Nonrigid Registration of Dynamic Renal MR Images Using a Saliency Based MRF Model

Nonrigid registration of contrast-enhanced MR images is a difficult problem due to the change in pixel intensity caused by the wash-in and wash-out of the contrast agent. In this paper we propose a novel saliency based Markov Random Field approach for effective nonrigid registration of contrast enhanced images. Saliency information obtained from the neurobiology-based saliency model alongwith intensity information is used to quantify the degree of similarity between images in the pre- and post-contrast stages. Information from these two features is combined by using an exponential function of the saliency difference such that it assigns low values to small differences in saliency and at the same time ensures that saliency information does not bias the energy term. Rotationally-invariant edge information from edge-orientation histograms was used to complement the saliency information resulting in better registration results. Tests on real patient datasets show that our algorithm results in accurate registration. We also simulated elastic motion on images, and the deformation field recovered by our algorithm was nearly the inverse of the simulated field.