像进行有效分割, 而且成功提取出了脑组织和脑骨结构。经对多达576幅可视人脑部图像作实验均获得若用单一算法无法实现的满意效果。 2.为适应MRI图象分割的问题提出了一种新的核聚类算法。该方法针对传统模糊核聚类算法当数据类差别很大时,小数据类被误分或被大数据类吞并的缺陷,通过定义一个新的目标函数,为每一个类分配了一个动态权值,可改善聚类效果。实验结果表明,本文算法对脑白质、脑灰质、脑脊液分割结果在降低误分率性能方面均比经典核聚类算法有5%以上的提高。 3.提出了一种分层Mumford-Shah模型的颅脑图像分割方法。该方法针对Mumford-Shah模型分割多目标时初始化的耦合对分割结果有明显影响的问题,提出应用分层Mumford-Shah模型进行图像分割,并运用了改进的图像分割方程进行计算求解的图像分割方法,经对MRI脑图像脑白质,脑灰质和脑脊液的分割实验,取得了良好的效果。更多还原

【Abstract】 Medical image segmentation is a hard-tough problem in medical image processing and analysis. Among it, brain medical image segmentation is the research focus for its important values. The primary task of my thesis is studying the Visible Human brain image segmentation and MRI brain image segmentation techniques in order to extract the tissues and structures. The achievements of my thesis include: 1. A Visible Human brain color image segmentation method based on maximum entropy has been proposed. Combing the maximum entropy method and Mathematical Morphology method that were only used for gray image originally, our method extracts brain tissues and brain bones that won’t be done if single method is used. The examples of 576 brain images show its effectiveness. 2.A new MRI kernel fuzzy clustering algorithm has been proposed for MRI image segmentation. When the difference of the date size is large. The small class may be misclassified or may be merged into big class in fuzzy kernel clustering algorithm. A new objective function is introduced and an additional weighting factor is assigned to each class. Based on this improvement the algorithm gets better performance. The results indicate it could reduce 5% misclassification rate of the white matter、 gray matter and CSF than original algorithm.3.When Mumford-Shah model is used for multi-object segmentation, the initialization influences the results seriously. A novel image segmentation approach based on hierarchical Mumford-Shah model has been proposed in this paper, which gets the solution by solving the improved image segmentation equations. The example of brain MRI image segmentation indicates its effectiveness.更多还原