High-performance Ultrasonic Beamforming Algorithm Based on Deep Learning

In this paper, a new deep neural network (DNN) ultrasonic beamformer was proposed to suppress off-axis scattering and improve image quality. The simulated channel signals from cysts and single point targets were decomposed by wavelet, and then the original signals and the features extracted by wavelet transform were combined into the input of DNN. DNN divided the input data into on-axis signals and off-axis signals, and the off-axis signals were suppressed by the network. The performance of DNN beamformer with parallel input of semantic information and ultrasonic signals was analyzed. According to the experimental results, the proposed DNN beamformer can significantly improve the CNR and CR while maintaining the SNRs.