Using total-variation regularization for deformable registration of the utero fetal MR images

Magnetic resonance imaging (MRI) has become an important tool in the assessment of fetal anomalies and the central nervous system. Accurate modeling of fetal motions remains illusive because of the lack of an effective mechanism to deal with the discontinuous movements of the involved anatomic structures. In this work, we propose an efficient deformable registration algorithm to deal with the fetal motion. Instead of directly applying the least square optimization, we include a totalvariation regularization to account for the motion discontinuity close to the contact surface between fetal and uterus wall. The term of total variation calculates the sum of absolute values of the derivatives, and the penalties drive the derivatives toward zeros and force the optimized displacement vector close to be piece-wise continuity. The proposed approach is evaluated using a series of fetal cases. A comparison with the Levenberg-Marquardt least square optimization showed that the registration accuracy was markedly improved. The deformable approach using total-variation provides a natural and logical solution to model the fetal motions in uterus and greatly improves the accuracy and robustness of deformable registration. Copyright © 2013 American Scientific Publishers.