CASIA OpenIR  > 毕业生  > 博士学位论文
图割算法的改进及其在三维医学图像分割中的应用研究
Alternative TitleResearch on Improved Graph Cut Algorithms for 3D Medical Image Segmentation
李秀丽
Subtype工学博士
Thesis Advisor田捷
2014-05-25
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
Keyword三维医学图像分割 图割 隐式形状配准 多表面图搜索 骨架割 3d Medical Image Segmentation Graph Cut Implicit Shape Registration Multiple Surfaces Graph Searching Skeleton Cuts
Abstract人体组织器官的三维图像分割是医学图像处理与分析的重要研究内容,三维医学图像分割的结果好坏在很大程度上影响计算机辅助诊断与治疗的效果。三维医学图像分割通过将计算机算法与临床解剖学和病理学先验知识相结合,对目标器官及病灶区域进行分割提取和定量分析,从而获得感兴趣区域的三维空间结构或三维功能特征分布,为疾病诊断、治疗方案规划和手术导航等临床应用提供辅助信息。三维医学图像分割存在两大主要难点:三维体数据计算量大和医学图像目标器官及病灶结构复杂,这对三维医学图像分割的快速性和准确性提出了挑战。基于图割理论的医学图像分割算法因其具有全局优化和同时结合边缘和区域信息的特点,近年来引起了广泛的关注。但是传统图割算法以体素为节点进行图构建,对于三维医学图像中的大量体素,图割算法效率很低且需要的内存非常大,因此图割算法在三维医学图像分割上的应用受到了一定的限制。 本研究针对三维医学图像分割的两大难点和图割算法的优缺点,深入研究并提出了图割算法的三种改进形式:图表现形式改进、图节点模型改进和局部图割算法,结合肾脏及肾皮质分割和肺实质及肺结节分割典型应用,实现了相应器官及病灶区域分割算法,并基于集成化医学影像处理与分析平台MITK&3DMed实现相关算法和设计软件。本文的工作与贡献概括为如下: 1) 提出一种基于隐式形状配准的肾脏表面分割算法 该算法创新性地提出了一种基于隐式形状配准的分割框架,不仅方便地实现了表面数据到体数据的配准,还能对表面数据的形状信息和体数据的位置信息进行整合。算法先利用训练数据生成肾脏平均形状模型,然后通过高斯中值滤波进行肾脏区域检测,最后对肾脏平均形状和肾脏区域进行隐式形状配准,完成肾脏表面分割。该算法在美国国立卫生研究院(NIH)临床中心的数据集上进行测试,并利用留一法进行性能评估,实验结果表明该算法有效整合了肾脏平均形状的形状信息和肾脏区域的位置信息,对临床数据能实现自动快速的肾脏表面分割。 2) 提出一种基于多表面图搜索的肾皮质分割算法 该算法创新性地提出对肾皮质外表面和内表面分别进行图构建和代价函数设计,并有效集成两层表面的不同属性和表面间的相互关系,引入网格稀疏度相关的采样间距和物理分离约束,由于以上对肾皮质外表面和内标的特异性处理,算法确保了两者的有效分割,从而提高了肾皮质分割的准确性。该算法在美国国立卫生研究院(NIH)临床中心的数据集上进行测试,并利用留一法进行性能评估,实验结果表明该算法有效地解决了因为肾脏特殊解剖结构而造成的肾皮质分割困难,对临床数据能实现高效和准确的肾皮质分割。 3) 提出一种基于骨架割的肺结节分割算法 该算法首先利用基于局部图割的算法对肺实质进行分割获得肺部灰度模板,在此灰度模板上采用骨架变换进行体素聚类,基于聚类后的超体素进行骨架割的图构建,并利用最大流最小割算法分割肺结节,最后我们利用骨架半径进行肺结节优化,获得最终肺结节分割结果。我们在美国佛罗里达州坦帕市的莫菲特癌症研究中心的临床数据集上对算法进行测试...
Other Abstract3D image segmentation of the human body tissues and organs is one of the most important research contents in medical image processing and analysis. The result of the 3D medical image segmentation will influence the effect of the computer assisted diagnose and treatment to a great extent. 3D medical image segmentation combines the computer algorithms and clinical anatomical and pathological priori knowledge, and carries out a segmentation and quantitative analysis for the target organs and lesion areas. Then we can obtain the 3D space structure and 3D functional feature distribution of the region of interest (ROI), and provide supplementary information for disease diagnosis, treatment planning and surgical navigation. There exist two main challenges for 3D medical image segmentation: the large computation cost for 3D volume data and the structure complexity of target organs and lesion areas in medical images. For the global optimization property and both of edge and region information representation, segmentation methods based on graph cuts have arisen widespread attention these years. The traditional graph cuts treat each voxel as a node during the graph construction. But when it comes to the large amount of voxels in 3D medical images, the efficiency becomes very low and the memory requirement becomes very big. These limit the application of graph cuts in 3D medical image segmentation. In this dissertation, aiming at the two challenges of 3D medical image segmentation and the pros and cons of graph cuts, we proposed three improvements for the graph cuts algorithm, which are graph representation improvement, graph node model improvement and local graph cuts. We realize corresponding segmentation algorithms for some typical applications, which include kidney and renal cortex segmentation, and lung parenchyma and pulmonary nodule segmentation. We also implement the algorithms based on the integrated medical image processing and analysis platform (MITK&3DMed). The main contributions of this thesis are listed as follows: 1) We proposed a kidney surface segmentation method based on implicit shape registration algorithm. The approach proposes a novel segmentation framework based on implicit shape registration. It not only realizes the registration from surface to volume, but also integrates the shape information and location information. The algorithm firstly generates the average kidney shape using training data. Then, it detects the rough kidney region ...
shelfnumXWLW2036
Other Identifier201118014629084
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6597
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
李秀丽. 图割算法的改进及其在三维医学图像分割中的应用研究[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20111801462908(6894KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李秀丽]'s Articles
Baidu academic
Similar articles in Baidu academic
[李秀丽]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李秀丽]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.