蜂窝状的交互式三维分割方法

Cellular Interactive 3D Segmentation Method

目的:蜂窝状肺病变是一种常见的肺部影像学表现,因其边缘模糊、分布弥散的特点,难于分割。为准确地对蜂窝状肺病变进行三维分割,提出一种基于支持向量机的交互式三维分割方法。方法:根据肺部CT图像灰度分布特性基于自适应阈值分割算法对肺部进行分割,然后采用区域生长法将气管剔除出肺部区域,并针对肺部区域采用修正后的分水岭算法按照纹理分割成小区域,对各区域进行纹理特征的提取,采用训练后的支持向量机对各区域进行判别是否为蜂窝状病变区域。最后根据切片层数据间的关联性基于面积重叠去除假阳性区域。结果:针对临床已确诊的30例病例参照医生分割的金标准进行测试,对分割算法进行了敏感性、特异性、准确率等指标的评估,该方法能分割出可靠的蜂窝状肺部病变区域。

Objective：Honeycombing are common findings in pulmonary medical imaging.Because of its characteristics of fuzzy edge and dispersed distribution,it＇s hard to segment accurately.In order to segment the 3D structure of honeycombing,we propose an interactive 3D segmentation method based on SVM algorithm.Method：According to the gray distribution characteristics of CT images,the lung area is segmented based on adaptive threshold algorithm.After removing the trachea from the lung area using region growing algorithm,a modified watershed algorithm is applied for segmenting the lung area into several regions in accordance with the lung texture.Kinds of the regional texture feature are extracted and used by the trained SVM to determine whether the region is honeycombing lesion area.At the end,we remove false positive areas according to the overlap relationship between slices.Results：A contrastive analysis of the proposed algorithm＇s and radiologist＇s segmentation result for 30 clinically diagnosed cases shows that the sensitivity,specificity and accuracy of the proposed algorithm are good enough to extract reliable puhmonary lesion from CT images.