基于三维动态区域生长算法的肝脏自动分割

摘要

目的:研究三维动态区域生长算法在肝脏自动分割中的可行性.方法:首先对CT图像进行预处理,包括插值、各向异性滤波和三维矩阵化处理;然后在预处理过的图像上使用三维动态区域生长算法,得到初始分割结果,算法在执行过程中使用种子点联合26邻域体素灰度均值代替传统的区域生长算法中选取的单一种子点灰度值,并结合动态阈值和双生长准则提高边缘分割精度;最后通过形态学后处理得到肝脏分割结果.结果:利用三维动态区域生长算法的肝脏自动分割结果接近专家手动勾画的结果(戴斯相似系数平均达到0.934),并且其分割速度(每幅图像用时0.64 s)和三维连续性优于手动勾画.结论:三维动态区域生长算法能够精确地分割肝脏,能在腹部肿瘤放疗计划制定中大幅度提高肝脏勾画效率.%Objective To evaluate the feasibility of three-dimensional (3D) dynamic region growing algorithm for liver autosegmentation.Methods The CT images were pro-processed using three methods,namely interpolation,anisotropic filtering,and 3D matrix processing.And then,initial segmentation results were obtained using 3D dynamic region growing algorithm.The algorithm combined the seed points with the grey average of 26 adjacent voxels,as comparison with one single seed point in traditional algorithm.Meanwhile,we used double growing criteria and dynamic thresholds to ensure the accuracy of segmentation.Finally,we obtained the liver segmentation results by morphological post-processing.Results The segmentation results in this study were almost consistent with the manual contour obtained by experts (the mean value of Dice's similarity coefficient was 0.934).The 3D dynamic region growing algorithm also had an improved segmentation speed (0.64 s/per image) and a better continuity in 3D directions than manual segmentation.Conclusion The 3D dynamic region growing algorithm can be used to accurately segment liver,which greatly improves the efficiency of live segmentation in abdominal tumor radiotherapy plan.