【摘要】 目的基于多回路图像配准的质量评估算法(assessing quality using image registration circuits,AQUIRC)通过构建多回路图像配准网络,可直接利用空间变换场信息估计配准误差。本文研究了AQUIRC算法估计仿射变换相关的配准误差的性能,旨在给出对该算法较为全面的认识。方法采用脑磁共振(magnetic resonance,MR)图像设计不同形变程度的旋转、缩放、错切、平移等线性变换构造模拟误差,并在无差别均匀采样的空间位置处分析AQUIRC算法估计的配准误差与目标配准误差(target registration error,TRE)的线性相关性,给出二者的区域分布情况。结果在旋转、缩放、错切误差实验中,算法估计的配准误差与TRE呈较强相关,且区域分布基本一致。在平移误差实验中,二者不存在线性相关关系。将平移变换与其他类型足够大的变换复合构造模拟误差,则可一定程度上提高二者的相关度。结论 AQUIRC算法能够准确估计旋转、缩放、错切变换相关的配准误差,但不能正确估计单一平移变换造成的配准误差。更多还原

【Abstract】 Objective The algorithm,assessing quality using image registration circuits( AQUIRC),can estimate registration error within an image network with multiple registration circuits relying solely on deformation fields. In this paper,the performance of AQUIRC in estimating registration errors associated with affine transformation is studied to give a more comprehensive understanding of the algorithm. Methods Linear transformations including rotation,scaling,shear and translation of different deformation degree are designed to simulate registration errors among brain magnetic resonance( MR) images. The linear correlation between the registration errors estimated by AQUIRC and target registration error( TRE) is analyzed at non-discriminatory uniform sampling locations,and the regional distributions of both are shown. Results In rotation,scaling and shear cases,the correlation between AQUIRC ’s estimation and TRE is quite high,and their regional distributions are almost the same. In translation cases,however,no correlation is observed. Nevertheless,the correlation can be improved to some extent if translation error is combined with other type of transformation errors whose deformation degree is big enough. Conclusions AQUIRC can accurately estimate registration errors associated with rotation,scaling,and shear transformation,yet cannot correctly estimate errors caused by translation.更多还原