(A) deep convolutional neural network for Rx-subsampled B-mode ultrasound imaging reconstruction

Abstract

There is an increasing demand for applications of B-mode ultrasound imaging, such as ultra-fast ultrasound or portable ultrasound. In order to meet such this demand, the technique of reconstructing high quality images using a limited number of RF data is needed. The existing methods use hardware changes of the ultrasound device or algorithms having the high complexity and large computations. Therefore, there are hardware limitations that it can not be applied to other ultrasound devices and software limitations that reconstruction time is long. To overcome these technical limitations, we propose the method of Rx-subsampling and learning receiver(Rx)-scanline(SC) 2-D data planes with a deep convolutional neural network. Because Rx-SC 2-D data planes represent the redundancy, the proposed method can interpolate Rx-subsampled Rx-SC 2-D data planes. In this paper, the data acquired directly by using the ultrasound device is learned by the deep convolutional neural network, and images with high quality are reconstructed at a high speed. In addition, by applying the same neural network irrespective of the transducers or the scanned regions, we show that the proposed method has the universality.