Beamspace-domain learning of minimum variance beamformer with fully convolutional network

In medical ultrasound systems, receiving beamforming is necessary to produce an ultrasonic image. Although minimum variance (MV) beamforming was developed to achieve higher image quality than commonly used delay-and-sum (DAS) beamforming, it is computationally expensive. Therefore, in this study, we investigated how to convert the beamforming profile of DAS to that of MV using deep learning. The results showed that a fully convolutional network could produce an image with comparable quality to that in MV beamforming in a shorter time than the conventional MV beamformer.