基于Transformer和CNN的低剂量CT图像去噪网络  

Low-dose CT Image Denoising Network Based on Transformer and CNN

作　　者：郝文强 崔学英[1] 郭映亭 HAO Wenqiang;CUI Xueying;GUO Yingting(School of Applied Sciences,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学应用科学学院,山西太原030024

出　　处：《海南师范大学学报（自然科学版）》2023年第2期176-182,共7页Journal of Hainan Normal University（Natural Science）

基　　金：国家自然科学基金项目(62001321);山西省自然科学基金项目(202103021224274)。

摘　　要：低剂量计算机断层扫描(Low-dose Computed Tomography, LDCT)在临床中有着广泛的应用,可以有效减轻对病人的辐射剂量。但是成像后的低剂量CT图像中含有明显的噪声和条形伪影,影响医师的诊断。提出了一种基于Transformer和CNN的去噪网络,该网络是一种改进的编解码网络架构,其编码端的每一层由卷积模块与Transformer模块融合而成,用来提取每一层的局部特征和全局特征,同时引入融合模块用来有效地融合提取的局部特征和全局特征。并把融合后的特征通过跳跃连接融入解码端对应的层,解码端的每一层通过卷积模块提取有效特征进而重建去噪后的图像。在真实数据集Mayo上的实验结果说明所提出的网络不仅可以有效去除噪声,还能够保持图像的边缘。Low-dose Computed Tomography(LDCT)is widely used in clinic,which can reduce the radiation dose of patients effectively.However,the low-dose CT images often contain obvious noise and streak artifacts,which affect the diagnosis of doctors.This paper proposes a denoising network based on Transformer and CNN,which is an improved codec network.Each layer in encoder includes the convolutional module and Transformer module,which are used to extract local and global features of each layer.At the same time,a fusion module is introduced to integrate the extracted local features and global features.The fused features are concatenated into the corresponding layer of the decoding terminal through skip connection.Each layer in the decoder extracts effective features through the convolutional module.The denoised image are reconstructed through the final convolutional layer.Experimental results on the real data set Mayo show that the proposed denoising network can not only effectively remove noise,but also maintain the edge of the image.

关 键 词：低剂量CT 图像去噪 U-Net TRANSFORMER 通道注意力 

分 类 号：O157.5[理学—数学]