基于CT体数据的人体肺气道树数学模型建立

发布时间：2018-03-17 11:30

本文选题：肺气道　切入点：计算机辅助诊断　出处：《中国科学技术大学》2010年博士论文　论文类型：学位论文

【摘要】： 肺部疾病是人类健康的重要威胁,建立肺部疾病计算机辅助诊断(CAD)系统已成为当前的研究热点。目前,有两种不同的实现肺部疾病CAD系统的方案：传统的异常依赖方案(ADA)和新兴的正常依赖方案(NDA)。异常依赖方案关注于图像目标中符合某种特定病变特征的区域,因此不能很好地应对临床诊断中多种病变同时存在的情况。正常依赖方案则更加符合专业医师的读片方法和诊断思路,那就是先识别并排除图像中的正常区域,然后对剩下的可能异常的区域进行精细分析。这样,多种病变的信息可以同时被保留下来,而不仅仅是某一种特定病变。鉴于这个原因,正常依赖方案已经成为未来肺部疾病CAD系统研究的发展趋势。本文遵循正常依赖方案的基本思想和原则,提出了一条新颖的构建人体肺部疾病CAD系统的技术路线,其基础性工作是在计算机里建立一个反映正常人群影像学特征的肺部数字化参数集合。我们将注意力主要放在肺气道参数上,因为肺气道疾病是最严重的肺部疾病之一,而且我们的工作将显示出肺气道参数能够从一定程度上反映整个肺的情况。本文最终建立了一个包含丰富肺气道参数的数学模型,并通过五个步骤来达成这一目标。在每个步骤中,都进行了深入的研究工作,取得了一定的成果。第1步：肺分割。肺分割是后续肺气道分割的基础。针对医学图像背景复杂、边界模糊、局部不均匀等特点,提出使用相对模糊连接度作为几何主动轮廓模型中曲线演化的驱动力,并从理论分析和实验验证两个角度证明了其对于肺部图像的适用性。本文方法在多目标图像和复杂图像的分割实验中取得了良好的效果,最终分割出了完整、正确的肺组织。第2步：气道分割。气道分割是整个模型建立的基础,并直接决定着模型的性能。以普通分割方法的结果作为本文分割的基础,采用气道分割的一般性框架,对其中的分割、评价等模块进行了改进。对于泄漏和细小气道提取这两大难题,提出了有针对性的策略。最终提取了10个层级中的104个气道段(相当于手动分割数量的45%),并完整保留了段支气道以上的全部气道信息。第3步：骨架提取。提取肺气道树的单像素宽中心骨架是进行参数测算的必须环节,几乎所有结构性参数的定义和测算方法都是基于气道骨架而提出的。针对肺气道树骨架化的特定需求,综合分析了常用的四类骨架提取算法,并最终选择了层次性一般势能场方法。整个骨架化过程被分为提取核心骨架(第一级)、加入细小分支(第二级)、连接末梢端点(第三级)这三个完整性逐渐增加的层级,其中第三级的结果即是最终结果。这种方法的骨架化效果在完整性上要优于其他方法。第4步：参数测算。这是与建立数学模型直接相关的步骤,所测算出的参数就是数学模型中的数据部分。整个肺气道树被解剖为整体、层级、段和层片四种结构,针对每种结构提取了一类参数。整体类参数有4种,层级类参数有4种(10层级),段类参数有5种(104段),层片类参数有5种(1916层片),因此总共提取了10144个肺气道参数。部分参数与影像学事实和真实解剖值进行了对比,结果显示这些测算结果是正确、可信的。第5步：模型建立。这是本文的最终目标。建立出的数学模型由两个部分组成：数据和操作。所有肺气道参数数据存储在具有五级结构的树中。树的每一级由一种类型的结构体组成,每种结构体分别存储整体参数、层级参数、段参数、层片参数和体素参数。对数据的操作是模型的第二部分,是对参数进行计算、存储、装载、查询、调用等的计算机算法。结构体中存有编码和指针,用于参数的查询和调用。在模型数据和模型操作的基础上,给出了三个实际应用的解决方案,作为后续建立肺部疾病CAD系统的参考。总之,本文所构建出的人体肺气道树数学模型中含有丰富而正确的参数,可以用于下一步建立肺部疾病CAD系统。本文作者特别感谢国家自然科学基金(60771007)和中科院研究生科技创新基金(2008年度)对本文的资助。
[Abstract]:Lung disease is an important threat to human health, the establishment of lung disease computer aided diagnosis (CAD) system has become the focus of current research. At present, a realization of lung disease CAD system for two kinds of abnormal dependence of traditional scheme (ADA) and the new normal dependent scheme (NDA) ADA focused. On the image with a specific lesion area, so it can not deal with at the same time, the presence of multiple lesions in clinical diagnosis. The NDA is more in line with the medical professional film reading method and diagnostic thinking, that is to identify and exclude the normal image area, then the remaining possible fine analysis the abnormal area. In this way, a variety of disease information can also be retained, rather than a particular disease. For this reason, the normal scheme has become dependent on the future The development trend of CAD system in lung disease.
This paper follows the basic ideas and principles of NDA, proposed a novel construction of human lung disease CAD system technical route, the basic work is to establish a set of lungs digital parameters reflect the normal people in the computer image features. We will focus mainly on parameters of lung airway, because of pulmonary airway the disease is the most serious lung diseases, and our work will show that airway parameters can reflect the entire lung from a certain extent. This paper established a mathematical model with rich airway parameters, and through the five steps to achieve this goal. In each step, are carried out deep research work, and achieved certain results.
The first step: lung segmentation. Lung segmentation is the basis for subsequent airway segmentation. For medical image with complex background, fuzzy boundaries, characteristics of local inhomogeneity, put forward relative fuzzy connectedness as a geometric active contour model driven curve evolution, and from the theoretical analysis and experimental verification of two proved that the the applicability of the lung images. This method has a good effect in multi object image segmentation experiments and complex image segmentation, finally the integrity, the right lung tissue.
