Phase Constraint Improves Ultrasound Image Quality Reconstructed using Deep Neural Network

Abstract

Plane wave (PW) imaging has become a popular ultrafast ultrasound (US) imaging method thanks to its high frame rate. However, PW imaging requires multiple steered transmissions for coherent compounding to improve the image quality, at the cost of decreased imaging frame rate. In this study, a phase constraint (PC) based loss function is proposed to train a deep neural network (DNN) to reconstruct B-mode images from single PW transmission, with quality equivalent to that of compounding with 75 PWs. To demonstrate the effectiveness of PC, the proposed method (U-Net-PC) was compared to the original U-Net without PC. Our preliminary data suggested that the addition of PC is useful for improving the quality of B-mode image and the accuracy of motion tracking. Results from simulations show that the proposed method achieves 9.21%, 15.84%, 61.55% and 18.80% improvements in root mean square errors (RMSEs) for displacement estimation. In vivo experiments show that the proposed method achieves better visual quality and higher peak signal-to-noise ratio (PSNR) (increased by 17.27%) and structural similarity (SSIM) in B-mode image (increased by 20.35%) than original U-Net without PC. These results suggest the superiority of the proposed PC-based loss over original loss function.