A streak artifact reduction algorithm in sparse-view CT using a self-supervised neural representation.

Sparse view - computed tomography (CT) for low dose and photon counting detector (PCD) for spectral imaging have been studied for improvement of image quality and quantification in medical imaging. The sparse view–CT can reduce dose, but there is a limitation that cannot be completely restored yet and PCD with physical phenomena such as charge sharing, K-escape and material characteristic can be difficult to material quantification due to different distribution of noise characteristics in a specific energy band. In this study, we propose a deep running-based wavelet-CNN for the efficient reduction of physical factors such as noise and streak artifact generated by fusion of sparse view-CT and PCD. The physical phenomena of the spatio-energetic cross-talks were reflected in PCD. We obtained images with a total of four energy thresholds with limited angles and trained through the proposed method. The proposed method was evaluated for the image quality by the peak signal to noise ratio (PSNR), the normalized mean square error (NMSE), the structural similarity (SSIM), the multi-scale SSIM (MS-SSIM), and the feature similarity (FSIM). The experimental results demonstrated that the sparse view-CT with PCD using proposed deep running structure effectively removes the streak artifacts and improves the image quality.

Low-dose computed tomography (CT) has attracted much attention in clinical applications since X-ray radiations can cause serious health risks to patients. Sparse-view CT imaging is one of the major ways to reduce radiation dose. However, if reconstructing the sparse-view CT images with conventional filtered backprojection (FBP) algorithm, image quality may be significantly degraded due to the severe streaking artifacts. Therefore, iterative sparse-view CT image reconstruction (IR) algorithms have been developed and utilized to improve image quality. One drawback of using the IR algorithms is they usually spend long computation time. Additionally, adjusting and optimizing the hyper-parameters that are needed during iteration procedures is also time-consuming, some- times may even depend on individual experience. These potential drawbacks strongly limit the wide applications of IR algorithms. Aiming at partially overcome such difficulties, in the present work, we propose a deep iterative reconstruction (DIR) framework to generalize the conventional IR algorithms by mapping them into the deep neural network (DNN) technique. With this proposed DIR algorithms, the prior term, the data fidelity term, and the hyper-parameters can all be represented and learned by network. By doing so, some generalized iterative models can be used to perform high quality sparse-view CT image reconstructions. Numerical experiments based on clinical patient data demonstrated that the proposed DIR algorithms can mitigate the streaking artifacts more effectively while well preserving the subtle structures.

This paper aims to investigate the problem of low-dose computed tomography (CT) reconstruction with prior image constraints using the compressed sensing (CS) theorem. The CS theorem states that images can be reconstructed from under-sampled data in an adequate or transfer domain without introducing noticeable artifacts by solving a convex optimization problem if the source signals are sparse. To describe the sparsity, a model of piecewise constant source distribution has recently been assumed for image reconstruction by minimizing the total variance (TV) of the image density distribution in the Fourier domain. However, the assumption may not hold for complicated image structures. It has been observed that a prior image from the same subject or anatomy can provide excellent information to the image to be reconstructed. Based on this observation, this study investigates the problem of image reconstruction from under-sampled data by minimizing the difference between the prior image and the concerned image to be estimated with data constraints in the Fourier domain. Compared to the TV criterion, this presented method doesn't require the piecewise constant assumption where the similarity between the two images specifies a new priori model for a new cost function. The presented method was tested by computer simulations using the Shepp-Logan phantom. In noise-free case, only 64 projections around the phantom are needed to produce an accurate reconstruction. The reconstruction remained excellent until the number of projections was reduced to 22 when a high similarity exists between the prior and concerned images while the well-known filtered backprojection reconstruction failed. In cases with noise variance at 1% level, the signal-to-noise of the reconstruction by presented CS-based approach dropped rapidly when the number of projections decreased from 64 to 22. This investigation reveals the high sensitivity of the CS-based approach for low-dose CT image reconstruction. Modification of the cost function to consider data statistics is needed.

