【作者】 樊鹏；

【导师】 霍修坤；

【作者基本信息】 安徽大学 ， 信号与信息处理， 2012， 硕士

【摘要】 肺部结构复杂,内部包含大量血管和气管,发于肺部的疾病也众多,严重影响了人类的健康。随着医疗影像的发展,多种医疗设备可用于检测肺部病情。对胸部CT进行详尽地分析,分割出肺实质、肺血管或具体病灶,意义重大,有助于医生精确诊疗,为病患减轻病痛。本文首先对图像分割知识进行了介绍,包括图像分割的原理、经典方法以及新兴方法。然后介绍了医疗图像分割的目的和重要意义以及国内外研究现状及未来发展趋势。因为分割的对象是DICOM格式的医疗图像,怎样将图像正确清晰地显示出来是一切后续处理的基础和前提。因此,本文详细介绍了DICOM的相关知识,怎样正确阅读DICOM文件及显示方法。本文的研究对象是CT胸部图像,面对这种很专业的医疗图像,必须对CT图像的成像原理、胸部CT的扫描方法有所了解,对CT读片前的重要知识熟悉,为后期对图像进行分割时,正确显示病灶选取恰当的图像做准备。随后对血管造影也进行了简单介绍。本文中,提出了图像分割在胸部CT的两种运用：一是对胸部CT进行处理,分割出肺实质区域；二是对分割出来的肺实质区域进行进一步的分割,分割出肺部血管。两次分割基于不同的分割对象给予不同的分割算法。对于将肺部实质从胸部CT中分割出来,本文首先介绍了经典的基于边界的边界追踪算法,基于区域的区域生长算法和基于阈值的Ostu阈值法,并运用此三种算法对胸部CT进行分割得到肺部实质区域,我们发现,经典的算法虽然优点很多,但分割出来的肺实质均有缺陷,分割效果不佳。最后重点介绍本文算法：将Mean-Shift算法和snakes模型相结合对胸部CT进行分割得到肺部实质区域,实验结果表明,本文算法明显优于经典的三种图像分割方法,分割出的肺实质效果完整,具有临床使用价值。随后,本文对分割出的肺实质进行进一步的研究,希望将肺部血管从肺部实质中分割出来。由于肺部血管对比度差、数量众多、形状复杂、分布无规律等特点,传统的分割方法分割效果均不佳,本文将Gabor小波变换与阈值法相结合分割提取肺部血管,经大量实验验证,Gabor小波变换能很好地增强肺部血管,提取肺部血管的方向特性,然后结合阈值法能够较好地提取出肺部血管,且本文所选阈值具有鲁棒性,无需人机交互,简单易行,方便医生使用,有助于临床诊疗。更多还原

【Abstract】 Lung has a complicated structure which contains a large number of vessels and tracheas. The disease in lung also is numerous which impacts on human health seriously. With the development of medical image, a variety of medical equipment can be used for the detection of lung disease. Give an exhaustive analysis to chest CT and segment out the lungs. Pulmonary vessels and specific lesions are signality for doctors to give accurate diagnosis and relieve patients’ pain.This thesis first has introduced the knowledges of image segmentation which including the principle of image segmentation and the classical and emerging methods of image segmentation. Then it has introduced the important purpose and meaning of medical image, also including the domestic and foreign research present situation and future development trend. Because the division of the object is DICOM format of medical image, it’s the basis and prerequisite of how to show out the images clearly for the follow-up processing. Therefore, this thesis has introduced the related knowledge of DICOM in detail and how to read DICOM image and its display methods rightly. This research object of this thesis is chest CT images. In the face of such a professional medical image, we must have a profound understanding in the imaging theory of CT image and the scanning method of chest CT. We also must know well about the knowledges before reading CT. It’s a preparetion for later image segmentation and the right image selection. Then this thesis has introduced angiogram simply. The thesis puts forward two kinds of use of image segmentation in CT image: One is to process chest CT, Two is to give a further division of lung parenchyma that has segmented out, that is the segmentation of pulmonary vascular. Two segmentations are based on the division of the different objects of different segmentation algorithms.For how to segment out the essence of the lungs from chest CT, First this thesis has introduced the classic boundary tracking algorithm based on the boundary, the region growing algorithm based on the region and the Ostu threshold value method based on the threshold value. Through using the three algorithms for chest CT segmentation to get lung essence area, We find that although classical algorithm has many advantages, it also hasn’t a good effect in segmenting out lung parenchyma. At last this thesis has introduced the algorithm mainly:Using the mean-shift algorithm and snakes model combined with chest CT segmentation to get lung essence area. The experimental results show that the algorithm is obviously better than the classical three image segmentation method. It has a full effect and clinical use value of the lung parenchyma which has segmented out.Then, this thesis gives a further study for the lung parenchyma which has segmented out, with hoping to segment the pulmonary vascular from the lung parenchyma. Because of the bad contrast ratio, fuzzy boundary and complex shape of the lung vascula, it’s not good to use the traditional segmentation. This thesis combines Gabor wavelet transform with threshold method to segment and extract lung blood vessels. The experiments show that, Gabor wavelet transform can strengthen the lungs blood vessels and extract pulmonary vascular’s direction characteristics well. Then it’s better to extract pulmonary vessels combined with threshold value method.And the selected threshold in this thesis has robustness, so that it’s simple and easy to do with no human-computer interaction. It’s also helpful for the clinical diagnosis and treatment and convenient for doctors.更多还原