Non-Rigid Registration of Multi-Modality Medical Image Using Combined Gradient Information and Mutual Information

Non-rigid multi-modality medical image registration is a vital problem in medical image analysis, and mutual information (MI) is a popular similarity measure. However, mutual information performs badly in the low-contrast regions due to the intensity discretization procedure before the accumulation of statistical entropy. A possible solution is to add spatial information into the criterion. In this paper, a novel intensity-based similarity measure is proposed for deformable medical image registration, and both mutual information and gradient information are considered. The basic idea is to distribute mutual information to each pixel, and the discrete mutual information is multiplied with a defined weighting term, which is a function of the gradient information. According to our definition, the proposed measure provides a strict restriction that gradients in reference and float images should have the same orientation, either identical or opposing directions. By adding of the weighting term, the registration of the strong gradient regions has a priority over the small gradient regions. In addition, several non-rigid registration experiments have been carried out to verify the effectiveness of the proposed method. The results indicate that one of the key advantages of the new similarity measure is its restriction in the low-contrast regions. The proposed method outperforms MI, sum of squared distances (SSD) and multifeature mutual information (MMI) with respect to registration accuracy and sensitivity to the grid spacing, and it performs better than residual complexity (RC) when the grid spacing becomes large and the registration results become stable.