Self-Supervised Learning of a Deep Neural Network for Ultrafast Ultrasound Imaging as an Inverse Problem

Compared with conventional delay-and-sum (DAS) beamforming method, sparse regularization (SR) techniques can reconstruct higher quality ultrasound (US) images from plane wave (PW) transmission. However, these methods suffer from long reconstruction time, owing to the high computational complexity inherently associated with its iterative process. In this work, a deep neural network (DNN) is trained to reconstruct plane wave ultrasound (PWUS) images from RF channel data with significantly reduced computational time. To overcome the lack of gold standard of ideal US images, a self-supervised learning scheme is employed to utilize the RF channel data as both the inputs and the labels during the training process. Using simulation data, the proposed method achieves comparable spatial resolution and 3.2-dB higher CNR, compared with the SR method.