【作者】 陈日晖；

【导师】 魏颖；

【作者基本信息】 东北大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 肺癌是当今死亡率最高的癌症疾病之一,严重地威胁着人类的生命健康。为了提高肺癌患者的生存率,肺癌的早期诊断与治疗是关键。CT扫描是肺癌诊断的重要手段。肺癌在CT图像中是以肺结节的形式表现的,而肺部本身结构的复杂性和肺结节形状大小的多样性,使得富有经验的医生也很难及时发现图像中所有可能病变区域。同时随着多层螺旋CT的出现,医生需要处理的图像信息急剧增加,因此迫切需要一种有效的技术来减轻医生的工作,计算机辅助诊断方法就这样慢慢发展起来。它可以辅助医生对病变及其它感兴趣的区域进行定性乃至准确的定量分析,从而提高医疗诊断的效率,减轻医生的负担。本文对面向肺部CAD的三维RO1分割、特征提取与分类方法进行研究。主要研究工作包括三部分：首先研究了肺部感兴趣区域(ROI)的三维分割算法。包括自适应三维区域生长算法和基于三维Hessian矩阵的算法,并对两种算法分割后的结果进行分析和比较。其次研究了感兴趣区域的三维特征提取算法。从全局考虑,提取了表面积、体积、球形度、凸形度、紧凑度和灰度均值6个特征；从局部考虑,提取了肺结节的局部体积形状指数特征。并根据肺结节的CT影像特征,对所提取的特征进行分析评价。最后根据感兴趣区域提取的有效特征,应用Fisher分类器和基于核的Fisher分类器对结节区域和正常组织进行分类。实验表明,改进的基于核的Fisher分类器比Fisher分类器具有更好的分类性能和准确性。整体而言,本文充分考虑了CT图像的上下层信息,运用三维的方法对感兴趣区域进行分割、特征提取及分类算法的研究,取得了较好的效果,可以为肺部结节的判别提供辅助诊断依据。更多还原

【Abstract】 Lung cancer is one of the most deadly diseases in the world, it threatens people’s lives. To improve lung cancer patients’viability, the earliest diagnosis and therapy of lung cancer is essential. CT scanning has been the most important method of lung cancer examination. The lung cancer is represented with the pulmonary nodules in CT images, and since the complexity of the lung structure and the multiplicity of nodule’s shape, even the seasoned doctor cannot easily discover all of the possible disease in time. However, with the advent of the multislice CT, the image data which were handled by the physician drastically increased. An assistance is cried for in order to lighten physician’s burdens, so computer-aided diagnosis (CAD) is developed gradually. It may help doctors to analyze pathological changes and other regions of interest in character and even in accurate quantity, and release the doctors’burden.In this thesis,3D ROI segmentation, feature extraction and classification methods for pulmonary CAD are researched. The main work includes the following three parts:first,3D segmentation algorithms of region of interest are researched. The whole ROI can be extracted by the proposed adaptive three-dimensional region growth algorithm and the algorithm based on three-dimensional Hessian filter. The results of two algorithms are analysed. Second, a scheme for three-dimensional feature extraction is presented. In the aspect of the spatial context features, six features which consisted of surface, volume, sphericity, convexity, compactness, mean are extracted; in the aspect of the local features, the volumetric shape index features are extracted. These extracted features are identified and evaluated according to medical symptoms. Last, the fisher classifier and the kernel-fisher classifier are designed to distinguish nodules from normal areas after the effective features extraction. The experiment results indicate that the performance of modified kernel-fisher is better than that of the fisher classifier.On the whole, the algorithms put forward in this thesis for ROI segmentation, feature extraction and classification by use of three-dimensional methods have good detected result according to the spatial context messages. And it could be used as supplementary information for the diagnosis of the pulmonary nodules.更多还原