一种基于双路径卷积神经网络的超声图像去噪方法

A Dual-Path Carotid Artery Ultrasound Image Denoising Method Based on Deep Learning

针对传统医学超声图像去噪方法的不足,本文提出了一种双路径卷积神经网络去噪方法。首先使用注意力机制网络来搭建噪声估计子网络用于训练噪声图像,然后将这个噪声图像与输入图像进行结合,送入一个类Unet网络并获得去噪后的图像。实验表明,在选取了最优网络结构后,该方法的去噪性能能够明显优于传统去噪方法,且相比于其他基于深度学习的去噪方法,该方法能在有效的去除超声图像斑点噪声的同时更好地保留了图像的细节信息。

Aiming at the shortcomings of traditional medical ultrasound image denoising methods,this paper proposes a dual-path convolutional neural network denoising method.First,use the attention mechanism network to build a noise estimation sub-network for training noise images,and then combine this noise image with the input image,send it to a Unet-like network and obtain the denoised image.Experiments show that after selecting the optimal network structure,the denoising performance of this method can be significantly better than traditional denoising methods,and compared with other denoising methods based on deep learning,this method can effectively remove ultrasonic image speckles.While noise,the details of the image are better preserved.