1 April 2016 Fast myocardial strain estimation from 3D ultrasound through elastic image registration with analytic regularization
Author Affiliations +
Abstract
Image registration techniques using free-form deformation models have shown promising results for 3D myocardial strain estimation from ultrasound. However, the use of this technique has mostly been limited to research institutes due to the high computational demand, which is primarily due to the computational load of the regularization term ensuring spatially smooth cardiac strain estimates. Indeed, this term typically requires evaluating derivatives of the transformation field numerically in each voxel of the image during every iteration of the optimization process. In this paper, we replace this time-consuming step with a closed-form solution directly associated with the transformation field resulting in a speed up factor of ~10-60,000, for a typical 3D B-mode image of 2503 and 5003 voxels, depending upon the size and the parametrization of the transformation field. The performance of the numeric and the analytic solutions was contrasted by computing tracking and strain accuracy on two realistic synthetic 3D cardiac ultrasound sequences, mimicking two ischemic motion patterns. Mean and standard deviation of the displacement errors over the cardiac cycle for the numeric and analytic solutions were 0.68±0.40 mm and 0.75±0.43 mm respectively. Correlations for the radial, longitudinal and circumferential strain components at end-systole were 0.89, 0.83 and 0.95 versus 0.90, 0.88 and 0.92 for the numeric and analytic regularization respectively. The analytic solution matched the performance of the numeric solution as no statistically significant differences (p>0.05) were found when expressed in terms of bias or limits-of-agreement.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Bidisha Chakraborty, Brecht Heyde, Martino Alessandrini, Jan D'hooge, "Fast myocardial strain estimation from 3D ultrasound through elastic image registration with analytic regularization", Proc. SPIE 9790, Medical Imaging 2016: Ultrasonic Imaging and Tomography, 979006 (1 April 2016); doi: 10.1117/12.2216781; https://doi.org/10.1117/12.2216781
PROCEEDINGS
7 PAGES


SHARE
Back to Top