High Resolution Ultrasound Image Reconstruction for Plane-Wave Imaging Using Deep Learning

The plane-wave imaging (PWI) has attracted researchers' attention owing to its ability to obtain high frame rate ultrasound imaging, which is crucial for cardiac applications. However, degraded image quality remains a drawback in PWI. Coherent plane-wave compounding (CPWC) using multiple plane-waves steered at different angles can be employed but it compensates the framerate. Thus, an imaging method that can provide both high image quality and frame rate is required. In this study, we propose a method to reconstruct high-resolution ultrasound images from plane-wave raw channel data (pre-beamform RF data) by adapting the modified U-Net as the deep learning network architecture. The training and test data consist of the combination of simulation and experimental data conducted on phantom and in-vivo study. The performance evaluation result shows improvement in terms of the structure of similarity, peak signal to noise ratio, and full width at half maximum compared to the conventional delay and sum (DAS) method. In conclusion, the proposed ultrasound image reconstruction method using the deep learning beamformer is capable of reconstructing a high-resolution image of the carotid artery for a single plane-wave ultrasound imaging.