B-spline explicit active tracking of surfaces (BEATS): Application to real-time 3D segmentation and tracking of the left ventricle in 3D echocardiography

We recently proposed a generic framework for realtime 3D segmentation (BEAS) and we demonstrated that it allows real-time segmentation of volumetric cardiac US sequences. Nevertheless, typical artifacts in RT3DE, such as lateral wall dropout and near-field artifacts still pose some challenges. The aim of the present study was to develop a real-time 3D LV automatic segmentation/tracking framework able to robustly deal with the aforementioned artifacts and to assess its performance in the extraction of volumetric functional parameters used in clinical routine. To this end, we propose a geometrical shape regularizer, penalizing the local curvature variation, and a contrast-dependent motion regularizer term based on the global motion of the LV which is applied only to areas of low local contrast thereby enhancing the tracking of drop-out regions. The proposed regularization terms contribute to a significant increase in tracking performance. Furthermore, the proposed algorithm (BEATS) is able to accurately extract relevant cardiac (functional) indexes in a clinical setting, fully automatically and in real-time.