基于Transformer增强型U-net的CT图像稀疏重建与伪影抑制  

Sparse Reconstruction of Computed Tomography Images with Transformer Enhanced U-net

作　　者：樊雪林 文昱齐 乔志伟[1] FAN Xuelin;WEN Yuqi;QIAO Zhiwei(School of Computer and Information Technology,Shanxi University,Taiyuan 030006,China;School of Materials Science&Engineering,Beijing Institute of Technology,Beijing 102401,China)

机构地区：[1]山西大学计算机与信息技术学院,太原030006 [2]北京理工大学材料学院,北京102401

出　　处：《CT理论与应用研究（中英文）》2024年第1期1-12,共12页Computerized Tomography Theory and Applications

基　　金：国家自然科学基金面上项目(模型与数据耦合驱动的快速四维EPRI肿瘤氧成像(62071281));中央引导地方科技发展资金项目(新型TV和学习先验联合约束的快速四维EPRI成像方法(YDZJSX2021A003));山西省回国留学人员科研资助项目(基于新型四维TV正则机理的快速EPRI肿瘤氧成像方法研究(2020-008))。

摘　　要：实现低剂量计算机断层成像(CT)的一个有效办法是减少投影角度,但投影角度较少会产生严重的条状伪影,降低图像的临床使用价值。针对该问题,提出一种耦合卷积神经网络(CNN)和多种注意力机制的U型网络(TE-unet)。首先采用U型架构提取多尺度特征信息;其次提出一个包含CNN和多种注意力的模块提取图像特征;最后在跳跃连接处加入Transformer块过滤信息,抑制不相关特征,突出重要特征。所提网络结合CNN的局部特征提取能力和Transformer的全局信息捕获能力,辅以多种注意力机制,实现了良好的去条状伪影能力。在60个投影角度下,与经典的Uformer网络相比,峰值信噪比(PSNR)高出0.3178 dB,结构相似度(SSIM)高出0.002,均方根误差(RMSE)降低0.0005。实验结果表明,所提TE-unet重建的图像精度更高,图像细节保留的更好,可以更好地压制条状伪影。An effective way to achieve low-dose computed tomography(CT)is to reduce the projection angle while maintaining the same radiation dose at each angle.However,a fewer projection angle can result in severe strip artifacts,reducing the practicality and clinical value of the image.To address this issue,a U-shaped network(Transformer Enhanced U-net,TE-unet)coupled with convolutional neural network(CNN)and multiple attention mechanisms was proposed.Firstly,a U-shaped architecture was adopted to fuse multi-scale feature information;Secondly,a module that includes CNN and multiple types of attention was proposed to extract image features;Finally,transformer blocks were added at skip connections to filter information,suppress irrelevant features,and highlight important features.This network combines the local feature extraction ability of CNN and the global information capture ability of Transformer,supplemented by various attention mechanisms,to achieve good ability to remove stripe artifacts.At 60 projection angles,compared to the classic uformer network,peak signal to noise ratio(PSNR)is 0.3178 dB higher,Structural Similarity(SSIM)is 0.002 higher,and Root Mean Square Error(RMSE)is 0.0005 lower.The experimental results show that the proposed TE-unet network reconstructs images with higher accuracy,preserves better image details,and can better suppress strip artifacts.

关 键 词：稀疏重建 计算机断层成像 TRANSFORMER 多注意力机制 条状伪影 

分 类 号：O242[理学—计算数学] TP391[理学—数学]