FSformer: A combined frequency separation network and transformer for LDCT denoising

Low-dose computed tomography (LDCT) has been widely concerned in the field of medical imaging because of its low radiation hazard to humans. However, under low-dose radiation scenarios, a large amount of noise/artifacts are present in the reconstructed image, which reduces the clarity of the image and is not conducive to diagnosis. To improve the LDCT image quality, we proposed a combined frequency separation network and Transformer (FSformer) for LDCT denoising. Firstly, FSformer decomposes the LDCT images into low-frequency images and multi-layer high-frequency images by frequency separation blocks. Then, the low-frequency components are fused with the high-frequency components of different layers to remove the noise in the high-frequency components with the help of the potential texture of low-frequency parts. Next, the estimated noise images can be obtained by using Transformer stage in the frequency aggregation denoising block. Finally, they are fed into the reconstruction prediction block to obtain improved quality images. In addition, a compound loss function with frequency loss and Charbonnier loss is used to guide the training of the network. The performance of FSformer has been validated and evaluated on AAPM Mayo dataset, real Piglet dataset and clinical dataset. Compared with previous representative models in different architectures, FSformer achieves the optimal metrics with PSNR of 33.7714 dB and SSIM of 0.9254 on Mayo dataset, the testing time is 1.825 s. The experimental results show that FSformer is a state-of-the-art (SOTA) model with noise/artifact suppression and texture/organization preservation. Moreover, the model has certain robustness and can effectively improve LDCT image quality.

Transferring U-Net between low-dose CT denoising tasks: a validation study with varied spatial resolutions

Background

Recently, deep learning techniques have been widely used in low-dose computed tomography (LDCT) imaging applications for quickly generating high quality computed tomography (CT) images at lower radiation dose levels. The purpose of this study is to validate the reproducibility of the denoising performance of a given network that has been trained in advance across varied LDCT image datasets that are acquired from different imaging systems with different spatial resolutions.

Methods

Specifically, LDCT images with comparable noise levels but having different spatial resolutions were prepared to train the U-Net. The number of CT images used for the network training, validation and test was 2,400, 300 and 300, respectively. Afterwards, self- and cross-validations among six selected spatial resolutions (62.5, 125, 250, 375, 500, 625 µm) were studied and compared side by side. The residual variance, peak signal to noise ratio (PSNR), normalized root mean square error (NRMSE) and structural similarity (SSIM) were measured and compared. In addition, network retraining on a small number of image set was performed to fine tune the performance of transfer learning among LDCT tasks with varied spatial resolutions.

Results

Results demonstrated that the U-Net trained upon LDCT images having a certain spatial resolution can effectively reduce the noise of the other LDCT images having different spatial resolutions. Regardless, results showed that image artifacts would be generated during the above cross validations. For instance, noticeable residual artifacts were presented at the margin and central areas of the object as the resolution inconsistency increased. The retraining results showed that the artifacts caused by the resolution mismatch can be greatly reduced by utilizing about only 20% of the original training data size. This quantitative improvement led to a reduction in the NRMSE from 0.1898 to 0.1263 and an increase in the SSIM from 0.7558 to 0.8036.

Conclusions

In conclusion, artifacts would be generated when transferring the U-Net to a LDCT denoising task with different spatial resolution. To maintain the denoising performance, it is recommended to retrain the U-Net with a small amount of datasets having the same target spatial resolution.

Structure-preserving low-dose computed tomography image denoising using a deep residual adaptive global context attention network

Background

Low-dose computed tomography (LDCT) scans can effectively reduce the radiation damage to patients, but this is highly detrimental to CT image quality. Deep convolutional neural networks (CNNs) have shown their potential in improving LDCT image quality. However, the conventional CNN-based approaches rely fundamentally on the convolution operations, which are ineffective for modeling the correlations among nonlocal similar structures and the regionally distinct statistical properties in CT images. This modeling deficiency hampers the denoising performance for CT images derived in this manner.

Methods

In this paper, we propose an adaptive global context (AGC) modeling scheme to describe the nonlocal correlations and the regionally distinct statistics in CT images with negligible computation load. We further propose an AGC-based long-short residual encoder-decoder (AGC-LSRED) network for efficient LDCT image noise artifact-suppression tasks. Specifically, stacks of residual AGC attention blocks (RAGCBs) with long and short skip connections are constructed in the AGC-LSRED network, which allows valuable structural and positional information to be bypassed through these identity-based skip connections and thus eases the training of the deep denoising network. For training the AGC-LSRED network, we propose a compound loss that combines the L1 loss, adversarial loss, and self-supervised multi-scale perceptual loss.

