【作者】 王瑞鑫；

【导师】 郭希娟；

【作者基本信息】 燕山大学 ， 计算机软件及理论， 2010， 硕士

【摘要】 肺部图像的自动分割技术,对于减少临床漏诊及误诊有重要的意义。肺部图像分割提取肺部图像中的肺实质,得到肺部肿瘤特征,给出肺部图像肿瘤的分割结果,帮助医生进行分析和判断。肺部图像中肺实质的提取在肺部图像自动分割系统中有重要的意义。首先,通过对现有的肺部自动检测技术进行分析,针对活动轮廓模型无法解决深凹陷问题,提出了基于改进能量公式的GVF Snake算法。通过引入肺部图像的梯度和曲率等信息,并在它们的共同合力作用下,改进了能量公式,确定肺部图像边界,解决了活动轮廓算法无法对深凹陷进行分割的问题。其次,改进能量公式后,活动轮廓曲线的停止条件将发生变化,针对改进后的GVF Sanke算法中,活动轮廓曲线将出现无法停止,并且越过真实边界,收敛于一点的现象,给出通过梯度信息和曲率信息对外力的引导方向。引入符号函数重新设定曲线运动的停止条件,解决曲线无法收敛于真实边界的问题。活动轮廓曲线的停止条件的重新设定提高了肺部图像分割的速度,降低算法的时间复杂度。然后,针对胸膜处粘连的肺部肿瘤在分割中出现肺部图像边界不连续或是连续却与真实的肺部图像边界相差很远的漏分割问题,提出自适应边界行进算法。该算法通过几何方式平滑肺边界,解决肺部图像中肺部粘连肿瘤的漏分割问题。最后,本文对计算机断层扫描的原图像分割进行模拟实验。根据大量实验,分析肺部图像分割算法中存在的问题,为进一步的研究提供方向和宝贵的经验。更多还原

【Abstract】 Automatic segmentation techniques of lung image has an important clinical significance for the reduce misdiagnosis and missed diagnosis. Lung image segmentation extracts the first source image of the lung parenchymal get lung nodule characteristics, the final analysis and screening through the classifier, given the findings and diagnoses to help doctors analyze and judge. Extraction of pulmonary parenchyma in the lung image segmentation system has an important significance.First of all, we aim at through the lungs of the existing automatic detection system for analysis, for active contour model can not solve the problem of deep depression is proposed to improve the energy formula based on GVF Snake algorithm, by introducing the image information such as gradient and curvature, in their common of forces, the improvement of the energy equation to determine the lung border, solve the active contour algorithm is unable to carry out the division of the deep depression.Secondly, we research deeply based on the improvement of the energy equation, the stop condition of the curve will change when GVF sanke algorithm stop, the curve will occur can not stop crossing the border and converge to the true point phenomenon, put forward by gradient information and curvature information to external guidance direction, the introduction of sign function to re-set the curve of the movement to stop the conditions to solve the boundary curve can not converge to the real problems and improve the speed of image segmentation and reduce the algorithm’s time complexity.Then,we proposed adaptive border marching algorithm for the pleural adhesion at segmentation of lung nodules in the lungs appeared discontinuous or continuous, boundary is the boundary with the actual lung very far split the leakage issue, it is a new frontier repair method, using geometric means smooth lung boundary, solve the leakage lung cancer division of adhesions.Finally, in this paper we use the original CT image segmentation to experiment. According to a large number of experiments, analysis of lung image segmentation problems, provide direction for further research and valuable experience.更多还原