文章摘要

文章信息/Info

作者:: 吴凉1; 吕晓琪1; 2△; 谷宇1; 3; 李菁1; 张文莉1; 任国印1; 张薇1; 1.内蒙古科技大学信息工程学院，内蒙古自治区模式识别与智能图像处理重点实验室，包头 014010；2.内蒙古工业大学，呼和浩特 010051；3.上海大学计算机工程与科学学院，上海 200444

Author(s):: WU Liang1 ; LU ?Xiaoqi1; 2; GU Yu1; 3 LI Jing1; ZHANG Wenli1; REN Guoyin1; ZHANG Wei1; 1. Inner Mongolia Key Laboratory of Pattern Recognition and Intelligent Image Processing , School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou 014010，Inner Mongolia,China;2.Inner Mongolia University of Technology,Hohhot 010051，Inner Mongolia；3.School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China

Keywords:: Lung parenchyma segmentation; Low dose CT; Hough transform; Algorithm of convexhull; Watershed; Detection

摘要:: 为了提高早期肺癌筛查中肺结节计算机辅助检测、辅助诊断的准确性，提出一种多种方法结合的低剂量CT图像肺实质自动分割算法。首先利用改进的多方向形态学滤波算法进行预处理；然后利用聚类法、flood-fill算法去除背景，实现粗分割；接着利用引入霍夫变换的改进三维区域生长算法去除气管和主支气管树；最后利用分水岭算法和二维凸包算法实现肺实质细分割。实验结果通过对ELCAP数据库中的50个低剂量CT序列利用本研究算法进行处理，验证了该算法的有效性，正确分割率达到95.75%。为肺结节检测等后期的诊断提供了有价值的参考信息。

Abstract:: To improve the accuracy of computer-aided detection of pulmonary nodules in early lung cancer screening, a fully automatic 3-D lung parenchyma segmentation algorithm based on hybrid processing was presented. Firstly, the low dose CT images were preprocessed by using the improved multi-directional morphological filtering algorithm. Secondly, the rough segmented images were achieved using cluster method and flood-fill algorithm to remove the background. Then trachea and the main bronchus tree were removed using improved 3D region with hough transform. Finally the fine segmented image was achieved by using watershed and two dimension convex hull algorithm. 50 low dose CT images in the ELCAP database were collected to segment the lung parenchyma with using the proposed method, obtaining the 95.75% of accuracy .The proposed algorithm can provide some valuable information for pulmonary nodule detection.