Results

Quantitative and qualitative experimental studies were performed to verify and validate the effectiveness of the proposed method. The simulation experiments demonstrated the proposed method exhibits the best result in terms of noise suppression [root-mean-square error (RMSE) =9.02; peak signal-to-noise ratio (PSNR) =33.17] and fine structure preservation [structural similarity index (SSIM) =0.925] compared with other competitive CNN-based methods. The experiments on real data illustrated that the proposed method has advantages over other methods in terms of radiologists' subjective assessment scores (averaged scores =4.34).

Conclusions

With the use of the AGC modeling scheme to characterize the structural information in CT images and of residual AGC-attention blocks with long and short skip connections to ease the network training, the proposed AGC-LSRED method achieves satisfactory results in preserving fine anatomical structures and suppressing noise in LDCT images.

Dynamic controllable residual generative adversarial network for low-dose computed tomography imaging

Background

Computed tomography (CT) imaging technology has become an indispensable auxiliary method in medical diagnosis and treatment. In mitigating the radiation damage caused by X-rays, low-dose computed tomography (LDCT) scanning is becoming more widely applied. However, LDCT scanning reduces the signal-to-noise ratio of the projection, and the resulting images suffer from serious streak artifacts and spot noise. In particular, the intensity of noise and artifacts varies significantly across different body parts under a single low-dose protocol.

Methods

To improve the quality of different degraded LDCT images in a unified framework, we developed a generative adversarial learning framework with a dynamic controllable residual. First, the generator network consists of the basic subnetwork and the conditional subnetwork. Inspired by the dynamic control strategy, we designed the basic subnetwork to adopt a residual architecture, with the conditional subnetwork providing weights to control the residual intensity. Second, we chose the Visual Geometry Group Network-128 (VGG-128) as the discriminator to improve the noise artifact suppression and feature retention ability of the generator. Additionally, a hybrid loss function was specifically designed, including the mean square error (MSE) loss, structural similarity index metric (SSIM) loss, adversarial loss, and gradient penalty (GP) loss.

Results

The results obtained on two datasets show the competitive performance of the proposed framework, with a 3.22 dB peak signal-to-noise ratio (PSNR) margin, 0.03 SSIM margin, and 0.2 contrast-to-noise ratio margin on the Challenge data and a 1.0 dB PSNR margin and 0.01 SSIM margin on the real data.

Conclusions

Experimental results demonstrated the competitive performance of the proposed method in terms of noise decrease, structural retention, and visual impression improvement.

Bilateral Weighted Relative Total Variation for Low-Dose CT Reconstruction

Low-dose computed tomography (LDCT) has been widely used for various clinic applications to reduce the X-ray dose absorbed by patients. However, LDCT is usually degraded by severe noise over the image space. The image quality of LDCT has attracted aroused attentions of scholars. In this study, we propose the bilateral weighted relative total variation (BRTV) used for image restoration to simultaneously maintain edges and further reduce noise, then propose the BRTV-regularized projections onto convex sets (POCS-BRTV) model for LDCT reconstruction. Referring to the spacial closeness and the similarity of gray value between two pixels in a local rectangle, POCS-BRTV can adaptively extract sharp edges and minor details during the iterative reconstruction process. Evaluation indexes and visual effects are used to measure the performances among different algorithms. Experimental results indicate that the proposed POCS-BRTV model can achieve superior image quality than the compared algorithms in terms of the structure and texture preservation.

Artifact-Assisted multi-level and multi-scale feature fusion attention network for low-dose CT denoising

BACKGROUND AND OBJECTIVE

Since low-dose computed tomography (LDCT) images typically have higher noise that may affect accuracy of disease diagnosis, the objective of this study is to develop and evaluate a new artifact-assisted feature fusion attention (AAFFA) network to extract and reduce image artifact and noise in LDCT images.

METHODS

In AAFFA network, a feature fusion attention block is constructed for local multi-scale artifact feature extraction and progressive fusion from coarse to fine. A multi-level fusion architecture based on skip connection and attention modules is also introduced for artifact feature extraction. Specifically, long-range skip connections are used to enhance and fuse artifact features with different depth levels. Then, the fused shallower features enter channel attention for better extraction of artifact features, and the fused deeper features are sent into pixel attention for focusing on the artifact pixel information. Besides, an artifact channel is designed to provide rich artifact features and guide the extraction of noise and artifact features. The AAPM LDCT Challenge dataset is used to train and test the network. The performance is evaluated by using both visual observation and quantitative metrics including peak signal-noise-ratio (PSNR), structural similarity index (SSIM) and visual information fidelity (VIF).

