U-Net based GAN with deformable convolution for CT image denoising†
Seong Hyun Han1 · Ji Hye Heo2 · Eon Seung Seong3 · Dong Hoon Lim4
1234Department of Information Statistics, Gyeongsang National University
Correspondence to: † This work was supported by the National Research Foundation of Korea(NRF) grant funded by the
Korea government(MSIT) (No. NRF-2022R1A2C1011140).
1 Undergraduate graduate, Department of Information Statistics, Gyeongsang National University, Jinju 52828, Korea.
2 Master’s graduate, Department of Information Statistics, Gyeongsang National University, Jinju 52828, Korea.
3 Undergraduate graduate, Department of Information Statistics, Gyeongsang National University, Jinju 52828, Korea.
4 Professor, Department of Information Statistics and RINS, Gyeongsang National University, Jinju 52828, Korea. E-mail: dhlim@gnu.ac.kr
Korea government(MSIT) (No. NRF-2022R1A2C1011140).
1 Undergraduate graduate, Department of Information Statistics, Gyeongsang National University, Jinju 52828, Korea.
2 Master’s graduate, Department of Information Statistics, Gyeongsang National University, Jinju 52828, Korea.
3 Undergraduate graduate, Department of Information Statistics, Gyeongsang National University, Jinju 52828, Korea.
4 Professor, Department of Information Statistics and RINS, Gyeongsang National University, Jinju 52828, Korea. E-mail: dhlim@gnu.ac.kr
Received May 5, 2023; Revised May 24, 2023; Accepted May 31, 2023.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
- Abstract
- Noise removal in CT images is an important technique that can help image interpretation while reducing radiation exposure. In this paper, we remove noise from CT images using the deep learning GAN (Generative Adversarial Networks) based U-DeCGAN (U-Net Based GAN with Deformable Convolution) model. The GAN model consists of two networks, a generator and a discriminator, and these two are trained in an adversarial competition relationship. The proposed U-DeCGAN model consists of a generator with a U-Net structure and a discriminator with deformable convolution, and the loss function of U-DeCGAN is a weighted sum of the adversarial loss and mean square error loss. To evaluate the performance of the U-DeCGAN model proposed in this paper, CT images damaged by various noises, that is, Gaussian noise, Poisson noise, and speckle noise, The BM3D method, CNN model’s DnCNN model, CDAE, and U-Net GAN model were compared. As a result of performance experiments, the proposed U-DeCGAN model showed good results in quantitative evaluation by PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index Measure) scales as well as qualitative evaluation.
- Keywords : CT image, deformable convolution, denoising, GAN, U-Net architecture
Full Text(PDF) Free
