PET与MRI三维脑图像高精度配准

目的 在PC机上实现高精度的PET与MRI三维脑图像配准。方法 采用最大互信息法对6例患者PET和MRI三维脑图像进行刚体配准。使用归一化互信息作为相似性量度。在互信息计算过程中，使用Powell多参数优化法和Brent一维搜索算法。为加快配准速度，使用了多分辨金字塔方法。采用基于坐标的阈值选取方法对PET图像进行分割预处理，消除星状背景伪影。结果 配准误差平均值为2．6mm，误差中位数平均为2．7mm。结论 配准视觉效果良好，评估证明该算法可达亚体元级配准精度。

High accuracy registration for three-dimensional brain images of PET and MRI

Objective To implement high accuracy registration for three-dimensional images of PET and MRI on PC computer. Methods Maximization of mutual information method was used in the rigid registration for PET and MRI images of 6 patients. Normalized mutual information was adopted as the similarity measure. During our processing with mutual information, Powell's multi-parameters searching methods and Brent's one-dimensional optimization algorithm were chosen. The multi-resolution approach was applied to speed up the matching process. As blurs occurred in PET images, the PET images were pre-segmented to reduce the radiated artifacts. Results According to the evaluation of multiresolution by the Vanderbilt University, in this study, the average of mean of registration error was 2.6 mm and the average of median of registration error was 2.7 mm. Conclusions The registration images with edge enhancement showed good matches by visual inspection. Sub-voxel accuracy in multi-modality registration has been achieved with this algorithm.