Golay-Net: Deep learning-based Golay coded excitation for ultrasound imaging

Abstract

Ultrasound imaging modality, which operates by transmitting and receiving short ultrasound pulses, offers a promising approach for real-time, high-resolution diagnostic imaging at relatively low cost. However, the conventional short-pulse approach is inherently limited by signal attenuation with increased imaging depth, leading to reduced penetration and a lower signal-to-noise ratio (SNR), which ultimately degrades diagnostic performance. Golay-coded excitation has been introduced to mitigate these issues by transmitting longer, coded pulses that use a pair of complementary sequences (Codes A and B) to enhance SNR and imaging depth. However, this technique requires two sequential transmissions to acquire two echoes related to the complementary codes, inevitably reducing the frame rate by half. In this work, we propose a novel deep learning framework that overcomes this limitation by generating the echo signal corresponding to Code B from the echo signal obtained after transmitting code A. For this, we developed Golay-Net, based on a 1-D U-Net architecture, which changes the phase of the range sidelobes of the Code A-related echo signals, thereby effectively synthesizing the echo signals that would have been obtained using Code B. In vitro and in vivo experiments demonstrate that the proposed Golay-Net can synthesize code B-related echo signals with high fidelity, enabling the reconstruction of ultrasound images with enhanced SNR and imaging depth, without compromising frame rate.