Locally adapted spatio-temporal deformation model for dense motion estimation in periodic cardiac image sequences

Abstract

We recently introduced a continuous state space parametric model of spatio-temporal transformations and an algorithm, based on Kalman filtering, to represent motion in an image sequence describing a periodic phenomena. One advantage of this method is to simultaneously take into account all the sequence frames to robustly estimate the parameters of a unique spatial and periodic-temporal model. However, in 3D+time, a large number of parameters is required. In this paper, we propose a criterion based on motion energy to locally adapt the trajectory model and thus the temporal complexity of the model. The influence of the model order is illustrated on true 2D+time Magnetic Resonance Images (MRI) of the heart in order to motivate the proposed adaptative criteria. Quantitative results of the proposed adapted spatio-temporal motion model are given on synthetic 2D+time MRI sequences. Preliminary experiments show a significant impact notably regarding the parameter saving while preserving the accuracy of the motion estimates. © Springer-Verlag Berlin Heidelberg 2007.