有限元弹性配准中的驱动外力及其网格细化  

The Driving Force and Mesh Refinement for Finite Element Elastic Registration

作　　者：张静亚[1] 李周雁 ZHANG Jingya;LI Zhouyan(School of Physics and Electronic Engineering,Changshu Institute of Technology,Changshu 215500;Wuzhong Branch,Industrial and Commercial Bank of China,Suzhou 215006,China)

机构地区：[1]常熟理工学院物理与电子工程学院,江苏常熟215500 [2]中国工商银行吴中支行,江苏苏州215006

出　　处：《常熟理工学院学报》2018年第5期73-79,共7页Journal of Changshu Institute of Technology

摘　　要：为了减小由于图像亮度差异和局部非均匀形变造成的配准误差,本文通过引入驱动点邻域的空间信息和灰度信息,并结合结构张量和鲁棒函数,得到了一个改进的驱动外力.在模型的有限元法求解中,提出了对细节区域进行网格细化的策略,该方法避免了计算误差在金字塔迭代中的传递放大.本文分别对20组肺部CT图像和20组脑部MR图像进行了弹性配准实验,采用归一化互信息、归一化互相关系数、峰值信噪比、均方误差,以及尺度不变特征变换(SIFT)匹配特征点平均距离来定量评估图像匹配度.实验结果表明,相对于传统的弹性配准,本文算法的配准精度获得了显著提高,t检验得到的P值小于0.05.In order to reduce the registration error caused by image brightness difference and local nonuniform deformation,an improved driving force is obtained by introducing spatial information and grayscale information of the neighborhood of the driving point in combination with structural tensor and robust function.In the finite element method of the model,a mesh refinement strategy for detailed areas is proposed,which avoids the propagation and amplification of computational errors in the pyramid iteration.In this paper,20 groups of lung CT images and 20 groups of brain MR images are tested for elastic registration respectively.The normalized mutual information,the normalized cross-correlation coefficient,the peak signal-to-noise ratio,the mean square error,and the scale-invariant feature transform(SIFT)are used to match the average distance of the feature points to quantitatively evaluate the image matching degree.The experimental results show that compared with the traditional elastic registration,the proposed algorithm accuracy is significantly improved at the 0.05 level.

关 键 词：图像配准 连续介质弹性力学 有限元 相似度 网格细化 

分 类 号：TN911.73[电子电信—通信与信息系统]