TY - GEN
T1 - A Dual-Channel Network Based GAN for Low-Dose CT Image Denoising
AU - Zhao, Yuanyuan
AU - Guo, Shuli
AU - Han, Lina
AU - Baris Cekderi, Anil
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - Denoising of low-dose computed tomography (LDCT)images is a hot issue in the field of medical images. In this paper, a dual-channel denoising network based on GAN architecture is proposed to solve the key problems of image blur and loss of detail structure in the process of LDCT denoising. In the generator part, a content learning channel and a noise learning channel are used to learn images at the same time. In the content learning channel, an improved U-Net network is used to learn the image content to generate a normal dose computed tomography (NDCT) image. In the noise learning channel, firstly, wavelet transform is used to obtain high-frequency components, and then the improved DnCNN network is used to learn the noise. Finally, the NDCT is obtained. Through the fusion layer, the results of the two are fused to obtain a better image. Through the ablation implementation and compared with the conventional methods, the method suggested in this research has a better denoising effect and a certain suppression effect on the blurring of detail structures.
AB - Denoising of low-dose computed tomography (LDCT)images is a hot issue in the field of medical images. In this paper, a dual-channel denoising network based on GAN architecture is proposed to solve the key problems of image blur and loss of detail structure in the process of LDCT denoising. In the generator part, a content learning channel and a noise learning channel are used to learn images at the same time. In the content learning channel, an improved U-Net network is used to learn the image content to generate a normal dose computed tomography (NDCT) image. In the noise learning channel, firstly, wavelet transform is used to obtain high-frequency components, and then the improved DnCNN network is used to learn the noise. Finally, the NDCT is obtained. Through the fusion layer, the results of the two are fused to obtain a better image. Through the ablation implementation and compared with the conventional methods, the method suggested in this research has a better denoising effect and a certain suppression effect on the blurring of detail structures.
KW - generative adversarial networks
KW - image denoising
KW - low-dose computed tomography
KW - noise learning
UR - http://www.scopus.com/inward/record.url?scp=85151760400&partnerID=8YFLogxK
U2 - 10.1109/CAC57257.2022.10054909
DO - 10.1109/CAC57257.2022.10054909
M3 - Conference contribution
AN - SCOPUS:85151760400
T3 - Proceedings - 2022 Chinese Automation Congress, CAC 2022
SP - 2943
EP - 2948
BT - Proceedings - 2022 Chinese Automation Congress, CAC 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 Chinese Automation Congress, CAC 2022
Y2 - 25 November 2022 through 27 November 2022
ER -
