Fast free-form registration based on Kullback-Leibler distance for multimodal medical image

Multimodal image registration is a great challenge in medical imaging which requires to aligning anatomically identical structures. Their appearance in images acquired with different multimodality images may be very different. In order to realize fast, precise and robust image registration, we proposed a new multimodal medical registration method, which is based on free-form deformation (FFD) and Kullback-Leibler distance (KLD) algorithm. We made the affine transform between the source image and target image as global transformation and FFD as local transformation took KLD as the similarity metric. Particle swarm optimization (PSO) and nonlinear least-squares (NLLS) algorithm were adopted to find the optimal transform parameters. Experimental results show that, as compared with the mutual information, normalized mutual information and cross correlation based registration methods, the proposed method has longer capture range at different image resolutions. Besides faster, its implementation can lead to a more robust performance and outperforms the state of the art methods for multimodal medical image registration.