基于向量分区模型和三维骨架化模型的肺裂分割算法  

Pulmonary Fissure Segmentation in CT Scans Based on Vector Partition Model and 3D Skeletonization Model

作　　者：彭圆圆 马招娣 彭兰 李雄[2] Peng Yuanyuan;Ma Zhaodi;Peng Lan;Li Xiong(School of Electrical and Automation Engineering,East China Jiaotong University,Nanchang 330000;School of Software,East China Jiaotong University,Nanchang 330000;School of Materials Science and Engineering,East China Jiaotong University,Nanchang 330000)

机构地区：[1]华东交通大学电气与自动化工程学院,南昌330000 [2]华东交通大学软件学院,南昌330000 [3]华东交通大学材料科学与工程学院,南昌330000

出　　处：《计算机辅助设计与图形学学报》2020年第7期1154-1161,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61172160,61571184,61602174);中国博士后科学基金(2017M620276);江西省自然科学基金(20161BAB212052,20171BAB202019,20192ACB21004);江西省教育厅科技计划项目(GJJ190356,GJJ150496,GJJ170385);长沙市科技计划项目(kq2001014);华东交通大学科研基金(2013044).

摘　　要：了解肺裂结构特征在肺部疾病诊断方面具有重要的应用价值.在CT图像中,肺裂会受到强度变化、噪声和形变等各种因素的影响,使得分割难度很大.为了克服此困难,提出一种基于向量分区模型和三维骨架化模型的肺裂分割算法.首先,受到肺裂在二维CT图像中呈现为线性结构特征的启发,通过借助方向窄条微分滤波器融合幅度信息和方向信息来构建肺裂向量场,此种途径能够有效地区分肺裂和其他组织.接着,考虑肺裂在三维CT图像中呈现为面状结构特征,通过改进的向量分区模型来分离肺裂和管状结构,此种方式能够高效率地抑制噪声.最后,为了得到完整的肺裂,引入三维骨架化模型理论来有效分割肺裂.提出的算法在公开数据集LOLA11上进行验证,其F1,FDR和FNR中值分别为0.893,0.068和0.128.实验结果表明,文中算法具有非常好的肺裂分割效果.Knowledge of structure features of pulmonary fissures is useful in diagnosis of lung diseases.In CT images,pulmonary fissure segmentation is a formidable task due to various of factors such as intensity variability,clutters and pathological deformation.To overcome the problem,a computerized scheme based on vector partition model and 3 D skeletonization model are introduced for pulmonary fissure segmentation.Motivated by the fact that fissures appear as line-profiles in 2 D space,we adopt an oriented derivative of stick filter by merging the intensity and orientation information to construct fissure vector field,which can effectively discriminate between fissures and other tissues.Considering the fact that fissures appear as planar patches in 3 D space,an orientation partition scheme is adopted to separate fissure patches and other tissues in different orientation partition,thus resulting a good performance in clutters suppression.To achieve a complete fissure segmentation,a post-processing pipeline based on 3 D skeletonization model is introduced for clutters removal.When applying the computerized scheme to publicly available LOLA11 datasets,the median F1,FDR and FNR are 0.893,0.068 and 0.128,respectively.Which indicates that the computerized scheme can efficiently segment pulmonary fissures in CT images.

关 键 词：CT图像 向量分区模型 骨架化模型 向量场 分割肺裂 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]