Motion Tracking by Transverse Oscillations in 3D Cardiac Ultrasound Imaging
MetadataShow full item record
Quantitative measurements of myocardial deformation has proved to be useful in assessment of cardiovascular diseases. Therefore, estimation of the motion of cardiac tissue has been an active field of research for many years. Tissue Doppler imaging is a technique widely used in clinical practice to estimate tissue motion. However, this method is only able to estimate the 1D motion along the beam axis. Several methods for 2D and 3D motion estimation of tissue have been proposed. In this thesis, four procedures for 3D motion estimation of tissue have been developed in Matlab and compared in terms of accuracy and computational efficiency. Three of the estimators are based on transverse oscillations to estimate the 3D motion, and exploits three different approaches for filtering the spatial frequency content of the images. The fourth method is a speckle tracking approach based on block matching. The first filtering approach is based on work done by Jensen's research group, and filters the spatial frequency content to introduce transverse oscillations in each transverse direction separately. The second method is based on filtering the spatial frequency content to keep four separate ranges of frequency in order to introduce transverse oscillations in both transverse directions simultaneously. The third filtering approach is a novel approach that proposes to filter the spatial frequency content to keep only three separate ranges of frequency. Phantom experiments with known motion either along a straight line or along a circular path were used for validation of the four methods. The motion estimation approaches were also tested on an in vivo example from the heart of a healthy volunteer. The transverse oscillation method based on filtering the spatial frequency content of the ultrasound images to keep three separate ranges of frequency, yielded the most accurate estimates of the z- and y-component of the circular velocity, with mean errors of 0.0062 mm/frame ± 0.0031 mm/frame and 0.0107 mm/frame ± 0.0085 mm/frame, respectively. The transverse oscillations approach based on the work done by Jensen's research group, yielded the most accurate estimate of the x-component of the circular motion with mean error of 0.0032 mm/frame ± 0.0025 mm/frame. On the other hand, block matching yielded significantly poorer estimates in the transverse directions than any of the methods based on transverse oscillations. Motion estimation based on block matching resulted in mean errors of 0.0071 mm/frame ± 0.0036 mm/frame in the z-direction, 0.0494 mm/frame ± 0.0255 mm/frame in the x-direction and 0.0396 mm/frame ± 0.0206 mm/frame in the y-direction. Results from this thesis have been submitted and accepted for oral presentation at the IUS 2017 conference.