Automatic mitral valve leaflet tracking in Echocardiography via constrained outlier pursuit and region-scalable active contours

Tracking the mitral valve leaflet in Echocardiography is of crucial importance to the better understanding of various cardiac diseases and is very helpful to assist the surgical intervention for mitral valve repair. In this paper, we present an automatic mitral leaflet motion tracking approach, which consists of two phases: constrained outlier pursuit for mitral leaflet detection and its shape refinement. In the former phase, we first learn a low-rank subspace which can gradually change over time to model the background sequence, and simultaneously detect sparse outliers through such low-rank representation. Then, we extract the supported states of the myocardial tissues to constrain the outlier pursuit for mitral leaflet detection, featuring on reliably removing the irrelevant outliers. In the latter phase, we further present a region-scalable active contour to refine the shapes of the detected mitral leaflet for final tracking. The proposed approach does not require any user-specified interactive information or pre-collected training data for learning. The robustness of its performance has been demonstrated against the fast mitral leaflet motions, shape deformation and unstable myocardial tissue appearance. Experimental results have shown that the proposed approach performs favorably on four challenging sequences in comparison with the state-of-the-art methods.