基于ct图像的肺实质自动分割的分析-analysis of automatic segmentation of lung parenchyma based on ct images.docxVIP

摘要CT 影像是肺部疾病临床检查的重要手段，随着 CT 技术的进一步发展,肺部检查 所获得的数据量越来越大,巨大的数据量将对医生的诊断工作带来了极大的挑战。为 了减轻医生的工作量，更快更准确的检查出肺部疾病，将计算机的快速计算能力应用 到肺实质分割中来辅助医生进行诊断。肺部图像主要由肺实质和肺实质外的空气、检查床等部分组成，我们提取的肺部 区域就是指肺实质，即包括在肺壁中的区域。快速有效的分割出肺实质区域，是提取 病变区域的基础。所以，要实现基于 CT 图像的计算机辅助肺部疾病诊断，第一步工 作就是将肺部区域从 CT 图像中分割出来。目前针对图像的分割方法主要包括阈值法、区域生长法、基于模式分类的分割方 法、分水岭分割法等几种。但由于医学图像具有噪声大，图像结构复杂等特点，这些 方法对于医学图像的应用仍然达不到实用的效果。目前的分割方法普遍存在两个问 题：一是方法大多是针对具体问题的，还没有一种通用的解决方法；二是自动分割的 问题，虽然近年来专家学者们提出了很多自动分割算法，但是其精确性还达不到临床 应用的要求。为了实现胸部 CT 图像的完全自动分割，本文结合阈值处理、数学形态 学运算提出了一种新的肺实质自动分割方法。在新的算法中对传统的阈值分割算法做 了一定的改进，可以取到更为理想的阈值。针对传统修补算法滚球法修补效果受半径 的取值影响的问题，本文运用数学形态学提出了一种新的修补算法，提高了修补的准 确率。实验表明，本文提出的算法可以自动准确地分割出肺实质区域。关键词：CT 图像肺实质分割辅助诊断AbstractCT images are inspection means of pulmonary diseases in clinical. With the further development of CT technology, the amount of data of lung inspection obtained become lager and larger, such a huge amount of data bring a great challenge to doctors diagnosis. In order to reduce the workload of doctors, check out the lung disease quicker and more accurate, the fast computing capabilities was applied to lung parenchyma segmentation to assist doctors’ diagnosis.The lung image was composed mainly of lung parenchyma, air outside the lung parenchyma, the examining table and so on. The lung region examined is the lung parenchyma, the part includes in the lung wall. The extraction of lung parenchyma has important value guidance for doctors’ clinical diagnosis and surgery. So that the first step of pulmonary diagnosis through computer-aided based on CT images, is separating lung region out from the CT images.Image segmentation methods include thresholding method, region growing algorithm, method for segmenting image based on pattern classification, watershed segmentation method and so on. But these methods still could not achieve the practical effect in the medicine images because medical image with large noise and image characteristics. Currently, there are two problems in all segmentation methods: one is the method for