Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (1): 252-258.DOI: 10.3778/j.issn.1002-8331.2107-0504

• Graphics and Image Processing • Previous Articles     Next Articles

LDCT Image Denoising Based on Edge Protection and Multi-Stage Network

GUO Zhitao, ZHOU Feng, ZHAO Linlin, YUAN Jinli, LU Chenggang   

  1. School of Electronic Information Engineering, Hebei University of Technology, Tianjin 300401, China
  • Online:2023-01-01 Published:2023-01-01

边缘保护与多阶段网络相结合的LDCT图像去噪

郭志涛,周峰,赵琳琳,袁金丽,卢成钢   

  1. 河北工业大学 电子信息工程学院,天津 300401

Abstract: Low dose CT is an effective method to reduce radiation and the impact on patients. However, reducing the radiation dose will lead to a lot of noise in the reconstructed image, which will affect the judgment of doctors in the process of diagnosis. To solve the problem, this paper constructs a multi-stage network denoising model and introduces an edge protection module composed of Sobel convolution to enhance the protection of image edge details. In addition, in order to comprehensively consider the influence of the output of each stage in the training process, Charbonnier loss is introduced in each stage and superimposed it as a loss function. Experimental results show that compared with the comparison algorithm, the proposed algorithm achieves the best denoising results, protects the image edge details, and achieves excellent performance in prediction time.

Key words: image processing, low dose CT denosing, deep learning, Sobel convolution

摘要: 低剂量CT是减少辐射剂量,降低对患者影响的有效方法。然而降低辐射剂量会导致重建之后的成像存在大量的噪声,并且会造成低剂量CT图像细节的丢失,从而影响医师在诊断过程中的判断。针对这个问题,构建了一种多阶段去噪网络模型,并在此基础上引入了由可训练的Sobel卷积构成的边缘保护模块来增强对图像边缘细节的保护。此外,为了综合考虑每个阶段的输出在训练过程中的影响,在每个阶段引入Charbonnier loss并将其叠加作为损失函数。实验结果表明,所提算法相较于对比算法取得了最优的去噪结果,并实现了对图像边缘细节的保护,同时在预测时间上也取得了优秀的表现。

关键词: 图像处理, 低剂量CT去噪, 深度学习, Sobel卷积