RAD-UNet: a Residual, Attention-Based, Dense UNet for CT Sparse Reconstruction.

PurposeTo assess whether the joint application of hybrid iterative reconstruction (HIR) and an adaptive filter (AF) could reduce streak artifacts and improve image quality of neck‐and‐shoulder computed tomography (CT).MethodsThis study included 96 patients with suspicious neck lesions who underwent a routine nonenhanced scan on a 64‐slice CT scanner. The raw data were reconstructed using four different settings: filtered back projection (FBP), HIR, FBP + AF, and HIR + AF. Regions of interest were manually drawn in erector spine, axillary fat, latissimus dorsi, and dorsal cervical fat. Mean and standard deviation (SD) of the CT number, signal‐to‐noise ratio (SNR), and contrast‐to‐noise ratio (CNR) were obtained and compared using Wilcoxon signed‐rank tests. The qualitative assessments of five factors were compared by two independent investigators.ResultsCompared to the other three settings, HIR + AF reduced noise in the area where the streak artifact of the lower neck were most serious (SD; all p ≤ 0.001). The SNR and CNR were improved significantly (all p ≤ 0.001). Compared to the other three settings, HIR + AF showed a significant improvement in CT image quality regarding the visibility of suspicious lesions, the extent of streaking artifacts, noise, soft‐tissue contrast, and visualization of small structures (all p ≤ 0.02).ConclusionsThe combination of HIR and AF can significantly reduce streaking artifacts and improve image quality in neck‐and‐shoulder CT imaging.

In recent years, health risks concerning high-dose x-ray radiation have become a major concern in dental computed tomography (CT) examinations. Therefore, adopting low-dose computed tomography (LDCT) technology has become a major focus in the CT imaging field. One of these LDCT technologies is downsampling data acquisition during low-dose x-ray imaging processes. However, reducing the radiation dose can adversely affect CT image quality by introducing noise and artifacts in the resultant image that can compromise diagnostic information. In this paper, we propose an artifact correction method for downsampling CT reconstruction based on deep learning. We used clinical dental CT data with low-dose artifacts reconstructed by conventional filtered back projection (FBP) as inputs to a deep neural network and corresponding high-quality labeled normal-dose CT data during training. We trained a generative adversarial network (GAN) with Wasserstein distance (WGAN) and mean squared error (MSE) loss, called m-WGAN, to remove artifacts and obtain high-quality CT dental images in a clinical dental CT examination environment. The experimental results confirmed that the proposed algorithm effectively removes low-dose artifacts from dental CT scans. In addition, we showed that the proposed method is efficient for removing noise from low-dose CT scan images compared to existing approaches. We compared the performances of the general GAN, convolutional neural networks, and m-WGAN. Through quantitative and qualitative analysis of the results, we concluded that the proposed m-WGAN method resulted in better artifact correction performance preserving the texture in dental CT scanning. The image quality evaluation metrics indicated that the proposed method effectively improves image quality when used as a postprocessing technique for dental CT images. To the best of our knowledge, this work is the first deep learning architecture used with a commercial cone-beam dental CT scanner. The artifact correction performance was rigorously evaluated and demonstrated to be effective. Therefore, we believe that the proposed algorithm represents a new direction in the research area of low-dose dental CT artifact correction.

To compare the quality of CT images of the lung reconstructed using deep learning-based reconstruction (True Fidelity Image: TFI ™; GE Healthcare) to filtered back projection (FBP), and to determine the minimum tube current–time product in TFI without compromising image quality. Four cadaveric human lungs were scanned on CT at 120 kVp and different tube current–time products (10, 25, 50, 75, 100, and 175 mAs) and reconstructed with TFI and FBP. Two image evaluations were performed by three independent radiologists. In the first experiment, using the same tube current–time product, a side-by-side TFI and FBP comparison was performed. Images were evaluated with regard to noise, streak artifacts, and overall image quality. Overall image quality was evaluated in view of whole image quality. In the second experiment, CT images reconstructed using TFI and FBP with five different tube current–time products were displayed in random order, which were evaluated with reference to the 175 mAs-FBP image. Images were scored with regard to normal structure, abnormal findings, noise, streak artifacts, and overall image quality. Median scores from three radiologists were statistically analyzed. Quantitative evaluation of noise was performed by setting regions of interest (ROIs) in air. In first experiment, overall image quality was improved, and noise was decreased in images of TFI compared to that of FBP for all tube current–time products. In second experiment, scores of all evaluation items except for small vessels in images of 25 mAs-TFI were almost the same as that of 175 mAs-FBP (all p > 0.31). Using TFI instead of FBP, at least 85% radiation dose reduction could be possible without any degradation in the image quality.

