Unpaired Image Denoising Using a Generative Adversarial Network in X-Ray CT

Park, Hyoung Suk; Baek, Jineon; You, Sun Kyoung; Choi, Jae Kyu; Seo, Jin Keun

doi:10.1109/access.2019.2934178

Cited by 63 publications

(58 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The proposed method was compared with the PWLS method with total variation regularization (PWLS-TV) [23], nonlocal mean (PWLS-NLM) [10], IM-DL [19], and IM-DL with the cycle consistency loss (IM-Cycle) [9]. The model of PWLS-TV and PWLS-NLM is given by arg min…”

Section: Resultsmentioning

confidence: 99%

“…Note that the difference between IM-Cycle and IM-DL is that it use the cycle-consistency loss (the last two terms in (22)). As suggested in [9], [19], the parameters in IM-DL and IM-Cycle were set to α = 10 (α, γ) = (10, 5), respectively. Fig.3 shows the graph of the objective function of (7) over the number of outer iterations for AAPM datasets.…”

Section: Resultsmentioning

confidence: 99%

“…The added cycle consistency loss leads to performance improvement in denoising task, whereas it increases training time by more than two times. As suggested in [9], [19], training of the both IM-DL, IM-Cycle methods can be performed in a patch-by-patch manner, which can also lead to an improvement in the performance of the IM-DL and IM-Cycle methods.…”

Section: Resultsmentioning

confidence: 99%

“…where parameter α = 10 as in [19], and training was run 200 epochs using Adam optimization. Unlike the proposed method, the IM-DL model contains the least squares term Algorithm 1: Algorithm of the proposed method Initialization:…”

Section: Implementation Detailsmentioning

confidence: 99%

“…In this paper, we propose a deep PWLS method by combining it with a deep learning prior. Unlike the previous work of [19], we estimate the prior distribution in the sense of Pearson χ 2 divergence, using several SDCT samples. We then convert the proposed model into a least squares-GAN (LS-GAN) [13] that adopts the least squares loss function for the discriminator, including a PWLS objective.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Low-Dose CT Image Reconstruction With a Deep Learning Prior

2020

Self Cite

View full text Add to dashboard Cite

In low-dose computed tomography (LDCT), a penalized weighted least squares (PWLS) approach that incorporates the Poisson statistics of X-ray photons can significantly reduce excessive quantum noise. To improve the quality of LDCT images, prior information such as the total variation, Markov random field, and nonlocal mean, can be imposed onto the target image. However, this information may be limited to reflect the characteristics of the target images, thereby resulting in unexpected side effects (e.g. blurry images). In this paper, we propose a PWLS method combined with a deep learning prior, which is learned from standard dose CT (SDCT) images. The proposed model learns a noise reduction function that maps an LDCT image to its corresponding SDCT images and estimates the prior distribution using a Pearson χ 2 divergence. The model can be converted to the least squares generative adversarial network with an added PWLS objective, where the optimal generator acts as the noise reduction function. We reformulate the proposed model as a constrained optimization problem and solve it using the alternating optimization (AO) algorithm. Clinical SDCT and simulated LDCT scans of ten patients were used to show the validity of the proposed method. Results show that the proposed method outperforms other PWLS methods, by imposing priors such as the total variation and the nonlocal mean. INDEX TERMS Low-dose computed tomography, penalized weighted least squares, deep learning, denoising Recently, Park et al [19] have proposed an image-based deep learning model that can learn a noise reduction function that maps LDCT images to their standard dose CT (SDCT)

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Implementation Detailsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Low-Dose CT Image Reconstruction With a Deep Learning Prior

