Multi-modal Medical Image Registration Based on Adaptive Combination of Intensity and Gradient Field Mutual Information

Mutual information (MI) is an effective criterion for multi-modal image registration. However the traditional MI function only includes intensity information of images and lacks sufficient spacial information to accurately measure the degree of alignment of two images, and besides, it is apt to be influenced by intensity interpolation, therefore presents many local maxima which frequently lead to misregistration. Our paper proposes a criterion of adaptive combination of intensity and gradient field mutual information (ACMI). Unlike the intensity MI computed from two original images, the gradient field MI of two images is calculated from their gradient code maps (GCM) constructed by coding the gradient field information of corresponding original image. Because of their complementary properties, these two MI functions are combined to form ACMI by a nonlinear weight function which can be adaptively regulated according to their performances and make the better dominant in the combination. Experimental results demonstrate that the ACMI outperforms the traditional MI and furthermore the former is much less sensitive than the latter to the reduction of resolution or overlapped region of images.