CT 영상에서 웨이블렛 변환 기반 U-Net GAN 모델을 이용한 잡음제거

초록

CT image denoising is an important technology for protecting patients from radiation and restoring high-quality images. In this paper, we propose a wavelet transformbased U-Net GAN model, or WT-UNetGAN model, to remove noise from CT images. The proposed WT-UNetGAN model’s generator is a structure that adds a long skip connection to the existing U-Net structure, the discriminator is a patchGAN structure with a deformable convolution layer, and the loss function is a weighted sum of the GAN loss (adversarial loss) and the loss function to alleviate the instability of the existing GAN learning. In order to evaluate the performance of the proposed WT-UNetGAN model in this paper, we compared its performance with that of the conventional BM3D, DnCNN using CNN, standard DCGAN, U-Net GAN, and wavelet-based DCGAN on CT images corrupted by various noises, namely, gaussian noise, poisson noise, and speckle noise. The performance experimental results show that the proposed WT-UNetGAN model exhibits high noise removal capability in qualitative evaluation, and also shows high values in quantitative evaluation by PSNR (peak signal-to-noise ratio) and SSIM (structural similarity index measure).

키워드