基于噪声估计与引导的低剂量CT图像盲去噪方法

Low-dose CT image blind denoising method based on noise estimation and guidance

提出了一种基于噪声特征估计与引导的低剂量CT盲去噪方法。首先,采用有监督特征学习的方式对输入图像的辐射剂量进行等级评估,并估计出图像潜在的噪声特征图。其次,提出了一种基于噪声引导的低剂量CT图像盲去噪模型,通过显性噪声特征引导的方式将噪声特征与原始图像进行融合,并采用残差编码-解码卷积神经网络实现CT图像噪声去除。实验结果表明,在真实数据集上噪声估计网络及特征融合网络能够大幅提升去噪网络的性能,并且在未知剂量CT图像去噪任务上取得了较好的去噪效果。

A low-dose CT blind denoising method based on noise feature estimation and guidance is proposed.Firstly,the supervised feature learning method is used to evaluate the radiation dose of the input image and estimate the potential noise feature map of the image.Secondly,a low-dose CT image blind denoising model based on noise guidance is proposed.The noise features are fused with the original image by means of explicit noise feature guidance,and a residual encoder-decoder convolutional neural network is used to remove noise from CT images.Experimental results show that the noise estimation network and feature fusion network on real data sets can greatly improve the performance of the denoising network,and achieve better denoising effects in the task of denoising CT images with unknown doses.