A Robust Deep Neural Network Approach for Ultrafast Ultrasound Imaging using Single Angle Plane Wave

Recently, deep learning-based methods have been proposed to reconstruct high-quality images from a single plane wave ultrasound data. A major problem with these methods is that they train the underlying network indiscriminately of the plane wave angle. This poses computational problems during training, as many plane waves at different angles must be mapped to a common ground-truth or reference image. To alleviate this problem, we propose a linear data transformation technique which reduces the intra-data variance among ultrasound Radio Frequency (RF) data at different angles. We further design a convolutional neural network, denoted by 'PWNet' which is trained using the transformed data to learn pixel weights for enhancing the image quality of the single plane wave delay and sum method. The results obtained on the experimental and simulated Plane-wave Imaging Challenge in Medical UltraSound data demonstrate the accuracy of our proposed method which would be beneficial for applications requiring high-quality images reconstructed at higher frame rates.