A delay-and-sum beamformer for 3D imaging using row-column arrays and written in CUDA is presented and compared to an existing similar GPU-based beamformer written in the MATLAB programming language. Data from a 192+192 rowcolumn array single element emission sequence is simulated and beamformed. The two beamformers’ performance is evaluated in two synthetic aperture setups comprised of 1) two orthogonal planes and 2) a full volume on three different NVIDIA GPUs: a 1050 Ti, a 1080 Ti, and a TITAN V. The execution time and the sample throughput (samples beamformed per second) are reported. The CUDA beamformer performs consistently better than the MATLAB beamformer with speed-ups ranging from 1.9 to 64.6 times, and the worst-case throughput of the CUDA beamformer exceeds the best-case of the MATLAB beamformer. High-resolution images of crossing planes can be beamformed at up to 13 Hz, while a 50-by-50-by-20 cubic-millimeter highresolution volume sampled at one quarter of a millimeter is beamformed in 3 seconds.
|Conference||2019 IEEE International Ultrasonics Symposium|
|Period||06/10/2019 → 09/10/2019|
Stuart, M. B., Jensen, P. M.
, Olsen, J. T. R., Kristensen, A. B., Schou, M., Dammann, B., Sørensen, H. H. B., & Jensen, J. A.
(2019). Fast GPU-beamforming of\\Row-Column Addressed Probe Data
. In Proceedings of 2019 IEEE International Ultrasonics Symposium
(pp. 1497-1500). IEEE. https://doi.org/10.1109/ultsym.2019.8925802