基于局部方差采样的2D/3D医学图像配准技术  

2D/3D medical image registration based on local variance sampling

作　　者：刘达[1] 谢永召[1] 倪自强[1] 王烁[2] 

机构地区：[1]北京航空航天大学机器人研究所,北京100191 [2]北京科技大学机械工程学院,北京100083

出　　处：《高技术通讯》2013年第8期827-832,共6页Chinese High Technology Letters

基　　金：863计划(2009AA045301)资助项目

摘　　要：为了提高医学图像引导治疗系统中2D/3D图像配准精度和鲁棒性,提出了一种基于灰度方差采样的局部等级相关的相似度测度方法。该方法首先将2D图像进行子区域分割,并根据子区域灰度方差进行采样,排除图像中对配准贡献小的像素区域,然后求取剩余子区域归一化局部等级相关和。由于等级相关不依赖图像之间的线性关系,所以对不同能量等级的图像配准收敛特性好。脊柱模型CT数据与模拟X射线图像配准实验结果表明,采样率为30%时,配准精度达到(0.21±0.03)mm,以平均目标误差小于5mm为准,算法的收敛率达到96%。A 2D/3D image registration method for image guided therapy using the metric function to obtain the sum of local rank correlations based on gray intensity variance sampling is proposed to improve the image registration＇ s precision and robustness.This method divides a 2D fixed image into subregions in the same size,and calculates intensity variances respectively,then those subregions with lower variances are excluded,and finally,the sum of local rank correlations is computed with the left subregions.The rank correlations are independent of linear relationship between two images,so a good converging result of the image registration with any energy level can be achieved.The experimental data from registration between CT image and simulative X-ray image of the spine showed that the registration algorithm converged at a probability of 96％ with an accuracy of （0.21 ＋ 0.03） mm at a sampling rate of 30％.

关 键 词：2D 3D配准 等级相关 手术导航 X射线图像 数字重建影像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]