Enhancing Ultrasound Imaging with Synthetic Aperture Data: A Deep Learning Approach for Superior Resolution and Contrast

Deep learning methods have been widely used in the field of ultrasound (US) imaging to improve lateral spatial resolution and contrast resolution of ultrasound images. However, current approaches have mainly focused on using traditional focused-wave (FW) and plane-wave (PW) ultrasound data for the training of deep neural networks. Although synthetic aperture ultrasound (SAU) data can enable high-quality ultrasound imaging with higher lateral spatial resolution and contrast resolution, they are rarely used in the study of deep learning-based ultrasound image enhancement algorithms. In this paper, we propose a unified framework that leverages SAU data for US image enhancement. First, we introduce a new multi-quality data generation approach that leverages SAU channel data to produce pairs of low-high-quality US images for network training. Subsequently, the obtained pairs of US image data are used to train efficient deep-learning models for various US image enhancement tasks. Experimental results demonstrate that deep learning models trained with the generated pairs of US image samples can greatly improve the lateral and contrast resolution. That is, our proposed approach achieves satisfactory performance on various ultrasound image enhancement tasks.