一种面向人体管状组织的三维骨架提取算法  

3D skeleton extraction algorithm for human tubular tissues

作　　者：耿欢[1,2] 杨金柱[2] 赵大哲[1,2] 覃文军[2] 袁玉亮[2] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110819 [2]东北大学医学影像计算教育部重点实验室,沈阳110179

出　　处：《仪器仪表学报》2014年第4期754-761,共8页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61172002;61302012);辽宁省自然科学基金(201320021);中央高校基金科研业务费(N120518001;N110718001)资助项目

摘　　要：针对人体管状组织的拓扑结构复杂、在影像上管壁存在噪声、骨架难以准确提取的问题,提出了一种新的基于快速步进方法(FMM)优化双距离场和改进SUSAN端点检测的全自动三维骨架提取方法。考虑近似欧拉距离计算会使骨架不居中,文中采用快速步进方法构造三维高精度的双距离场,然后用最小代价路径机制搜索连通的、单像素宽的分支;融合源距离场极值和三维SUSAN形态特征检测分支端点,从而在抑制边界噪声的同时保持细微分支拓扑结构。将该方法用于对肺气管、脑血管、冠状动脉进行骨架提取,实验结果表明该方法更居中、更平滑、对边界噪声不敏感,较传统的双距离场方法、细化方法提高了骨架的准确性和计算效率。Aiming at the problems of complex topological structure of human tubular tissues,the noise existing on the tubular wall in the image and the difficulty in accurate skeleton extraction,a novel automatic 3D skeleton extraction method based on the double distance transform fields optimized by the fast marching method （FMM） and the improved SUSAN endpoint detection is presented in this paper.Considering that approximate calculation of Euler distance will cause the skeleton not centered,the 3D double distance transform fields with high precision are constructed with the fast marching method,and then the minimum cost path mechanism is used to search for the connected,single pixel wide branch.The branch endpoints are detected through fusing the local extreme value of distance from source （DFS） and the 3D morphological characteristics of SUSAN comer detection; the topological structure with fine branches is preserved while the boundary noise is also suppressed at the same time.The proposed method was used in the skeleton extractions of patient-specific lung trachea,cerebral vascular and coronary artery.Experiment results show that the proposed method can obtain more centered and smoother skeleton,is less sensitive to boundary noise and can improve the accuracy and computation efficiency of skeleton extraction compared with conventional double distance field method and thinning method.

关 键 词：骨架 中心线 中心路径 水平集 快速步进 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]