The image quality of deep-learning image reconstruction of chest CT images on a mediastinal window setting

The objective of this study was to assess the value of an integrated circuit (IC) detector, potentially improving spatial resolution by means of reduced cross talk between detector channels, in coronary computed tomographic (CT) angiography regarding image quality and stenosis quantification compared with conventional detector technology. In the ex vivo part of the study, a coronary phantom including 63 defined stenoses and 7 plaque densities (degree of stenosis, 10%-90%; plaque densities, -100 to 1000 Hounsfield unit [HU]) was loaded with contrast agent diluted to 300 HU and placed in an anthropomorphic chest phantom. The phantom was scanned in 0-, 45-, and 90-degree orientations to the z-axis of the CT scanner table. Images were acquired using 128-section dual-source CT equipped with IC and with conventional detector technology. Data were reconstructed with filtered back projection (FBP) and with sinogram-affirmed iterative reconstruction (IR) at a slice thickness of 0.6 mm (increment, 0.4 mm). Data acquired with the IC detector were additionally reconstructed with a slice thickness of 0.5 mm (increment, 0.3 mm) combined with IR. Two readers rated image quality; image noise and degree of stenosis were assessed. In the in vivo part of the study, phantom observations were validated in 30 consecutive patients (11 women; mean [SD] age, 62 [13] years; mean [SD] heart rate, 81 [17] beats per minute) undergoing coronary CT angiography with IC for clinical indications. Images of the patients were reconstructed with FBP (slice thickness, 0.6 mm) and with IR (slice thickness, 0.5 mm) and were assessed for image quality and degree of stenosis. Interreader agreement for image quality was evaluated using intraclass correlation coefficients. The image quality was compared with the Wilcoxon signed rank test. The image noise and the degree of stenosis were compared with the Student t test for paired samples. The interreader agreement for the assessment of image quality was substantial (intraclass correlation coefficients, 0.79). The image quality was significantly (P < 0.001) higher for the images acquired with the IC detector as compared with the conventional detector. The image noise with IR was significantly (P = 0.020) reduced for the IC detector as compared with the conventional detector. The IC detector yielded significantly more accurate results regarding stenosis grading when compared with the images acquired with the conventional detector regarding both FBP (mean [SD] error FBP, 12.1% [7.6%] vs 13.7% [7.6%]; P = 0.043) and IR (mean [SD] error IR, 10.5% [6.6%] vs 13.0% [6.9%]; P < 0.001). The images with a slice thickness of 0.5 mm reconstructed with IR (mean [SD] error, 8.8% [5.9%]) obtained by the IC detector significantly (P < 0.001) improved measurement accuracy in the phantom as compared with FBP with a slice thickness of 0.6 mm (mean [SD] error, 12.1% [7.6%]). In the patients, we found a significantly (P < 0.001) higher image quality, and stenoses were quantified significantly (P = 0.009) smaller with FBP as compared with IR (mean stenosis, 47.6% vs 42.1%; mean difference, 5.5%). Our ex vivo and patient study indicates significantly reduced image noise and more accurate stenosis quantification in coronary CT angiography when acquiring data using an IC detector and combining IR with high-resolution images as compared with conventional detector technology and FBP reconstructions.

A newly-developed metal artifact reduction algorithm improves the visibility of oral cavity lesions on 320-MDCT volume scans

