Fast and Fully Automatic 3-D Echocardiographic Segmentation Using B-Spline Explicit Active Surfaces: Feasibility Study and Validation in a Clinical Setting

Abstract

A novel framework to efficiently deal with three-dimensional (3-D) segmentation of challenging inhomogeneous data in real-time has been recently introduced by the authors. However, the existing framework still relied on manual initialization, which prevented taking full advantage of the computational speed of the method. In the present article, an automatic initialization scheme adapted to 3-D, echocardiographic data is proposed. Moreover, a novel segmentation functional, which explicitly takes the darker appearance of the blood into account, is also introduced. The resulting automatic segmentation framework provides an efficient, fast and accurate solution for quantification of the main left ventricular volumetric indices used in clinical routine. In practice, the observed computation times are in the order of 1 s.

Real-time 3D interactive segmentation of echocardiographic data through user-based deformation of B-spline explicit active surfaces

“…Although generic data attachment terms can be used, such as the Chan-Vese [30] and Yezzi [31] segmentation functionals and their localized counterparts [32], the segmentation energy can be tailored for more specific applications. For instance, the darker appearance of the blood in RT3DE data can be explicitly taken into account when designing the energy to be minimized during the segmentation of RT3DE data, which improves the robustness of the method in this particular application [25]. Taking this an example, it can be shown that this energy can be directly minimized wrt.…”

“…1. The segmentation algorithm is automatically initialized with an ellipsoid using the algorithm proposed in [25], allowing fully automatic segmentation of the left ventricle. Further details on the foundations of the BEAS framework can be found in [23].…”

“…Starting from the fully automatic segmentation results provided by the algorithm presented in [25], the user is allowed to introduce points in order to manually steer the segmentation where the automatic algorithm failed to capture the correct position of the LV boundary. The update of the segmented surface was achieved through the direct minimization of the energy terms with respect to the B-spline coefficients c[k], yielding:…”

“…In spite of obtaining promising results, the initial semi-automated algorithm relied on manual input (6 clicks per 3D volume) to fit an ellipsoid used to initialize the segmentation algorithm [24]. Since manual initialization accounted for the vast majority (∼95%) of the total analysis time, a fast and automatic initialization method was posteriorly developed, allowing fully automatic segmentation [25].…”

A Combined Multi:scale Deep Learning and Random Forests Approach for Direct Left Ventricular Volumes Estimation in 3D Echocardiography

“…Based on echocardiograph, several different methods have been proposed to estimate the LV volumes [1][2][3][4][5][6][7][8].…”