【作者】 李鹏；

【导师】 邵虹；

【作者基本信息】 东北大学 ， 生物医学工程， 2012， 硕士

【摘要】 随着经济增长和人们生活水平提高,肺部疾病开始日益突出。近年,肺癌发病率逐年增长,已成为危害和影响人类生命安全的主要因素。X线检查由于具有辐射小、价格低、操作简单、实用强等特点,仍是放射科影像检查的首选手段。x线胸片记录着胸部健康与病灶的信息,目前,基于X线胸片的肺部疾病检查占到影像诊断领域的40%左右,因此X线胸片有着重要的医学应用研究价值。然而,X线胸片图像存在很多缺陷和不足,如：分辨率较低、图像中各个组织之间相互重叠,致使x线胸片中几乎没有一种组织存在着明确的边界,导致病灶很容易被其它组织掩盖,难以分辨或很容易遗漏,严重影响肺部疾病的诊断。因此,有关肺部疾病的计算机辅助诊断已成为当代医疗科学研究领域的重大课题。本文以X线胸片计算机辅助诊断为背景,结合X线胸片图像中的一些问题,对其中的X线胸片图像肺野分割部分进行研究。在分析和总结现阶段肺野分割方法基础上,结合X线胸片图像的特点,采用改进二维Otsu算法,提出一种基于Otsu与数学形态学相结合的肺野分割方法,首先,利用二维Otsu算法得到胸片图像的全局阈值对胸片图像进行初次分割；接着,在全局阈值基础上逐渐缩小阈值,选取肺野局部区域分割的最佳阈值,对肺野图像进行二次分割,通过两次分割处理得到较完整的肺野二值图像；最后,通过对肺野二值图像进行填补、边缘平滑、轮廓提取等操作,实现胸片图像肺野分割。由于该方法实现肺野分割操作复杂、阈值选取不精准,导致分割的实效性和精确度不高。因此,为了进一步提高分割的效率和精确度,本文又提出一种基于遗传算法的Otsu肺野分割方法,该方法是以改进的二维Otsu算法作为遗传算法的适应度函数,同时结合X线胸片图像特点,对遗传算法的编码、选择、交叉、变异等遗传操作进行相应改进,充分发挥了遗传算法的全局搜索能力、整体寻优策略和对全局最优解逼近性等优点,搜索X线胸片图像肺野分割的全局和局部的最优阈值,通过全局和局部最优阈值实现肺野分割,提高分割的效率和准确度。经过对数据库中选取的54张x线胸片图像进行肺野分割测试,本文提出的基于遗传算法的Otsu肺野分割方法,快速有效的完成肺野分割,并取得较好的分割效果。结果表明,本文提出的肺野分割方法具有分割速度快、精确度高、灵敏度强等特点,在实际X线胸片图像肺野分割应用中具有可行性。更多还原

【Abstract】 With the development of economic growth and people’s living standards, lung disease becomes increasingly. In recent years, the incidence of lung cancer is increasing year by year which has become the main factor of the harm and impact the safety of human life. X-ray has the features of small radiation, low price, simple operation, strong usefulness and so on, which is still the preferred method of radiology imaging inspection. Chest X-ray radiograph records the information of health or lesions, it is accounted for40%field of diagnostic imaging of lung disease based on chest X-ray inspection, as a result, X-ray has important medical applications for research value. However, chest X-ray radiograph has many disadvantages and deficiencies, such as:low resolutions, the organizations in images are overlap with each other which leads to almost any organizations in the chest X-ray radiographs do not have clear edge, and the lesions are very easily to be covered up by other organizations, that results in lesions are difficult to distinguish or easily omitted, so it seriously impact on the diagnosis of lung disease. Therefore, computer-aided diagnosis of lung disease has become a major issue in contemporary medical and scientific research.Based on the background of chest X-ray radiograph computer-aided diagnosis, combine with some problems of the chest X-ray radiographs, and research on segmentation of lung fields in chest X-ray radiographs. At the basis of the present lung field segmentation method analysis and summary, combine the characteristics of chest X-ray images and use the improved2d Otsu algorithm, propose a lung field segmentation method which based on combining Otsu and mathematical morphology. First of all, using2d Otsu algorithm obtain image global threshold to implement splitting the original segmentation chest X-ray radiograph; Furthermore, gradually reduce the threshold based on global threshold, and select the optimal threshold value to segment lung fields, as a result of which, to get a more perfect lung field binary image. Finally, by filling up the lung field binary image, smoothing edge and contouring extraction, implement the lung segmentation of chest X-ray radiograph. However, this segmentation of lung field method is complex and the threshold selection is also not accurate, result in the low effectiveness and the low accuracy. Therefore, in order to improve the efficiency and the accuracy of segmentation, in this article, I propose a lung field segmentation method based on genetic algorithm and Otsu algorithm. This method uses improved2d Otsu algorithm as the sufficiency function of genetic algorithm, and at the same time, combine the characteristics of X-ray images to improve the genetic algorithm operation, such as:encoding, selection, crossover, mutation operation and so on, which also fully shows the advantages of genetic algorithm, such as:global search ability, overall optimization strategy and global optimal approximate solution, search the global and local optimal threshold for segmentation of lung field in chest X-ray radiograph. Using global and local optimal threshold to achieve segment lung field and improve the efficiency and accuracy of segmentation.In a conclusion, I select54chest X-ray radiographs from the databases to test segment lung field segmentation, the segmentation method based on combination of genetic algorithm with Otsu could fast and efficient to implement lung field segmentation, and could also get ideal segmentation results. The results showed the lung field segmentation method has high speed, high accuracy, and strong sensitivity. Furthermore, this method is feasible in the practical chest X-ray radiographs in lung field segmentation applications.更多还原