RxNet: Learning for Receive Element Reduction in Synthetic Transmit Aperture Imaging

Ultrasound beamforming with reduced receive channel count is necessary when the system contains fewer channels than the elements in the probe to be controlled. While synthetic transmit aperture (STA) imaging with two-way dynamic focusing can improve image quality, it requires sequential full-aperture receives. This paper proposes a self-supervised network to adaptively select a group of optimized receive elements from a limited number of channels in each STA transmit-receive event. The selection procedure is achieved by modifying a fully-connected encoder to characterize the relationship between the full-aperture channel data and their sub-sampled components. The decoder of this network can also recover the full-aperture channel data from the subsamples. The network trained by phantom data can be utilized to evaluate the data from computer simulations and in-vivo experiments. Compared with a compressed-sensing based method, the proposed method achieves more accurate image reconstruction and better image contrast performance.