3D Intra-cardiac flow estimation using speckle tracking: A feasibility study in synthetic ultrasound data

In recent years, echo particle image velocimetry (EPIV) has been proposed to characterize intra-cardiac flow fields noninvasively. To date, these methods remain two dimensional (2D). As volumetric cardiac ultrasound has become more readily available and as this may avoid some of the problems inherent to 2D imaging (i.e. out-of-plane flow), the aim of this study was to test whether EPIV could be extended to three dimensions using synthetic data sets. A typical inflow velocity profile and a dynamic LV volume change profile combined with a three-dimensional (3D) anatomical model of the LV were used to build a realistic computational fluid dynamics (CFD) flow vector field (Fluent 12.1, ANSYS). 3D B-mode image sequences were acquired using a fast ultrasound simulator (COLE) of virtual contrast bubbles moving along this 3D flow field (frequency = 4.5MHz; sampling frequency = 50MHz; 50×50 degrees opening angle; frame rate = 113 Hz; image size 120×60×60mm). Inter-frame motion during diastasis was estimated by volumetric speckle tracking (kernel 0.65×6.70×6.70mm; search region 4.96×9.71×8.85mm; overlap 40%×90%×90%) using 3D normalized cross-correlation as a similarity metric. Cubic spline interpolation was embedded for subsample motion estimation prior to median filtering (6.34×4.73×4.75mm) of the resulting velocity vector field. Regression and correlation coefficients as well as the estimate error of the tracking velocity field with the CFD reference ground truth were calculated. The relative point-wise velocity vector error averaged over the period of E wave, A wave and systole were 14.66 ± 2.66%, 12.61 ± 1.15%, 14.00 ± 7.60% respectively. The flow field obtained through speckle tracking agreed well with the ones measured from CFD when the velocity amplitude was below 0.35 m/s. However, for larger velocities, the regional motion was underestimated.