Computed tomography (CT) is widely used in medicine for diagnostics or for image-guided therapies, and is also popular in industrial applications for nondestructive testing. CT conventionally requires a large number of projections to produce volumetric images of a scanned object, because the conventional image reconstruction algorithm is based on filtered-backprojection. This requirement may result in relatively high radiation dose to the patients in medical CT unless the radiation dose at each view angle is reduced, and can cause expensive scanning time and efforts in industrial CT applications. Sparse- view CT may provide a viable option to address both issues including high radiation dose and expensive scanning efforts. However, image reconstruction from sparsely sampled data in CT is in general very challenging, and much efforts have been made to develop algorithms for such an image reconstruction problem. Image total-variation minimization algorithm inspired by compressive sensing theory has recently been developed, which exploits the sparseness of the image derivative magnitude and can reconstruct images from sparse-view data to a similar quality of the images conventionally reconstructed from many views. In successive CT scans, prior CT image of an object and its projection data may be readily available, and the current CT image may have not much difference from the prior image. Considering the sparseness of such a difference image between the successive scans, image reconstruction of the difference image may be achieved from very sparsely sampled data. In this work, we showed that one can further reduce the number of projections, resulting in a super-sparse scan, for a good quality image reconstruction with the aid of a prior data. Both numerical and experimental results are provided.

Computed tomography (CT), as a noninvasive method first invented in the early 1970s, has become a crucial technique in medical imaging with a variety of applications including detections of cancer, cardiovascular, neurological disorders, and so on. However, conventional CT scans are inherited with the issue of a high radiation dose, which may raise health concerns against radiologists and patients. Thus, CT radiation dose is intentionally decreased in exchange for human well-being. However, a lower radiation dose causes the problem of producing a CT image with degraded quality compared with a normal dose. Furthermore, quick and small-scale CT scans are often required when young patients or dynamic organs are being inspected. Therefore, CT scans may fail to capture full-angular coverage or acquire a sufficient number of scans for the patients, thereby having limited-angle and sparse-view of CT images, which makes it hard for them to reflect the accurate features of the target object. To fix these issues while not raising the time and amount of dose of CT scans, low-dose computed tomography (LDCT), limited-angle computed tomography (LACT), and sparse-view computed tomography have emerged as promising solutions for CT image reconstruction. However, traditional approaches in implementing LDCT, LACT, and sparse-view CT may struggle with insufficient data or secondary artifacts. Therefore, it is a foreseeable trend for LDCT, LACT, and sparse-view CT to combine with deep learning so that less data is required to reconstruct CT images with fewer secondary artifacts and improved qualities. For instance, DDPNet model with global residual learning + UNet-structured network and DuDoUFNet with blocked residual learning + UNet-structured network for LDCT; DIOR with residual block + UNet-structured network and DOLCE with denoising diffusion probabilistic model (DDPM) for LACT; PINER with black-box deep model + physical-consistency optimization and DreamNet with residual block + UNet-structured network for sparse-view CT. This paper aims to introduce state-of-the-art deep models in LDCT, LACT, and sparse-view CT based on workflow and data analysis.

PurposeRepeated computed tomography (CT) scans are prescribed for some clinical applications such as lung nodule surveillance. Several studies have demonstrated that incorporating a high-quality prior image into the reconstruction of subsequent low-dose CT (LDCT) acquisitions can either improve image quality or reduce data fidelity requirements. Our proposed previous normal-dose image induced nonlocal means (ndiNLM) regularization method for LDCT is an example of such a method. However, one major concern with prior image based methods is that they might produce false information when the prior image and the current LDCT image show different structures (for example, if a lung nodule emerges, grows, shrinks or disappears over time). This study aims to assess the performance of the ndiNLM regularization method in situations with change in anatomy.MethodWe incorporated the ndiNLM regularization into the statistical image reconstruction (SIR) framework for reconstruction of subsequent LDCT images. Because of its patch-based search mechanism, a rough registration between the prior image and the current LDCT image is adequate for the SIR-ndiNLM method. We assessed the performance of the SIR-ndiNLM method in lung nodule surveillance for two different scenarios: (1) the nodule was not found in a baseline exam but appears in a follow-up LDCT scan; (2) the nodule was present in a baseline exam but disappears in a follow-up LDCT scan. We further investigated the effect of nodule size on the performance of the SIR-ndiNLM method.ResultsWe found that a relatively large search-window (e.g., 33×33) should be used for the SIR-ndiNLM method to account for misalignment between the prior image and the current LDCT image, and to ensure that enough similar patches can be found in the prior image. With proper selection of other parameters, experimental results with two patient datasets demonstrated that the SIR-ndiNLM method did not miss true nodules nor introduce false nodules in the lung nodule surveillance scenarios described above. We also found that the SIR-ndiNLM reconstruction shows improved image quality when the prior image is similar to the current LDCT image in anatomy. These gains in image quality might appear small upon visual inspection, but they can be detected using quantitative measures. Finally, the SIR-ndiNLM method also performed well in ultra-low-dose conditions and with different nodule sizes.ConclusionsThis study assessed the performance of the SIR-ndiNLM method in situations in which the prior image and the current LDCT image show substantial anatomical differences, specifically, changes in lung nodules. The experimental results demonstrate that the SIR-ndiNLM method does not introduce false lung nodules nor miss true nodules, which relieves the concern that this method might produce false information. However, there is insufficient evidence that these findings will hold true for all kinds of anatomical changes.

