Multi-resolution three-dimensional multi-modality image registration by maximization of mutual information

Maximization of mutual information is a very powerful criterion for 3D medical image registration, allowing robust and accurate fully automated rigid registration of multi-modal images in a various applications. In this paper, we presented a method based on normalized mutual information with sub-sampling of the images for 3D image registration on the images of CT, MR and PET. Powell's direction set method and Brent's one-dimensional optimization algorithm were used as optimization strategy. A multi-resolution approach was applied to speedup the matching process. For PET images, preprocessing of segmentation was performed to reduce the background artifacts. According to the evaluation by Vanderbilt University, the average of mean of registration error for CT-MR task was 1.47 mm and for MR-PET task was 3.22 mm. The registration images with edge extraction showed good matches by visual inspection. Sub-voxel accuracy in multi-modality registration had been achieved with this algorithm.