Recovery of Full Synthetic Transmit Aperture Dataset with Well-preserved Phase Information by Self-supervised Deep Learning

Thanks to two-way dynamic focusing, synthetic transmit aperture (STA) imaging can obtain images with high spatial resolution. However, STA imaging suffers from low signal-to-noise ratio (SNR) and low frame rate. In our previous study, the array elements were spatially encoded with partial Hadamard matrix to transmit plane wave for higher energy and to reduce the number of transmissions for higher frame rate. Thereafter, a compressed-sensing algorithm (CS-STA) was used for the recovery of full STA dataset. However, CS-STA is time-consuming and may destroy the phase information of radio-frequency (RF) signals. In this study, a self-supervised learning (SSL) method was proposed to recover the STA dataset from the encoded PW transmissions, to both accelerate the reconstruction and better preserve the RF phase information for motion estimation. Phantom experiments show that the proposed method achieves similar resolution and improved contrast, compared with STA. Simulation results demonstrate that the proposed method obtains higher performance in lateral strain estimation than STA and CS-STA, in noisy situation. Moreover, the proposed method significantly reduces the reconstruction time from 1 hour (CS-STA) to 10 seconds.