CT and MR multimodal registration guided by weighted gradient images

An effective registration algorithm for computerized tomography (CT) and magnetic resonance (MR) images is introduced in this paper. This algorithm constructs a guided layer in a same modal space for the image pair through the introduction of weighted gradient images to navigate the registration behaviors of the original reference image and the moved image. Subsequently, a fine registration is implemented to get a final aligned result. As a result, both the convergence in optimization and the accuracy in registration can be guaranteed. The normalized mutual information has been adopted as the similarity measure and the hierarchical B-spline grid is used to define the freeform deformation field. Numerous contrastive implements in image registration have indicated that the proposed algorithm is able to guarantee the convergence in the process of iterations and lead to higher accuracy with more robust property to noise.