基于卷积神经网络的超声造影图像去噪方法

针对超声造影图像包含大量噪声的问题, 提出一种基于卷积神经网络的超声图像去噪方法. 首先, 通过图像平移、 翻转、 旋转等数据增强方法扩充稀缺的超声造影图像样本数量； 其次, 通过重叠切割小图像块, 进一步扩充样本数量； 最后, 以图像块和人工噪声为输入训练集, 训练基于卷积网络结构的去噪模型. 实验结果表明, 该方法可有效扩展至不同大小的超声造影图像, 对于超声造影图像去噪后的峰值信噪比高于传统的图像去噪方法.

Denoisng Method of Contrast-Enhanced Ultrasound Image Based on Convolutional Neural Networks

Aiming at the noise problem of contrast-enhanced ultrasound (CEUS) image, we proposed a denoisng method of contrast-enhanced ultrasound image based on convolutional neural networks. Firstly, the number of rare contrast-enhanced ultrasound image samples was expanded by data enhancement methods such as image translation, image flipping and image rotation. Secondly, the number of samples was further expanded by overlapping small image blocks. Finally, image blocks and artificial noise were used as input training sets to train the denoising model based on convolutional network structure. The experimental results show that the proposed method can be effectively extended to different sizes of CEUS images, and the peak signal-to-noise ratio of CEUS images after denoising is higher than that of traditional image denoising methods.