Adaptive myocardial border tracking in M-mode echocardiograms

Abstract This paper describes a sequential approach for the detection of myocardial borders in M-mode echocardiograms, which takes into account physiological knowledge. Initial border estimates are obtained from the output of a cross-correlation detector. These initial observations are improved by combining the extraction of the border pairs associated with the same myocardial wall. A lattice from joint process predictor whose parameters are estimated adaptively is used for this purpose. In addition, a physiological constraint expressed as a linear relationship between the ventricular diameter and the posterior wall thickness is used to improve the detection of the endocardium during systole. Finally, the correlation templates are updated recursively using the intensity profiles surrounding the final border positions. This is accomplished using a new adaptive algorithm that is optimal in the least squares sense. This procedure turns out to be quite efficient in tracking the temporal variations of the myocardial patterns and improves the continuity of the trajectories. The present method is compared with previous approaches and examples are identified for which it improves border detection significantly.