基于跨尺度边缘增强深度卷积神经网络的低剂量CT图像去噪  

Low-dose CT image denoising based on cross-scale edge enhanced deep convolutional neural networks

作　　者：王同罕 吴通 贾惠珍 李沛钊 谢婷 舒华忠[3] Wang Tonghan;Wu Tong;Jia Huizhen;Li Peizhao;Xie Ting;Shu Huazhong(School of Information Engineering,East China University of Technology,Nanchang 330013,China;College of Engineering and Design,Hunan Normal University,Changsha 410081,China;Laboratory of Image Science and Technology,Southeast University,Nanjing 210096,China)

机构地区：[1]东华理工大学信息工程学院,南昌330013 [2]湖南师范大学工程与设计学院,长沙410081 [3]东南大学影像科学与技术实验室,南京210096

出　　处：《东南大学学报（自然科学版）》2023年第2期363-369,共7页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(62261001,62266001);江西省教育厅科学技术研究资助项目(GJJ200746);东华理工大学江西省放射性地学大数据技术工程实验室开放基金资助项目(JELRGBDT202001)。

摘　　要：为了提高去噪网络的可解释性,将传统滤波算子的优势融入到网络设计中,提出了基于跨尺度边缘增强的深度卷积神经网络(CEDCNN).将传统边缘算子与卷积相结合,设计出轻量化的边缘增强模块,增强边缘信息对网络结果的影响.基于自适应一致性先验算法构建深度迭代网络,进一步提取边缘增强特征,从而实现端到端可训练、可解释的深度去噪网络.将均方误差和多尺度注意残差感知损失相结合,解决了重建图像过平滑的问题.实验结果表明,CEDCNN去噪网络的PSNR、RMSE、SSIM评价指标分别为43.647 5 dB、0.006 8、0.987 5,说明该方法可显著提高去噪后的图像质量,有效保证低剂量X射线下CT图像的成像质量和精度,去噪效果与正常剂量CT图像所展现的人体组织细节相当.To improve the interpretability of denoising network,the advantages of traditional filtering operators were incorporated into network design,and a cross-scale edge-enhanced deep convolutional neural network(CEDCNN)was proposed.The traditional edge operator was combined with convolution to design a lightweight edge enhancement module,which enhanced the influence of edge information on the network results.A deep iterative network was constructed based on an adaptive consistency prior algorithm,and edge enhancement features were further extracted to achieve an end-to-end trainable and interpretable deep denoising network.The composite loss combining mean square error(MSE)and multi-scale attention residual perception were combined to solve the problem that the reconstructed images were easily over-smoothed.The experimental results show that the peak signal to noise ratio(PSNR),the root mean square error(RMSE),the structural similarity(SSIM)of the CEDCNN denoising network are 43.6475 dB,0.0068,0.9875,respectively.The image quality after denoising is significantly improved.The imaging quality and accuracy of the CT images can be effectively guaranteed at low dose of X-ray.The denoising results are comparable to the details of human tissues shown in normal dose CT image.

关 键 词：低剂量CT图像 深度学习 跨尺度边缘增强 医学图像去噪 自适应一致性先验 

分 类 号：TN911.73[电子电信—通信与信息系统]