介绍了一种基于最大互信息原理的图象配准技术.并就实施最大互信息配准法的一些重要技术问题进行了研究,其中包括不增加新数据点的格点采样子集、不产生分数灰度值的PV插值技术和出界点策略等.该方法在搜索策略上采用了无需计算梯度的Powell算法.由于计算互信息的关键技术与有效的搜索策略的结合,使得该方法能快速、准确地实现多模医学图象的配准.用该方法对7个病人的41套CT-MR和35套MR-PET 3D全脑数据进行了配准,结果经美国Vanderbilt大学评估,全部达到亚象素级配准精度.该方法可以临床应用.

In this paper a maximization of mutual information based multi-modality medical image registration method is described. The method presented in this paper applies mutual information to measure the information redundancy between the intensities of corresponding voxels in both images, which is assumed to be maximal if the images are geometrically aligned. MI is used as a measure of similarity of two images. There exist many important technical issues to be solved about the method such as how to compute MI more accurately and how to obtain the maximization of MI, which are seldom mentioned in published papers. In this paper we provide some implementation issues, for example, subsampling, PV interpolation, outlier strategy. Powell searching algorithm is used which does not compute gradients. The combination of these computation techniques and searching strategy leads to a fast and accurate multi-modality image registration. The registration results of 3D human brain volume data of 41 CT-MR and 35 PET-MR from seven patients are validated to be subvoxel. The registration method is promising in clinical use.