2D RF-based non-rigid image registration for cardiac motion estimation: Comparison against block matching

Cardiac strain can be estimated from echocardiographic images by motion estimation techniques such as block matching (BM) or non-rigid image registration (NRIR). In case of small motions, it has already been shown that RF-based motion estimation is more accurate compared to B-mode tracking due to the presence of high frequency signals. In a feasibility study, we previously demonstrated the performance of RF-based NRIR in case of a 1D uniaxial compression. The primary goal of this study was to extend the application of RF-based NRIR to 2D motion estimation and perform a comparison with the RF-based BM method. Given the higher amount of RF data to be processed as compared to B-mode tracking, we also demonstrate the performance of a more computationally efficient analytic regularizer to speed up the registration time. To validate this, an in-silico phantom of size 78 mm × 56 mm (axial × lateral) was created and an inter-frame lateral compression of 0.6% and 1% was applied in Cartesian space. The phantom was scanned with a 3.4 MHz phased array with an opening angle of 90° to generate 2D synthetic RF images (7193 samples × 200 lines) for 10 different realizations. The tracking was performed for a set of points in RF space for both methods and strains were computed in Cartesian space. For a lateral compression of 0.6% and 1% the mean estimated lateral strain was more accurate for NRIR vs. BM (0.63 % vs. 0.48 %) and (1.03 % vs. 0.87%) respectively. The eSNR (mean strain over its standard deviation) was also computed to assess tracking performance. The comparison of the mean lateral eSNRs between NRIR vs. BM, for 0.6% (3.68 vs. 2.02) and 1% (6.04 vs. 3.10), indicates that less noisy lateral strain estimates were obtained for NRIR.