基于CT图像的肺结节分割方法研究

发布时间：2018-08-30 12:21

【摘要】：研究表明,肺癌是人类目前死亡率最高的疾病之一,且肺癌的发病率和死亡率呈现不断增长的趋势,如何预防及治疗肺癌已经成为人们必须解决的课题。通过研究肺癌病例发现,及早发现肺癌病状可以延长患者的生存时间,相反,晚期肺癌患者死亡率非常高。肺结节作为肺癌早期主要表现形式,因此,肺结节分割成为发现肺癌的重要手段。近年来,活动轮廓模型(Active Contour Model,ACM)凭借其灵活的结构、优越的性能以及多样的形式,在医学图像分割领域受到了越来越广泛的研究和应用。本文对目前已有的基于活动轮廓模型算法在医学图像分割中的应用进行了深入的学习和研究。提出改进的活动轮廓模型算法,并将改进算法在分割磨玻璃型(Ground Glass Opacity,GGO)肺结节和血管粘连型(Juxta-vascular,JV)肺结节中进行应用。首先,针对局部区域活动轮廓模型(Local Region-based Active Contour Model,LRACM)模型分割GGO肺结节存在的缺陷:对初始轮廓敏感、分割精度低和依赖梯度信息等。本文在LRACM模型基础上提出了改进的基于局部和全局隶属的活动轮廓模型(Local and global membership based on the active contour model,LGMACM)分割方法:1)采用全局隶属度构造边界停止函数。通过全局隶属度来构造边界停止函数,解决了边界泄露的问题;2)采用局部隶属度构造轮廓模型的数据项,加强了GGO肺结节与背景的对比度;3)采用全局隶属度来获取初始轮廓曲线,提高了模型的效率和收敛速度。其次,针对基于边界的轮廓模型(Edge-based Active Contour Model,EACM)和基于区域活动轮廓模型(Region-based Active Contour Model,RACM)分割 JV 型肺结节存在的问题:边界泄露、对噪声敏感、收敛速度慢和没充分考虑图像先验知识等。本文在EACM和RACM模型基础上提出了改进的基于模糊速度的活动轮廓模型(Fuzzy velocity function based on active contour model,FVFACM)的分割方法:1)构造亮度和形状特征的二维向量并结合模糊聚类算法计算模糊隶属度,并根据隶属度构造速度函数。2)将模糊速度函数引入到活动轮廓模型中,将模糊速度作为内部能量和外部能量的权重系数。基于以上两点的改变,该模型有效的处理了边界泄露、对噪声敏感和收敛速度慢的问题,同时也提高了模型的分割精度。实验结果表明,本文提出的两种改进模型,通过实验仿真结果对比和数据统计对比,能够达到预期的效果。
[Abstract]:The research shows that lung cancer is one of the highest mortality diseases, and the morbidity and mortality of lung cancer are increasing. How to prevent and treat lung cancer has become a problem that must be solved. Through the study of lung cancer cases, early detection of lung cancer symptoms can prolong the survival time of patients, on the contrary, the mortality rate of advanced lung cancer patients is very high. Pulmonary nodules as the main early manifestations of lung cancer, lung nodules segmentation has become an important means of finding lung cancer. In recent years, the active contour model (Active Contour Model,ACM) has been more and more widely studied and applied in the field of medical image segmentation because of its flexible structure, superior performance and various forms. In this paper, the application of active contour model algorithm in medical image segmentation is studied. An improved active contour model (ACM) algorithm is proposed and applied to segmented glass-grinding (Ground Glass Opacity,GGO) pulmonary nodules and vascular adhesion type (Juxta-vascular,JV) pulmonary nodules. Firstly, the defects of local active contour model (Local Region-based Active Contour Model,LRACM) model for segmenting GGO pulmonary nodules include sensitivity to initial contour, low segmentation accuracy and dependent gradient information. In this paper, an improved (Local and global membership based on the active contour model,LGMACM segmentation method based on local and global membership is proposed based on LRACM model. The global membership degree is used to construct the boundary stop function, and the problem of boundary leakage is solved. 2) the data item of the contour model is constructed by using the local membership degree. The contrast between GGO pulmonary nodules and background is enhanced. The global membership degree is used to obtain the initial contour curve, which improves the efficiency and convergence speed of the model. Secondly, the problems of boundary based contour model (Edge-based Active Contour Model,EACM) and region based active contour model (Region-based Active Contour Model,RACM) for segmenting JV pulmonary nodules include boundary leakage, sensitivity to noise, slow convergence rate and inadequate prior knowledge of image. On the basis of EACM and RACM models, an improved segmentation method of active contour model (Fuzzy velocity function based on active contour model,FVFACM) based on fuzzy velocity is proposed in this paper. The segmentation method: 1) constructs two-dimensional vectors of luminance and shape features, and calculates fuzzy membership degree with fuzzy clustering algorithm. The fuzzy velocity function is introduced into the active contour model and the fuzzy velocity is used as the weight coefficient of internal and external energy. Based on the above two changes, the model effectively deals with the problems of boundary leakage, sensitivity to noise and slow convergence rate, and also improves the segmentation accuracy of the model. The experimental results show that the two improved models proposed in this paper can achieve the desired results through the comparison of the experimental simulation results and the statistical data.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R734.2;TP391.41

【相似文献】