CUDA-GPU Implementation Strategies for non-linear Beamforming for Synthetic Transmit Aperture Technique: Initial Results

Synthetic transmit aperture (STA) ultrasound imaging offers dynamic focusing during both transmission and reception, which provides high-resolution images. However, its implementation demands high data processing capability. Nonlinear beamforming techniques have been developed to improve ultrasound image quality; these methods make use of signal correlation to increase image contrast and resolution. Here again, the computational complexity in implementation poses a huge challenge for applicability to real-time applications. Numerous works utilized GPU programming to address large computations in advanced beamforming and STA imaging. This paper investigates the GPU implementation strategies of STA imaging with non-linear beamforming and suggests a method for realistic imaging with improved image quality. In particular, we studied approaches that utilized GPU tensor technology, beamforming methods, and beamforming grid sizes. To evaluate different strategies, experimental radiofrequency (RF) data obtained from a CIRS general purpose phantom was used. The results showed that the filtered delay multiply and sum (FDMAS) provides better quality images than delay and sum (DAS) beamforming for all combinations. Moreover, the results suggest that the 1024x128 grid size provided better-quality images at an acceptable beamforming time.