基于胸部CT图像肺结节分割算法研究与实现

发布时间：2018-02-09 10:09

本文关键词： 肺结节分割测地距离光线投射凸包操作竞争扩散　出处：《东北大学》2012年硕士论文　论文类型：学位论文

【摘要】：近年来在世界范围内肺癌的发病率和死亡率持续攀升,发病率和死亡率已均居各类癌症首位。针对肺癌还无法根治的现状,降低其死亡率的主要措施是早发现、早诊断、早治疗。肺癌在早期阶段表现为肺结节,肺结节通常比较小,很容易漏检。CT成像技术是目前肺结节诊断和鉴别的首选方法,因其具有良好的密度分辨率、三维成像特征和很高的敏感度,可以很好显示结节的位置、形态、大小、内部结构、密度、边缘特征及其周围的改变。目前,对肺结节良恶性的判定考虑的主要指标有体积、倍增率、形态等。肺结节的精确分割是进行结节定量分析和良恶性鉴定的前提和基础,因此,基于胸部CT图像的结节分割已成为肺计算机辅助诊断技术中重要内容之一。由于结节自身的复杂性和CT成像技术的特点,不同的结节在灰度值、形态、与周围组织粘连程度等方面表现出很大差异,使得结节的分割成为一项极具挑战性的工作。针对结节的分割,本文提出了三种不同的分割方法,前两种能够对特定种类的结节进行分割,第三种能够对不同种类的结节进行分割,包括：(1)基于局部形状分析的血管粘连型结节分割方法,首先利用距离变换找到结节的内核,然后利用测地距离作为步长进行区域生长,探测血管和结节发生粘连的位置,剔除血管；(2)基于三维光线投射的肺壁粘连型结节分割方法,首先利用光线投射探测靠近肺实质一侧结节的轮廓,然后利用凸包操作对探测到的轮廓进行重建；(3)基于最速下降思想的肺结节分割方法,首先利用竞争扩散系统对图像进行二值化,距离变换找到结节内核,然后从结节的内核开始进行最速下降,找到结节的边界,实现不同种类结节的分割。通过实验验证,这三种算法对结节的分割均取得较高的准确率和效率,和金标准的平均重叠率依次达到85.16%、93.03%和85.35%,平均分割一个结节的时间为2s、12.7s和1.5s。表明了这些算法的分割结果可以用于结节的定量分析,并为结节的诊断和治疗提供辅助信息。
[Abstract]:In recent years, the incidence and death rate of lung cancer has been rising continuously in the world, and the morbidity and mortality have been ranked first in all kinds of cancers. In view of the present situation that lung cancer cannot be cured, the main measures to reduce the mortality rate are early detection and early diagnosis. Early treatment. Lung cancer in the early stage of pulmonary nodules, pulmonary nodules are usually relatively small, it is easy to miss the CT imaging technology is the first choice for the diagnosis and differentiation of pulmonary nodules, because of its good density resolution, Three-dimensional imaging features and high sensitivity can well display the location, shape, size, internal structure, density, marginal features and the changes around the nodules. At present, the main criteria for judging benign and malignant pulmonary nodules are volume. The accurate segmentation of pulmonary nodules is the premise and foundation of quantitative analysis and identification of benign and malignant nodules. Therefore, the segmentation of nodules based on chest CT images has become one of the important contents of computer-aided diagnosis of lung. Due to the complexity of the nodules themselves and the characteristics of CT imaging techniques, different nodules show great differences in terms of gray value, morphology, adhesion to the surrounding tissues, and so on. In this paper, we propose three different segmentation methods, the first two can segment a specific type of nodules. The third one is able to segment different types of nodules, including: 1) a method for segmenting vascular adherent nodules based on local shape analysis. Firstly, distance transformation is used to find the kernel of the nodules, and then geodesic distance is used as the step size for regional growth. To detect the location of adhesion between blood vessels and nodules, and to remove the blood vessels. (2) the segmentation method of pulmonary wall adhesion nodules based on 3D ray casting. First, the contours of nodules near one side of the lung parenchyma were detected by ray casting. Then the detected contour is reconstructed by convex hull operation. Based on the idea of steepest descent, the lung nodule is segmented. Firstly, the image is binarized by competitive diffusion system, and the kernel of the nodule is found by distance transformation. Then starting from the kernel of the nodule, the fastest descent is carried out, the boundary of the nodule is found, and the segmentation of different kinds of nodule is realized. The experimental results show that these three algorithms achieve high accuracy and efficiency in the segmentation of the nodule. The average overlap rate of the standard and the gold standard was 85.16% and 85.35% respectively, and the average time of dividing a single node was 2 s-1. 7 s and 1. 5 s. The results showed that the segmentation results of these algorithms could be used for quantitative analysis of nodules and provide auxiliary information for the diagnosis and treatment of nodules.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R734.2;R730.44

【参考文献】