胸部CT中肺实质的自动分割及粘连肿瘤检测  

Automatic lung parenchyma segmentation and attached lung nodule detection in thoracic CT images

作　　者：裴晓敏[1] 郭红宇[1] 戴建平[1,2] 

机构地区：[1]东北大学中荷生物医学与信息工程学院,辽宁沈阳110004 [2]首都医科大学附属北京天坛医院神经影像中心,北京100050

出　　处：《哈尔滨工程大学学报》2010年第5期679-682,共4页Journal of Harbin Engineering University

基　　金：辽宁省自然科学基金资助项目(20072038)

摘　　要：为了实现胸部CT图像肺区的完全自动分割与辅助诊断,提出了一种自动肺实质分割算法,即引入基于量子粒子群优化的二维直方图阈值分割算法结合3-D区域生长,通过分割背景及胸腔实现肺实质分割,并提出行扫描曲率分析法实现粘连肿瘤检测及左右肺分离.该方法有效解决了肺实质分割中高密度特征易丢失、边缘肿瘤易遗漏等问题.通过多组胸部CT序列图像的实验,证明该方法对于肺实质分割非常精确有效,与传统分割算法相比,在分割精度上有明显提高.Completely automatic segmentation of lung parenchyma and computer-aided diagnosis （CAD） of lung diseases should be achievable. In this study, an automated method for lung parenchyma segmentation was pro- posed. First, two dimensional threshold image segmentation was done based on a quantum-behaved particle swarm optimization. Combined with 3-D region growing, it was used to separate a 2-D computed tomography （CT） image into chest and background. Next, 3-D lung parenchyma segmentation was done by segmenting chest and back- ground from the image. Line search curvature analysis was then applied to detect attached lung nodules and sepa- rate the left lung from the right. The proposed method solved problems such as high-density information being easy to lose, attached nodules being easy to miss, etc. resulted from such segmentation. Experimental results from many sets of CT images verified that this method is precise and efficient for lung parenchyma segmentation. Compared with traditional algorithms, this method is more robust and accurate.

关 键 词：自动肺实质分割 量子粒子群优化 三维区域生长 行扫描曲率分析 粘连肿瘤 胸部CT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]