Image registration by maximization of combined mutual information and gradient information

Despite generally good performance, mutual information has also been shown by several researchers to lack robustness for certain registration problems. A possible cause may be the absence of spatial information in the measure. The present paper proposes to include spatial information by combining mutual information with a term based on the image gradient of the images to be registered. The gradient term not only seeks to align locations of high gradient magnitude, but also aims for a similar orientation of the gradients at these locations. Results of combining both standard mutual information as well a normalized measure are presented for rigid registration of three-dimensional clinical images (MR, CT and PET). The results indicate that the combined measures yield a better registration function than mutual information or normalized mutual information per se. The accuracy of the combined measures is compared against a screw marker based gold standard, revealing a similar accuracy for the combined measures to that of the standard measures. Experiments into the robustness of the measures with respect to starting position imply a clear improvement in robustness for the measures including spatial information.