The second step: airway segmentation. Airway segmentation is the basis of the model establishment, and directly determines the performance of the model. The common segmentation method results as the basis for this division, the general framework of the airway segmentation, segmentation, evaluation module is improved. For small airway leakage and extraction of the two the problem, proposed a targeted strategy. Finally from the 104 section of the 10 levels in the airway (equivalent to 45% of the number of manual segmentation), and retain the integrity of the airway section supporting the whole airway information above.
The third step: skeleton extraction. Single pixel wide skeleton extraction center airway tree is the essential part for parameters measurement, definition and calculation method of almost all structural parameters are proposed. Based on the skeleton of airway for the needs of specific airway tree skeleton, a comprehensive analysis of the four kinds of commonly used skeleton extraction algorithm, and chose the level of general potential field method. The whole skeleton extraction process is divided into core skeleton (first level), with small branches (second), connecting peripheral points (third) of the three integrity level gradually increased, the third level is the result of the final result. The effect of this skeleton the method is better than other methods in integrity.
The fourth step: parameter estimates. This is directly related to the establishment of mathematical model of the steps, the data portion of the measured parameters is the mathematical model. The airway tree was dissected as a whole, level, structure of four kinds of segments and layers for each extraction structure for a class of parameters. The whole class has 4 parameters a class hierarchy, there are 4 parameters (10 levels), 5 kinds of segment parameters (104), 5 kinds of slice parameters (1916 slices), so the total extraction of 10144 pulmonary airway parameters. Some parameters and imaging anatomic facts and real values were compared, the results showed these calculation results are correct and reliable.
The fifth step: to establish the model. This is the ultimate goal of this paper. The established model consists of two parts: data and operation. All the parameters of lung airway data stored in a structure with five levels in the tree. The tree of each level by a type of structure body, each storage structure respectively. The overall level of parameters, parameters, parameters, slice parameters and voxel parameters. The operation of the data is the second part of the model, the calculation of parameters, loading, storage, query, computer algorithm calls etc. the structure being encoding and pointers, query and call for parameters based on the model. The data and operation of the model, gives three solutions for practical applications, as the subsequent establishment of lung disease CAD system.
In conclusion, the mathematical model of the human lung airway tree constructed in this paper contains abundant and correct parameters, which can be used to establish the CAD system for the next step of the lung disease.
The author of this article is particularly grateful to the National Natural Science Foundation (60771007) and the graduate science and Technology Innovation Fund (2008) of the Chinese Academy of Sciences for the funding of this article.

【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2010
【分类号】：R311;TP391.41

【引证文献】