计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (13): 201-205.

• 图形图像处理 • 上一篇    下一篇

结合区域生长与模糊连接度的肺气管树分割

彭  双,肖昌炎   

  1. 湖南大学 电气与信息工程学院,长沙 410082
  • 出版日期:2016-07-01 发布日期:2016-07-15

Segmentation of pulmonary airway tree by combining region growing and fuzzy connectedness

PENG Shuang, XIAO Changyan   

  1. College of Electrical and Information Engineering, Hunan University, Changsha 410082, China
  • Online:2016-07-01 Published:2016-07-15

摘要: 针对CT图像中因噪声、密度分布不均匀和边界模糊等因素造成肺气管树难以准确分割的问题,提出了一种区域生长与模糊连接度相结合的肺气管树分割流程。通过阈值化及形态学闭操作提取出肺实质以定义感兴趣区域;采用改进迟滞阈值区域生长法预分割出较粗气管并结合局部体积突变指标抑制侧向泄漏;将预分割的结果进行骨架化及修剪来进一步提取出分支点,并以此作为后续分割的新种子点;根据CT图像的灰度均匀性与气管的管状结构特征来构造亲和力函数以计算种子点与其他体素的模糊连接度,并选取合适的阈值对模糊连接度进行阈值分割以提取出完整气管树。实验采用了20例来自EXACT’09竞赛提供的公开数据,分别从分支点、分支数量和分支数比率等方面进行了量化评估。该方法能在较低泄漏情况下成功检测出参考标准中一半以上的分支,平均分支数比率达到59.7%。实验结果表明,该方法可对肺气管树进行较精确的分割。

关键词: 气管树分割, 区域生长, 模糊连接度, CT图像

Abstract: Considering the noise, intensity inhomogeneity and boundary fuzzy in CT image, it is difficult to segment pulmonary airway tree accurately. To improve the segmentation results, 3D multi-seeded fuzzy connectedness algorithm is proposed. Firstly, the region of interest is defined by extracting the lung parenchyma with a global threshold and a morphological closing operation. Secondly, the trachea and big bronchi are pre-segmented using an improved region growing method on basis of an iterative hysteresis threshold, and a local volume explosion index is adopted to suppress the lateral leakage. Then, branch points are extracted from the skeleton which is extracted and pruned from the pre-segmentation results, and these points as the new seed points for subsequent segmentation. Finally, the fuzzy connectedness between the seed and any point is calculated by constructing affinity function according to intensity homogeneity and tubular structure characteristics of the trachea. Besides, the fuzzy connectedness is segmented by choosing an appropriate threshold. The algorithms are tested with the publicly available data of the EXACT’09 challenge, and the quantitative evaluation is conducted with the node number, branch number and branch ratio on the 20 test CT cases by comparing with a manual reference. The proposed method is able to detect more than half of branches in the reference and the mean numbers of detected branches reach 59.7% under a relatively low leakage rate. The experimental results show that the method has more accurate segmentation results.

Key words: airway tree segmentation, region growing, fuzzy connectedness, CT image