Global search for image registration based on normalized mutual information

Multimodality medical image registration gives insight to anatomical and functional information, which assist diagnosis and therapy planning in medicine. Registration and consecutive visualization allow combining the image data and thereby conveying more meaningful images to the clinician. In recent years, many approaches used information theory as registration criterion. In this paper, a voxel-based approach based on normalized mutual information is applied; accurate and retrospective registration is provided. A global search approach is proposed here, which combined genetic and Powell algorithm in optimization. The results show that it is a feasible way of determining the global solution, which is robust to local maximum and insensitive to initial positioning.