We propose a subpixel registration algorithm to deal with the motion induced by breathing for renal perfusion MR image sequences. Our approach minimizes an energy functional that integrates a motion model and temporal smoothness constraints. We then present a multistage level set method that segments different kidney structures in the registered MR image sequence. Tests with a rat kidney transplantation study show that the algorithm successfully compensates for the breathing motion and extracts the kidneys, cortex, medulla, and pelvis.
Chien Ho, José M. F. Moura, Ying Sun