Enhanced Radon Domain Beamforming Using Deep-Learning-Based Plane Wave Compounding

In recent years, ultrafast ultrasound imaging has received a lot of attention. However, ultrafast imaging requires large data transfers in short periods of time. Therefore, methods to reduce this data load, while maintaining image quality, are of crucial importance. In the present study, a neural net (NN) is developed that processes ultrasound data in the Radon domain (RD). By using RD data as input, the NN infers an RD pixel-wise weight mask. As such, the NN makes an informed decision on which values it negates to enhance images. The NN is trained to approximate an image of 51 compounded plane waves (PWs) from a 3 PW input. This study shows that the proposed method can match the gCNR of a 51 PW compounded image, using only 3 PWs. This method can be employed in ultrasound systems to reduce data transfer rates in ultrafast imaging and enhance image quality.