RESULTS

Using AAFFA network improves the averaged PSNR/SSIM/VIF values of AAPM LDCT images from 43.4961, 0.9595, 0.3926 to 48.2513, 0.9859, 0.4589, respectively.

CONCLUSIONS

The proposed AAFFA network is able to effectively reduce noise and artifacts while preserving object edges. Assessment of visual quality and quantitative index demonstrates the significant improvement compared with other image denoising methods.

Low-dose CT noise reduction based on local total variation and improved wavelet residual CNN

BACKGROUND

Low-dose computed tomography (LDCT) is an effective method for reducing radiation exposure. However, reducing radiation dose leads to considerable noise in the reconstructed image that can affect doctor's judgment.

OBJECTIVE

To solve this problem, this study proposes a local total variation and improved wavelet residual convolutional neural network (LTV-WRCNN) denoising model.

METHODS

The model first introduces local total variation (LTV) to decompose the LDCT image into cartoon and texture image. Next, the texture image is filtered using the non-local mean (NLM). Then, the cartoon image is added to the filtered texture image to obtain the preprocessing image. Finally, the pre-processed image is fed into the improved wavelet residual neural network (WRCNN) to obtain an improved image. Additionally, we also introduce a compound loss in wavelet domain that combines mean squared error loss and directional regularization loss to separate the structural details from noise more thoroughly.

RESULTS

Compared with state-of-the-art methods, the peak-signal-to-noise ratio (PSNR) value and the structure similarity (SSIM) value of the processed CT images using the new proposed model are 33.4229 dB and 0.9158. Study also shows that applying new model obtains better results visually and numerically, especially in terms of the preservation of structural details.

CONCLUSIONS

The proposed new model is feasible and effective in improving the quality of LDCT images.

A phantom study comparing low-dose CT physical image quality from five different CT scanners

BACKGROUND

To systematically evaluate the physical image quality of low-dose computed tomography (LDCT) on CT scanners from 5 different manufacturers using a phantom model.

METHODS

CT images derived from a Catphan 500 phantom were acquired using manufacturer-specific iterative reconstruction (IR) algorithms and deep learning image reconstruction (DLIR) on CT scanners from 5 different manufacturers and compared using filtered back projection with 2 radiation doses of 0.25 and 0.75 mGy. Image high-contrast spatial resolution and image noise were objectively characterized by modulation transfer function (MTF) and noise power spectrum (NPS). Image high-contrast spatial resolution and image low-contrast detectability were compared directly by visual evaluation. CT number linearity and image uniformity were compared with intergroup differences using one-way analysis of variance (ANOVA).

RESULTS

The CT number linearity of 4 insert materials were as follows: acrylic (95% CI: 120.35 to 121.27; P=0.134), low-density polyethylene (95% CI: -98.43 to -97.43; P=0.070), air (95% CI: -996.16 to -994.51; P=0.018), and Teflon (95% CI: 984.40 to 986.87; P=0.883). The image uniformity values of GE Healthcare (95% CI: 3.24 to 3.83; P=0.138), Philips (95% CI: 2.62 to 3.70; P=0.299), Siemens (95% CI: 2.10 to 3.59; P=0.054), Minfound (95% CI: 2.35 to 3.65; P=0.589), and Neusoft (95% CI: 2.63 to 3.37; P=0.900) were evaluated and found to be within ±4 Hounsfield units (HU), with a range of 0.99-2.76 HU for standard deviations. There was no statistically significant difference in CT number linearity and image uniformity across the 5 CT scanners under different radiation doses with IR and DLIR algorithms (P>0.05). The resolution level at 10% MTF was 6.98 line-pairs-per-centimeter (lp/cm) on average, which was similar to the subjective evaluation results (mostly up to 7 lp/cm). DLIR at all 3 levels had the highest 50% MTF values among all reconstruction algorithms. For image low-contrast detectability, the minimum diameter of distinguishable contrast holes reached 4 mm at a 0.5% resolution. Increasing the radiation dose and IR strength reduced the image noise and NPS curve peak frequency while improving image low-contrast detectability.

CONCLUSIONS

This study demonstrated that the image quality of CT scanners from 5 different manufacturers in LDCT is comparable and that the CT number linearity is unbiased and can contribute to accurate bone mineral density quantification.

