【作者】 李冰；

【导师】 张石； 佘黎煌；

【作者基本信息】 东北大学 ， 信号与信息处理， 2015， 硕士

【摘要】 随着CT等新医学成像方法的广泛应用,医学图像处理与分析已经成为医学技术中发展最快、成果最显著的领域之一。如何对获得的影像学数据进行分析成为至关重要的问题,因此分割技术成为后续图像处理和分析中的关键技术。本文对目前国内外的医学图像分割方法进行了综述,将研究目标定位在基于水平集方法的肝脏CT图像分割算法研究上。针对肝脏CT图像边缘模糊,器官间对比度差、肝实质灰度不均匀等问题,提出了自己的肝脏CT图像分割方法。本文的主要工作如下:(1)针对肝脏CT成像中产生的噪声和肝实质灰度不均的问题,采用Mean Shift方法对肝脏CT图像进行平滑,由于传统的肝脏图像平滑方法在平滑的同时对严重丢失了肝脏区域的信息,且对肝实质的灰度不均问题没有很好的处理,增加了之后分割难度。Mean Shift方法通过沿灰度梯度的方向进行平滑,保留了肝脏图像的边缘信息,在平滑的同时保证了肝脏区域图像的信息,通过实验表明该方法对肝脏图像具有较好的平滑效果。(2)针对肝脏CT图像中存在的器官间对比度差、边缘模糊的问题,提出了基于GVFGAC模型的交互式肝脏分割方法。采用区域生长进行粗分割,排除肝脏图像中其他器官的影响,然后通过形态学变换和边缘提取,修复区域生长过程中产生的空洞和过分割的问题,获得梯度图后用GVFGAC模型进行细分割,GVFGAC模型通过引入GVF的矢量场,增加了进入凹陷的能力,能够更细致的完成对肝脏的分割。实验表明该方法优于现存的一些交互式肝脏图像分割方法。(3)为了排除人为参与肝脏CT图像分割带来的影响,提出了基于DRLSE模型的全自动肝脏CT图像分割方法,通过K-means的聚类分类,然后对图像进一步通过形态学变换和寻找最大连通区域完成粗分割,而后通过C-V模型和DRLSE模型相结合的水平集方法完成对肝脏CT图像的分割,由于全自动的分割方法没有人为引导,算法复杂,运算量大,采用不需要初始化的DRLSE模型可以大幅降低水平集的运算时间,结合基于C-V模型中的能量项,能够快速准确的完成对肝脏区域的细分割。实验表明该方法优于现存的一些全自动肝脏图像分割方法。更多还原

【Abstract】 With the wide application of new medical imaging methods such as CT,medical image processing and analysis has become one of the most rapidly developing and most notable achievements in medical technology.How to analyze the acquired image data is the key problem,so the segmentation technology becomes the key technology in the following image processing and analysis.In this thesis,the medical image segmentation method is reviewed,and the research object is based on the segmentation algorithm of the liver CT image based on the level set method.Aiming at the problems of the edge blur of CT image,the contrast between the organs and the gray level of liver parenchyma,the segmentation method of CT images of the liver is presented.The main work of this thesis is as follows:(1)In liver CT imaging noise and liver parenchyma gray uneven problems.For the first time,the mean shift method for smoothing the liver CT images,due to the traditional liver image smoothing method in smooth and drastically reduces the image quality and treatment of hepatic parenchyma of gray uneven problem is not very good,the increased after segmentation difficulty.Smooth along the gradient direction by mean shift method,preserves the edge information of images of the liver,while smoothing the image quality changes little,through experiments show that the method of liver image has good smoothing effect.(2)The problem of the contrast between the contrast and the edge of the organs in the liver CT images is presented,and the method of the interactive liver segmentation based on the GVFGAC model is proposed.Using region growing coarse segmentation,excluding liver image in other organs,and then through morphological transform and edge detection,repair area growth empty and over segmentation problem generated in the process,to obtain the gradient map with GVFGAC model fine segmentation and model GVFGAC through into the GVF model,the vector field,increase the ability to enter into depression,can complete the segmentation of liver in more detail.Experiments show that this method is better than some existing methods for liver image segmentation.(3)In order to eliminate the effect of human liver CT image segmentation,segmentation method is proposed for automatic liver CT images based on DRLSE model,through the clustering and classification of K-means,and then the image further by morphological transformation and find the biggest connected area to complete the coarse segmentation,and then through the C-V model and DRLSE model combined with level set method the completion of the liver CT image segmentation,the segmentation method automatically without human guidance,complex algorithm,the large amount of computation,the DRLSE model does not need to initialize the level set can greatly reduce the computation time,based on the energy in the C-V model,can quickly and accurately complete the segmentation of liver region.Experiments show that the proposed method is better than some existing automatic liver image segmentation method.更多还原