结肠形态仿真及中心线提取算法评价  

Simulation of Colon Structure and Evaluation of Center Line Extraction Algorithms

作　　者：张滢[1] 冯筠[1] 崔磊[1] 赵翊凯[1] 贺小伟[1] 

机构地区：[1]西北大学信息科学与技术学院,西安710127

出　　处：《系统仿真学报》2014年第3期601-608,614,共9页Journal of System Simulation

基　　金：国家自然科学基金(61372046);陕西省科技计划项目(2012KJXX-29);陕西省教育厅自然科学计划(2011JK1026)

摘　　要：结肠管腔中心线的提取在结肠疾病的计算机辅助检测中起到重要的作用,对中心线提取结果的定量评价一直是辅助检测中的难点。从中心线参数方程和截面方程两方面考虑,生成了20套仿真数据,模拟了结肠高曲率、褶皱和管腔的形态特点,提出了一系列评价准则。还对最小路径(DIJ)和最大生成树(MST)算法及作者提出的基于生成树的中心线快速提取算法(Fast centerline extraction algorithm based on Maximal Spanning Tree,简称FMST)进行了性能比较。实验结果验证了最大生成树算法解决了中心线在高曲率的拐角问题,保持中心线提取准确率同时,提取速度平均提高80%以上,是三种算法中性能最好的。Centerline extraction of colon lumen is one of the important procedures for computer aid detection of colorectal diseases. In the mean time, the evaluation of extracted centerlines is still an open problem. Based on centerline and profile parameter functions, 20 analytic geometric models with centerline standard were generated, which simulated the colon structure characteristics, i.e. tubular shape, ellipse lumen and high curvature. Furthermore, Dijkstra＇s shortest path algorithm （DIJ）, Maximal Spanning Tree algorithm （MST） and the proposed fast centerline extraction algorithm based on Maximal Spanning Tree （FMST） were compared based on simulated models and a set of proposed quantitative criteria for evaluation. The experimental results demonstrate that MST and FMST achieve better accurate rates than DIJ, while avoiding hugging the corners around sharp turns. The proposed FMST is the best extraction algorithm which can speed up MST 80%.

关 键 词：结肠形态 中心线提取 仿真数据 评价准则 参数方程 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]