Sparse-view low-dose computed tomography (LDCT) imaging poses difficulties in preserving image quality while reducing radiation exposure. Recent research has focused extensively on artificial intelligence (AI) to reduce artifacts in LDCT. This paper presents a unique integration based on a conditional generative adversarial network (CGAN) with metaheuristic optimization to improve the reconstruction of sparse-view computed tomography (CT) images. A Pix2Pix CGAN-based model was integrated with Particle Swarm Optimization (PSO), Differential Evolution (DE), and Cuckoo Search (CS) to improve essential hyperparameters, such as learning rate and beta values. The LDCT-P and LUNA16 datasets were used, producing seven levels of sparse-view CT images (10, 16, 32, 64, 128, 256, and 512 views) for assessment. The findings indicated a substantial improvement in image quality with an increase in the number of view projections. Pix2Pix+PSO demonstrated superior performance, with the Structural Similarity Index metric (SSIM) rising from 0.900 (10 views) to 0.972 (512 views) for abdominal CT and from 0.801 to 0.971 for lung CT, respectively. The results underscore the capability of the Pix2Pix model integrated with metaheuristic optimization to enhance sparse-view CT reconstruction. This method adeptly reconciles computing economy with image integrity, enabling improved LDCT imaging applications in clinical settings.

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.

Computed tomography (CT) is a valuable imaging modality for pulmonary imaging. Fast acquisition times and sharp cross-sectional images guarantee high diagnostic confidence. With the introduction of low-dose CT, it has been established as standard for lung screening of heavy smokers in many countries around the world. However, at some point the limits for dose reduction with conventional CT are reached and further reduction would suffer from poor image quality. Sparsesampling CT is one technology that would allow a further radiation dose reduction by reducing the number of acquired projection images. Recently, the feasibility of a fast pulsing X-ray tube for CT has been demonstrated, indicating that sparse sampling could become available in future generations of CT scanners. Therefore, we investigated the effect of sparse sampling by a stepwise reduction of the projection images. A lung phantom with synthetic pulmonary nodules was scanned with a clinical CT system. Sparse sampling was simulated by removing projection images prior to reconstruction. The phantom was scanned at the iso-center and at the highest possible table position (off-center). The modulation transfer function (MTF) was determined for different degrees of sparse sampling. Image quality was evaluated by comparing the reduced dose simulations against the full dose image using the structural similarity index (SSIM). MTF was stable down to using 1/4th of the projection images (4-times sparse sampling, SpS-4) with high degradation at the off-center position (full sampling (FS) 10% MTF, iso-center: 0.64; off-center: 0.47). SSIM indicates a small image quality degradation of FS images compared to sparse-sampling images at low radiation doses at the iso-center (35 mAs; FS: 0.91; SpS-4: 0.93) and stronger degradations at the off-center position (35 mAs; FS: 0.65; SpS-4: 0.84). In conclusion, sparse sampling provides stable MTF results down to 1/4th of the projection images. At low dose levels (iso-center: ≤43 mAs; off-center: ≤86 mAs), sparse sampling performs better in terms of SSIM compared to FS.

Sparse-view computed tomography (CT) holds great promise for speeding up data acquisition and reducing radiation dose in CT scans. Recent advances in reconstruction algorithms for sparse-view CT, such as iterative reconstruction algorithms, obtained high-quality image while requiring advanced computing power. Lately, deep learning (DL) has been widely used in various applications and has obtained many remarkable outcomes. In this paper, we propose a new method for sparse-view CT reconstruction based on the DL approach. The method can be divided into two steps. First, filter backprojection (FBP) was used to reconstruct the CT image from sparsely sampled sinogram. Then, the FBP results were fed to a DL neural network, which is a DenseNet and deconvolution-based network (DD-Net). The DD-Net combines the advantages of DenseNet and deconvolution and applies shortcut connections to concatenate DenseNet and deconvolution to accelerate the training speed of the network; all of those operations can greatly increase the depth of network while enhancing the expression ability of the network. After the training, the proposed DD-Net achieved a competitive performance relative to the state-of-the-art methods in terms of streaking artifacts removal and structure preservation. Compared with the other state-of-the-art reconstruction methods, the DD-Net method can increase the structure similarity by up to 18% and reduce the root mean square error by up to 42%. These results indicate that DD-Net has great potential for sparse-view CT image reconstruction.

