Left ventricular motion estimation under stochastic uncertainties

Reliable estimation of regional cardiac deformation is of great importance for the clinical assessment of myocardial viability. Given partial, noisy, image-derived measurements on the cardiac kinematics, prior works on model-based motion estimation have often adopted deterministic constraining models of mathematical or mechanical nature. In this paper, we present a novel estimation framework for motion analysis under stochastic uncertainties. The main novelty is that the statistical properties of the model parameters, system disturbances, and measurement errors are not treated as constant but rather spatio-temporally varying. An expectation-maximization (EM) framework, in both space and time domains, is used to automatically adjust the model and data related matrices in order to better fit a given measurement data set, and thus provides more accurate tissue motion estimates. Physiologically meaningful displacement fields and strain maps have been obtained from in vivo cardiac magnetic resonance phase contrast image sequences.