2020

Self Cite

View full text Add to dashboard Cite

show abstract

A fidelity‐embedded learning for metal artifact reduction in dental CBCT

et al. 2022

Self Cite

View full text Add to dashboard Cite

Purpose: Dental cone-beam computed tomography (CBCT) has been increasingly used for dental and maxillofacial imaging. However, the presence of metallic inserts, such as implants, crowns, and dental braces, violates the CT model assumption, which leads to severe metal artifacts in the reconstructed CBCT image, resulting in the degradation of diagnostic performance. In this study, we used deep learning to reduce metal artifacts. Methods: The metal artifacts, appearing as streaks and shadows, are nonlocal and highly associated with various factors, including the geometry of metallic inserts, energy-dependent attenuation, and energy spectrum of the incident X-ray beam, making it difficult to learn their complicated structures directly. To provide a step-by-step environment in which deep learning can be trained, we propose an iterative learning approach in which the network at each iteration step learns the correction error caused by the previous network, while enforcing the data fidelity in the projection domain. To generate a realistic paired training dataset, metal-free CBCT scans were collected from patients without metallic inserts, and then simulated metal projection data were added to generate the corresponding metal-corrupted projection data. Results: The feasibility of the proposed method was investigated in clinical metal-affected CBCT scans, as well as simulated metal-affected CBCT scans. The results show that the proposed method significantly reduces metal artifacts while preserving the morphological structures near metallic objects and outperforms direct image domain learning. Conclusion:The proposed fidelity-embedded learning can effectively reduce metal artifacts in dental CBCT compared with direct image domain learning.

show abstract

Unpaired low‐dose computed tomography image denoising using a progressive cyclical convolutional neural network

et al. 2023

Medical Physics

View full text Add to dashboard Cite

BackgroundReducing the radiation dose from computed tomography (CT) can significantly reduce the radiation risk to patients. However, low‐dose CT (LDCT) suffers from severe and complex noise interference that affects subsequent diagnosis and analysis. Recently, deep learning‐based methods have shown superior performance in LDCT image‐denoising tasks. However, most methods require many normal‐dose and low‐dose CT image pairs, which are difficult to obtain in clinical applications. Unsupervised methods, on the other hand, are more general.PurposeDeep learning methods based on GAN networks have been widely used for unsupervised LDCT denoising, but the additional memory requirements of the model also hinder its further clinical application. To this end, we propose a simpler multi‐stage denoising framework trained using unpaired data, the progressive cyclical convolutional neural network (PCCNN), which can remove the noise from CT images in latent space.MethodsOur proposed PCCNN introduces a noise transfer model that transfers noise from LDCT to normal‐dose CT (NDCT), denoised CT images generated from unpaired CT images, and noisy CT images. The denoising framework also contains a progressive module that effectively removes noise through multi‐stage wavelet transforms without sacrificing high‐frequency components such as edges and details.ResultsCompared with seven LDCT denoising algorithms, we perform a quantitative and qualitative evaluation of the experimental results and perform ablation experiments on each network module and loss function. On the AAPM dataset, compared with the contrasted unsupervised methods, our denoising framework has excellent denoising performance increasing the peak signal‐to‐noise ratio (PSNR) from 29.622 to 30.671, and the structural similarity index (SSIM) was increased from 0.8544 to 0.9199. The PCCNN denoising results were relatively optimal and statistically significant. In the qualitative result comparison, PCCNN without introducing additional blurring and artifacts, the resulting image has higher resolution and complete detail preservation, and the overall structural texture of the image is closer to NDCT. In visual assessments, PCCNN achieves a relatively balanced result in noise suppression, contrast retention, and lesion discrimination.ConclusionsExtensive experimental validation shows that our scheme achieves reconstruction results comparable to supervised learning methods and has performed well in image quality and medical diagnostic acceptability.

show abstract

Unpaired Image Denoising Using a Generative Adversarial Network in X-Ray CT

Cited by 63 publications

References 33 publications

Low-Dose CT Image Reconstruction With a Deep Learning Prior

Low-Dose CT Image Reconstruction With a Deep Learning Prior

A fidelity‐embedded learning for metal artifact reduction in dental CBCT

Unpaired low‐dose computed tomography image denoising using a progressive cyclical convolutional neural network

Contact Info

Product

Resources

About