Parallel registration of multi-modal medical image triples having unknown inter-image geometry

A method is proposed to register three multimodal medical data, where none of the images are superimposed. Contrary to previously presented solutions that perform more simultaneous registrations after one-by-one, present method registers all images in parallel. The method minimizes the registration error by seeking the optimum of a vector including rigid transformation parameters of both reslice images. To measure the similarity among all three images, a higher dimensional extended normalized mutual information have been adopted. Comparison with simultaneous methods have been performed on brain and femoral multi-modal image triples. Based on the comparative results, presented parallel method significantly outperforms the simultaneous methods in both translation and rotation registration error minimizations. On the contrary, the simultaneous methods need less computational time to converge.