Kidney Segmentation In Mri Sequences Using Temporal Dynamics

We propose an energy-based image segmentation algorithm that uses the correlation information among pixels in the same image as well as the temporal correlation across the images in the sequence. We focus on MRI sequences that are extremely difficult to segment on the basis of single images. Our method detects motion-free objects whose intensities change across the image sequence. We introduce an energy functional that exploits the difference in the dynamics of the temporal signals associated with distinct pixels, We develop a level set approach and a region-growing algorithm to minimize the energy functional. Our tests in a transplantation study show that we successfully extract automatically the kidneys and their structures in magnetic resonance (MR) image sequences.