【作者】 宋志强；

【导师】 王大玲；

【作者基本信息】 东北大学 ， 计算机软件与理论， 2008， 硕士

【摘要】 随着环境污染的加重,全球肺部疾病的发病率都呈明显的上升趋势,越来越多的人开始关注自己的肺部健康。CT作为检查肺部疾病的一种重要手段,在临床实践中广泛被采用。对于肺部疾病的诊断,对肺结节的识别尤其重要。本文对基于肺部CT图像的肺结节检测技术进行了研究,首先对肺实质进行了分割,然后在提取的肺实质内对可疑区域进行标记并对其特征提取,最后对可疑区域分类,最终实现了肺结节的检测。应用Level Set方法在图像分割中的优势,本文提出了一种将形态学与水平集相结合的肺实质分割方法。该方法的最大特点在于利用形态学的方法得到肺实质的轮廓窄带和初始轮廓,使水平集只在轮廓窄带内演化,实现了肺实质的分割。实验表明该方法明显的提高了分割性能。对于特征提取问题,由于特征提取的质量是决定分类性能的关键因素,所以选择适当的特征至关重要。描述图像特征的方法有很多种,如颜色特征、纹理特征、形状特征等。本文进一步研究了临床诊断中对肺结节诊断的依据,通过对肺结节和图像的分析,提取了29个不同种类的特征。对于肺结节分类问题,本文将支持向量机的机器学习方法引入其中。分析了支持向量机的理论基础和数学模型,特别是支持向量机的推广能力和核函数理论,在此基础上应用支持向量机方法、采用肺结节提取出来的特征对样本进行分类。最后的实验结果表明本文采用的算法能够应用于肺结节计算机辅助诊断中。更多还原

【Abstract】 Because of circumstance pollution, the incidence of pulmonary disease is increasing obviously. More and more persons pay attention to their lung health. As an important means of detecting lung disease, CT has been applied in clinic. For the diagnosis of lung disease, identifying pulmonary nodules is very important.In this thesis, detection of pulmonary nodules based on lung CT images is studied. Firstly, the lung segmentation is segmented, in which some suspicious regions are marked. Then the features of the regions are extracted which are used for classification. Finally, the pulmonary nodules will be detected.In this thesis, with the advantages of Level Set, a new method based on morphology and Level Set is proposed for lung segmentation. The most important character of this method is that methods of morphology is used to gain narrowband of contour and the initialization contour of lung, making Level Set function evolution only in the narrowband. The results of experimentation show that with this method performance of segmentation is improved obviously.For feature extraction, since the quality of feature extraction is a crucial element of the performance of classification, choosing a proper way to extract features of image is extremely important. There are a lot of ways to describe image features, such as color, texture, and shape. Symptoms of pulmonary nodule in clinical diagnosis are further researched, and 29 different features of pulmonary nodules are determined by use of outline analysis and image analysis.For classification of pulmonary nodules, supported vector machine (SVM) is imported, which is a machine learning method. Firstly, the theoretical principle and mathematical model of SVM, especially the popularizing ability and kernel function theory are analyzed. Then the feature extraction results are taken as input and SVM is applied to classify pulmonary nodules.The experimental results show that the method in this thesis can be regarded as a technique for CAD systems in CT pulmonary nodules.更多还原