A Coarse-to-Refined Approach of Medical Image Registration Based on Combining Mutual Information and Shape Information

Mutual information (MI) is currently one of the most effective similarity metric in medical image registration. It is an automatic measure, and suitable for multimodal image registration. However it ignores the global spatial information inherent in the images. In addition, the mutual information-based registration is a time-consuming work and can lead to misalignment. A coarse-to-refined medical image registration method is presented in this paper, by combining mutual information with shape information of the images. The gradient images of the registration images generate the principal axes and centroids, and then the two images can be aligned coarsely according to these shape parameters. Mutual information is used to refine the registration. Experiment results demonstrate the presented method reduces the time consumed by 97 percent than the registration using MI alone, which is helpful in clinical applications