Blind Phase-aberrated Baseband Point Spread Function Estimation Using Complex-valued Convolutional Neural Network

Many methods used to improve clinical ultrasound image quality, e.g. deconvolution, require precise estimation of point spread function (PSF). However, the PSF cannot be well estimated even with prior knowledge of the system setting because the unknown property of inhomogeneous sound velocity in human tissue leads to phase-aberrated PSF. In addition, most image quality improving techniques are performed over beamformed baseband data (i.e., IQ data) and most portable ultrasound systems only allows the access of beamformed baseband data because of limited data transfer bandwidth. Thus, blind phase-aberrated PSF estimation directly from the beamformed baseband data is beneficial for portable ultrasound to leverage these image quality improving techniques. For this purpose, we introduce a novel complex-valued convolutional neural network (CNN) based blind estimator of phase-aberrated PSF using beamformed baseband data. Simulation results show that the proposed complex-valued U-Net estimator produces an aberrated PSF with higher similarity to the ground truth PSF.