While the recent advancements of computed tomography (CT) technology have contributed in reducing radiation dose and image noise, an objective evaluation of image quality in patient scans has not yet been established. In this study, we present a patient-specific CT image quality evaluation method that includes fully automated measurements of noise level, structure sharpness, and alteration of structure. This study used the CT images of 120 patients from four different CT scanners reconstructed with three types of algorithm: filtered back projection (FBP), vendor-specific iterative reconstruction (IR), and a vendor-agnostic deep learning model (DLM, ClariCT.AI, ClariPi Inc.). The structure coherence feature (SCF) was used to divide an image into the homogeneous (RH) and structure edge (RS) regions, which in turn were used to localize the regions of interests (ROIs) for subsequent analysis of image quality indices. The noise level was calculated by averaging the standard deviations from five randomly selected ROIs on RH, and the mean SCFs on RS was used to estimate the structure sharpness. The structure alteration was defined by the standard deviation ratio between RS and RH on the subtraction image between FBP and IR or DLM, in which lower structure alterations indicate successful noise reduction without degradation of structure details. The estimated structure sharpness showed a high correlation of 0.793 with manually measured edge slopes. Compared to FBP, IR and DLM showed 34.38% and 51.30% noise reduction, 2.87% and 0.59% lower structure sharpness, and 2.20% and -12.03% structure alteration, respectively, on an average. DLM showed statistically superior performance to IR in all three image quality metrics. This study is expected to contribute to enhance the CT protocol optimization process by allowing a high throughput and quantitative image quality evaluation during the introduction or adjustment of lower-dose CT protocol into routine practice.

In many practical situations, images are not only down sampled but also compressed for efficient transmission and storage. JPEG and MPEG-2 compressions often introduce blocking artifacts because they process the data as 8 × 8 blocks. Many of the existing super resolution (SR) methods assume low resolution images as a down sampled version of high resolution (HR) image, and neglect the degradation due to compression. This exacerbates artifacts in the SR image and reduces the user experience. To address the joint deblocking and SR (DbSR), a novel deep network with dense skip connections and dilated convolutions is proposed in this paper, and we name it as DenseDbSR. Recently, many researchers have proposed deeper networks and achieved improvement in the SR performance. However, training deeper networks is very challenging because of the problem of vanishing gradients. Simply increasing the depth of the network leads to cumbersome computational costs. To enlarge the field-of-view (FOV) without increasing the computational cost, the dilated convolution is used. The dilated convolution exponentially expands the FOV and helps to exploit the contextual information efficiently. Moreover, the dense skip connections create short paths for gradients to be back-propagated efficiently and alleviates the problem of vanishing gradients. Furthermore, the network is relieved from the training burden by learning residuals of the SR image instead of learning raw images. From the conducted extensive experimentation, the proposed DenseDbSR network produced better performance in terms of PSNR and SSIM than the compared state-of-the-art methods.

Blind super-resolution for handheld ultrasound image: Two-stage degradation based unpaired deep learning.

The purpose of the study was to assess the impact of iterative model reconstruction (IMR) on reader confidence with respect to stone detection and image quality in comparison to filtered back-projection (FBP) and iDose level 4 (iDose4) in abdominal MDCT with radiation doses below 2 mSv. For 32 consecutive patients with suspected ureteral stone disease, the raw data of unenhanced 256 slice MDCT (120 kV, 40 reference mAs, mean CTDIvol: 2.7 ± 0.8 mGy, mean DLP: 126 ± 38 mGy × cm) were reconstructed using a prototype version of IMR (levels 1 - 3), iDose4 (level 4) and FBP at a 3 mm slice thickness. Image analysis was independently performed by two radiologists in a blinded fashion. The reader confidence level with respect to stone detection was recorded based on a 5-point scale (1 - certain exclusion; 5 - concrement definitely present) as well as for the evaluation of image quality regarding the depiction of anatomical details (1 - poor; 5 - excellent). A clinical reference standard for stone detection was not established. Statistical evaluation included weighted kappa analysis and Wilcoxon test. 17 pelvic and ureteral stones were found. 11 further concrements were located within the ostium of the urinary bladder or the bladder itself. Applying IMR, a distinct improvement in image quality was observed at every level (mean value for FBP, 2.0; iDose4, 2.9; IMR L1, 4.2; IMR L2, 4.0; IMR L3, 3.9; all p < 0.001). Applying the higher IMR levels L2 and L3, a certain level of so-called "blotchiness" of anatomical contours was observed. Reader confidence was significantly improved and was independent of IMR level (certain stone detection FBP, 69 %; iDose4, 81 %; IMR L1 to L3, 95 %; all p > 0.001). With increasing IMR levels, the reduction in streak artifacts was quantified by a decrease in image noise. A loss of anatomical information was not observed. The sensitivity rates for stone detection were equivalent for all MDCTs reconstructed with FBP, iDose4 and IMR. A mean effective dose of 1.9 ± 0.6 mSv was calculated. In comparison to FBP and iDose4, a significant increase in mean image quality, reduction in image noise and improvement in subjective reader confidence can be achieved by applying IMR even at significantly reduced dose settings below 2 mSv. Results indicate that a further dose reduction might be possible with IMR. · Urinary tract. · urolithiasis. · iterative reconstruction. · Schmidt-Holtz J, Laqmani A, Butscheidt S et al. Iterative Model Reconstruction (IMR) in MDCT Below 2mSv for the Detection of Urinary Calculi: Diagnostic Accuracy and Image Quality in Comparison to Filtered Back-Projection and 4th Generation Iterative Reconstruction (iDose4). Fortschr Röntgenstr 2018; 190: 630 - 636.

