摘要
目的本研究提出一种基于局部边缘特征的加权水平集演变算法,并应用于医学图像分割。方法首先,计算演变轮廓内外邻域的图像局部边缘特征;接着,计算演变轮廓邻近轮廓的平均边缘强度和图像梯度向量流场,构造加权函数项;然后,由此构建新的水平集算法能量函数的长度项和区域项,借助偏微分方程求得最小值获得图像目标理想边界。实验图像选用人工合成图像和临床实例图像,不同水平集算法分割性能采用Dice相似性系数。结果视觉分析显示,基于本研究算法获得的图像目标轮廓与真实图像目标区域边界吻合度最高。定量分析显示,基于本研究算法所得分割图像能获得更高的Dice相似性系数。此外,迭代次数较少时,本研究算法即可获得最佳目标轮廓,且增加迭代次数,本研究算法Dice相似性系数变化微弱不溢出。最后,初始轮廓位置不同,其他三种算法所得Dice相似性系数变化较大且低于本研究算法。结论本研究算法较其他水平集算法收敛速度快,对初始轮廓位置敏感度低,稳定性强,是一种可行的医学图像分割算法。
Objective To propose a weighted level set evolution based on local edge features for medical image segmentation.Methods Firstly,the local edge features from the adjacent region located inside and outside of the evolving contour were calculated.Then,the average edge intensity and the image gradient vector flow field of the adjacent contour of the evolving contour were calculated,and the weighted function term was constructed.Finally,the length term and the area term of the energy function of the new level set algorithm were constructed,and the minimum value was obtained by using the partial differential equation to obtain the ideal boundary of the image target.Synthetic images and clinical images were selected as experimental images.Dice similarity coefficient(DSC)was used for segmentation performance of different level set algorithms.Results Results of visual analysis showed that the image target contour obtained based on the algorithm had the highest degree of coincidence with the real image target area boundary.Results of quantitative analysis showed that our proposed method could obtain higher DSC than other methods based on this algorithm.Additionally,when the number of iterations was relatively small,the optimal target contour could be obtained by the algorithm in this study,and when the number of iterations was increased,the DSC changed slightly and did not overflow.Finally,with different initial contour positions,the DSC obtained by the other three algorithms varied greatly and were lower than those obtained by this algorithm.Conclusion Compared with other level set algorithms,this algorithm has faster convergence speed,lower sensitivity to the initial contour position and strong stability,so it is a feasible medical image segmentation algorithm.
作者
魏应敏
王薇
张媛
WEI Yingmin;WANG Wei;ZHANG Yuan(Department of Radiology,Nanjing First Hospital,Nanjing Medical University,Nanjing Jiangsu 210006,China)
出处
《中国医疗设备》
2021年第1期94-98,共5页
China Medical Devices
关键词
图像分割
医学图像
水平集
活动轮廓
相似性系数
image segmentation
medical images
level set
active contour
similarity coefficient
