Estimating and resolving uncertainty in cardiac respiratory motion modelling

We present a novel method for cardiac respiratory motion estimation in image-guided interventions. The technique combines a preprocedure affine motion model and intra-procedure real-time images to estimate and correct for the respiratory motion of the heart. As well as making motion estimates, the model is able to quantify the uncertainty in these estimates. This uncertainty is resolved using a Bayesian approach based on a prior probability from the motion model and a likelihood term derived from the intraprocedure images. The proposed method is validated using MR-derived motion fields and simulated 3D real-time echocardiography data for 4 volunteers and compared to 3 other motion estimation techniques. Our Bayesian approach shows improvements in respiratory motion estimation for each volunteer of 7.0%, 4.6%, 7.7% and 5.3% respectively, compared to the use of a motion model only.