基于体素相似性的三维多模脑图像配准研究

摘要

The registration method based on voxel similarity is proposed to study the 3D multi-modality brain image registration. The 3D space transformation model and the 3D volume data are further studied and the reason of mutual information algorithm dropping into local maxima is analyzed. In this paper, Mattes mutual information algorithm with multi-resolution strategy is introduced to register 3D multi-modality brain image. Mattes mutual information algorithm with multi-resolution strategy has the advantages of high accuracy and good robustness. And the method can also reduce the possibility of the measurement function dropping into a local optimal solution. All the registration errors are smaller than a pixel size and the registration accuracy is close to sub-pixel. Mattes mutual information algorithm with multi-resolution strategy can improve the registration speed, accuracy and robustness of 3D multimode brain image.%　　为了对不同模态下的三维脑图像进行配准研究，引入一种基于体素相似性的配准方法。深入研究三维空间变换模型与三维配准体数据，并分析了互信息算法陷入局部极值的原因；本研究使用具有多分辨率策略的 Mattes 互信息算法对三维多模脑图像进行配准，通过对多模三维体图像进行三维单模和多模配准发现这种混合算法精度高，鲁棒性强，有效地降低了测度函数在收敛过程中陷入局部最优的可能性，所有的配准误差都小于一个像素大小，配准的精度达到亚像素级标准。即加入多分辨率策略的 Mattes 互信息算法较好地提高了三维多模脑图像配准的速度、精度和鲁棒性。