This paper presents an integrated image registration algorithm to correct the motion induced by patient breathing for dynamic renal perfusion MR images. Registration of kidneys through the MR image sequence is a challenging task due to rapidly changing image contrast over the course of contrast enhancement. Our algorithm achieves temporal image registration in a multi-step fashion. We first roughly register the images by detecting large-scale motion, and then refine the registration results by integrating region information and local gradient information with auxiliary image segmentation results. We have tested the proposed algorithm on several real patients and obtained excellent registration results.
José M. F. Moura, Marie-Pierre Jolly, Ying