Robust recovery of volumetric cardiac motion with physically constrained H∞ filtering

We present a robust methodology for volumetric heart motion recovery from medical image sequences, where mechanically motivated spatial constraints of myocardium is combined with the temporal filtering of the H^∞ strategy. Cardiac system dynamics is transformed into state space representation, and the heart kinematics are then estimated using the min-max criterion. Unlike the popular Kalman filtering paradigm which assumes known Gaussian statistics of the process and measurement noises, our effort more realistically assumes no prior knowledge on the noise statistics but only noise energy levels, and it employs a minimum worst-case error estimation philosophy. Experiments with two types of magnetic resonance images, the tagging and the phase contrast data, are conducted with physiologically sensible results.