Lung cancer screening: who pays? who receives?-the Chinese perspective

The National Lung Screening Trial (NLST) revealed that low-dose computed tomography (LDCT) screening reduced lung cancer mortality by 20.0%. In China, LDCT is very cheap and easy to access. As a result, LDCT screening is not limited to "high-risk" population defined by the NLST trial. The results of LDCT screening in China are also quite different from that in Western countries. LDCT detected lung cancer in a significant proportion of young, female and non-smokers in China. There is also a higher proportion of adenocarcinoma (ADC), a lower proportion of squamous cell carcinoma, and a higher proportion of early-stage 0/I disease among LDCT-detected lung cancer in China. The issue of overdiagnosis and overtreatment is discussed. Finally, we call the global attention to clarify the etiology of lung cancer in young female non-smokers.

A quality improvement method for lung LDCT images

BACKGROUND

Low dose computed tomography (LDCT) reduces radiation damage to patients. However, with the decrease of radiation dose, LDCT images of the lung often appear some serious problems such as poor contrast, noise and streak artifacts.

OBJECTIVE

To improve the quality of lung LDCT images, this study proposed and investigated a new denoising method based on classification training structure combined dictionary for lung LDCT images.

METHODS

First, top-hat transform and anisotropic diffusion with a shock filter (ADSF) algorithm are used to enhance the image contrast and image details. Second, an adaptive dictionary is trained and used for noise reduction. Third, more image details are extracted from the residual image by using the atom activity measurement. The final result is obtained by combining the dictionary denoising result with the extracted detail information. The proposed method is then validated by both simulated and clinical lung LDCT images. Four metrics including Contrast-to-Noise Ratio (CNR), Noise Suppression Index (NSI), Edge Preserving Index (EPI), and Blurring Index (BI) are computed to quantitatively evaluate image quality.

RESULTS

The results showed that the CNR, NSI, EPI, and BI of our method reached 8.953, 0.9500, 0.7230 and 0.0170, respectively. Noise and streak artifacts can be removed from lung LDCT images while keeping and retaining more details.

CONCLUSIONS

Comparing with the results of other methods tested using the same dataset, this study demonstrated that our new method significantly improved quality of the lung LDCT images.

Screening values of carcinoembryonic antigen and cytokeratin 19 fragment for lung cancer in combination with low-dose computed tomography in high-risk populations: Initial and 2-year screening outcomes

OBJECTIVES

To assess added screening value of Carcinoembryonic Antigen (CEA) and Cytokeratin 19 Fragment (CYFRA 21-1) in combination with LDCT beyond LDCT alone and likelihood ratio of positive (LHR+) of their combination for lung cancer in high-risk populations with indeterminate and positive LDCT after initial screening and 2-year follow up.

MATERIALS AND METHODS

LDCT was performed annually at baseline and for 2 years in 634 heavy smokers (>30 pack-years) who were aged 50-70 years, and it was classified as negative, indeterminate, or positive (suspicious for lung cancer). Serum CEA and CYFRA 21-1 were examined and followed with LDCT in the indeterminate and positive LDCT groups and defined as positive with an abnormal level of either CEA or CYFRA 21-1.

RESULTS

A total of 17 lung cancer cases were diagnosed (9 from initial screening and 8 from follow-up cycles). Seventy and 22 patients had indeterminate and positive baseline LDCT, respectively. Among indeterminate baseline LDCT, the LHR+ for lung cancer diagnosed after initial screening with a positive marker was 6.61 (p = .039) and 1.51 (p = .502) with a negative marker. After 2 years follow up, the LHR+ was 6.31 (p = .004) and 0.86 (p = .677), respectively. Among positive baseline LDCT, the LHR+ for lung cancer diagnosed after initial round with positive and negative markers was 69.44 (p < 0.001) and 11.57 (p = .015), respectively. The corresponding LHR+ after 2-year round was 13.61 (p = .002) and 18.15 (p = .001), respectively. The combinations of CEA/CYFRA 21-1 and LDCT, and CEA and LDCT had crude and adjusted added value beyond LDCT alone (crude: 8%, p = .033 and 7%, p = .038; adjusted: 4%, p = .019 and 4%, p = .029, respectively).

CONCLUSIONS

CEA in combination with LDCT significantly increases the value of lung cancer screening compared with using LDCT alone particularly in participants with indeterminate baseline LDCT in both initial and 2-year screening outcomes.

Current estimate of costs of lung cancer screening in the United States

The United States Preventive Services Task Force recently endorsed the use of low-dose computed tomography for lung cancer screening in high-risk patients because of the potential to reduce deaths. Before implementation on a national level, it will be important to ensure that a safe, high-quality, and accessible service can be adequately provided. It will also be important to make sure that screening is cost-effective. This article summarizes the published analyses of lung cancer screening cost, provides a contemporary estimation of the annual cost of screening in the United States, and identifies areas for improvement in the future.