Magnetic Particle Imaging (MPI), as an emerging non-invasive imaging technique, acquires high-resolution images by detecting the response of magnetic nanoparticles. Reconstruction based on System Matrix (SM) is an important component of MPI research. However, when scanning parameters, particle types, or environmental conditions change, SM requires repetitive and time-consuming calibration scans, significantly reducing its practicality. In this study, we propose a novel SM super-resolution recovery method, PGSM-net, based on Generative Adversarial Network (GAN) and Pyramid Attention Mechanism (PAM) to enhance SM calibration in MPI. The generator utilizes enhanced ResNet blocks and a PAM module for multi-scale feature extraction, attention weighting, and feature fusion to recover finer details from low-resolution inputs. The discriminator, an unpooled VGG network, assesses high-dimensional feature consistency to aid the generator in producing more realistic and higher-quality images. Our approach represents the first application of GAN and perceptual loss in MPI system matrix calibration. Experimental results show significant improvements in SM recovery quality and detail, highlighting its potential in MPI applications.

Deep Feature Blend Attention: A New Frontier in Super Resolution Image Generation

Sinogram domain metal artifact correction of CT via deep learning

Micro X-ray computed tomography (micro-CT) provides higher spatial resolution than medical CT scanners. To observe fine structural details, it is important to fully leverage the capabilities of micro-CT. One effective approach is to decrease the source-object distance (SOD) to obtain highmagnification images. However, decreasing the SOD can lead to physical interference between the rotating object and the X-ray source. As a result, the scanning angle becomes restricted, resulting in a limited-angle CT. Incomplete projection data leads to severe streak artifacts in CT images when using analytical reconstruction algorithms. To deal with this problem, deep learning methods recently have been explored. However, most of these methods require paired limitedangle and full-view datasets for supervised training. In magnified CT imaging, acquiring full-view data at a short SOD is difficult. We propose a weakly supervised-based artifacts reduction method for limited-angle CT using limited-angle data acquired at a short SOD with full-view data acquired at a long SOD. The model generates a CT image from a random noise. This generated CT image is forward-projected using both short- and long-SOD geometries to produce corresponding sinograms. The model parameters are optimized by minimizing the losses between these projected sinograms and the measured sinograms at each SOD. Experimental results demonstrated that the proposed method can effectively reduce streak artifacts. Thus, the proposed method has the potential to reconstruct high-quality CT images from limited-angle sinograms without requiring short-SOD full-view data.

X-ray computed tomography (CT) is widely used in radiation therapy treatment planning in recent years. However, metal implants such as dental fillings and hip prostheses can cause severe bright and dark streaking artifacts in reconstructed CT images. These artifacts decrease image contrast and degrade HU accuracy, leading to inaccuracies in target delineation and dose calculation. In this work, a metal artifact reduction method is proposed based on the intrinsic anatomical similarity between neighboring CT slices. Neighboring CT slices from the same patient exhibit similar anatomical features. Exploiting this anatomical similarity, a gamma map is calculated as a weighted summation of relative HU error and distance error for each pixel in an artifact-corrupted CT image relative to a neighboring, artifact-free image. The minimum value in the gamma map for each pixel is used to identify an appropriate pixel from the artifact-free CT slice to replace the corresponding artifact-corrupted pixel. With the proposed method, the mean CT HU error was reduced from 360 HU and 460 HU to 24 HU and 34 HU on head and pelvis CT images, respectively. Dose calculation accuracy also improved, as the dose difference was reduced from greater than 20% to less than 4%. Using 3%/3mm criteria, the gamma analysis failure rate was reduced from 23.25% to 0.02%. An image-based metal artifact reduction method is proposed that replaces corrupted image pixels with pixels from neighboring CT slices free of metal artifacts. This method is shown to be capable of suppressing streaking artifacts, thereby improving HU and dose calculation accuracy.