Deep learning based minimum variance beamforming for ultrasound imaging

Abstract

Deep learning has been applied to ultrasound imaging recently, and it needs to be further studied to improve ultrasound beamforming methods. According to the latest research, deep neural network was able to suppress off-axis scattering signals in ultrasound channel data, which enhanced the performance of beamforming and improved the contrast of the output ultrasound images. Minimum variance beamforming was capable to present high lateral resolution, but lacked of high image contrast of ultrasound images. In order to effectively improve the contrast of minimum variance beamforming, this work investigated the combination of deep neural network and minimum variance beamforming. In the experiments, the simulated point target and cyst scenarios were adopted to evaluate the performance of the proposed methods. The results demonstrated that combining deep neural network and minimum variance beamforming can effectively reduce the side lobe level and thus can improve the contrast of the ultrasound images while maintaining the lateral resolution performance.