Low-Dose CT Image Denoising Using a Generative Adversarial Network With Wasserstein Distance and Perceptual Loss

Abstract

The continuous development and extensive use of computed tomography (CT) in medical practice has raised a public concern over the associated radiation dose to the patient. Reducing the radiation dose may lead to increased noise and artifacts, which can adversely affect the radiologists' judgment and confidence. Hence, advanced image reconstruction from low-dose CT data is needed to improve the diagnostic performance, which is a challenging problem due to its ill-posed nature. Over the past years, various low-dose CT methods have produced impressive results. However, most of the algorithms developed for this application, including the recently popularized deep learning techniques, aim for minimizing the mean-squared error (MSE) between a denoised CT image and the ground truth under generic penalties. Although the peak signal-to-noise ratio is improved, MSE- or weighted-MSE-based methods can compromise the visibility of important structural details after aggressive denoising. This paper introduces a new CT image denoising method based on the generative adversarial network (GAN) with Wasserstein distance and perceptual similarity. The Wasserstein distance is a key concept of the optimal transport theory and promises to improve the performance of GAN. The perceptual loss suppresses noise by comparing the perceptual features of a denoised output against those of the ground truth in an established feature space, while the GAN focuses more on migrating the data noise distribution from strong to weak statistically. Therefore, our proposed method transfers our knowledge of visual perception to the image denoising task and is capable of not only reducing the image noise level but also trying to keep the critical information at the same time. Promising results have been obtained in our experiments with clinical CT images.

Fig. 1

The overall structure of the proposed WGAN-VGG network. In Part 1, n stands for the number of convolutional kernels and s for convolutional stride. So, n32s1 means the convolutional layer has 32 kernels with stride 1.

Fig. 2

The structure of the discriminator network. n and s have the same meaning as in Fig. 1

Fig. 3

Optimization procedure of WGAN-VGG network.

Fig. 4

Plots of validation loss versus the number of epochs during the training of the 5 networks. (a) MSE loss convergence, (b) VGG loss convergence and (c) Wasserstein estimation convergence.

Fig. 5

Transverse CT images of the abdomen demonstrate a low attenuation liver lesion (in the red box) and a cystic lesion in the upper pole of the left kidney (in the blue box). This display window is [−160, 240]HU.

Fig. 6

Zoomed ROI of the red rectangle in Fig. 5. The low attenuation liver lesion with in the dashed circle represents metastasis. The lesion is difficult to assess on quarter dose FBP recon (b) due to high noise content. This display window is [−160, 240]HU.

Fig. 7

Transverse CT images of the abdomen demonstrate small low attenuation liver lesions. The display window is [−160, 240]HU.

Fig. 8

Zoomed ROI of the red rectangle in Fig. 7 demonstrates the two attenuation liver lesions in the red and blue circles. The display window is [−160, 240]HU.

Fig. 9

VGG feature maps of full dose and quarter dose images in Fig. 5 and their absolute difference.