Dual-domain sparse-view CT reconstruction with Transformers

Background: Radiation exposure from computed tomography (CT) remains a significant concern in medical imaging, prompting continuous technological innovation to optimize dose reduction while maintaining diagnostic image quality. This systematic review synthesizes recent developments in CT technology for radiation dose reduction from 2020 to 2025. Objective: To systematically review and analyze technological advancements in CT imaging that reduce X-ray radiation exposure, evaluate their clinical effectiveness, and identify future research directions. Methods: A comprehensive literature search was conducted across four major databases (SciSpace, PubMed, Google Scholar, and full-text repositories) for publications from January 2020 to November 2025. Search terms included combinations of “computed tomography,” “CT,” “radiation dose reduction,” “dose optimization,” “low dose,” and related technological terms. Studies were included if they reported on technological innovations for CT dose reduction with quantitative outcomes. A total of 99 unique papers were identified, combined, and reranked by relevance. Results: Three primary technological tracks emerged: (1) Advanced reconstruction algorithms (iterative reconstruction and deep learning-based methods) achieving 36-89% dose reductions; (2) Photon-counting detector CT (PCCT) providing superior image quality at significantly lower doses compared to conventional energy-integrating detectors; (3) Optimized acquisition techniques including automatic exposure control, tube current modulation, and spectral shaping with tin filters achieving up to 89% dose reduction in specific protocols. Real-world clinical implementations demonstrated CTDIvol reductions up to 82% and DLP reductions up to 83.9% while maintaining or improving diagnostic accuracy. Conclusions: Substantial progress has been made in CT dose reduction through convergent advances in reconstruction algorithms, detector hardware, and acquisition strategies. Deep learning reconstruction and photon-counting detectors represent paradigm shifts with proven clinical benefits. However, challenges remain in external validation, standardization across vendors, and task-specific optimization. Future research should focus on prospective multicenter trials, standardized benchmarking protocols, and integrated AI-driven dose optimization systems.

Characterization of anatomical change and other differences is important in sequential computed tomography (CT) imaging, where a high-fidelity patient-specific prior image is typically present, but is not used, in the reconstruction of subsequent anatomical states. Here, we introduce a penalized likelihood (PL) method called reconstruction of difference (RoD) to directly reconstruct a difference image volume using both the current projection data and the (unregistered) prior image integrated into the forward model for the measurement data. The algorithm utilizes an alternating minimization to find both the registration and reconstruction estimates. This formulation allows direct control over the image properties of the difference image, permitting regularization strategies that inhibit noise and structural differences due to inconsistencies between the prior image and the current data. Additionally, if the change is known to be local, RoD allows local acquisition and reconstruction, as opposed to traditional model-based approaches that require a full support field of view (or other modifications). We compared the performance of RoD to a standard PL algorithm, in simulation studies and using test-bench cone-beam CT data. The performances of local and global RoD approaches were similar, with local RoD providing a significant computational speedup. In comparison across a range of data with differing fidelity, the local RoD approach consistently showed lower error (with respect to a truth image) than PL in both noisy data and sparsely sampled projection scenarios. In a study of the prior image registration performance of RoD, a clinically reasonable capture ranges were demonstrated. Lastly, the registration algorithm had a broad capture range and the error for reconstruction of CT data was 35% and 20% less than filtered back-projection for RoD and PL, respectively. The RoD has potential for delivering high-quality difference images in a range of sequential clinical scenarios including image-guided surgeries and treatments where accurate and quantitative assessments of anatomical change is desired.

Prior-image-based Reconstruction for On-board CBCT Imaging: Application in Image-guided Radiotherapy for Head and Neck Cancer

To investigate the feasibility of low-dose CT (LDCT) with tin filtration and third-generation iterative reconstruction (IR) in evaluating interstitial lung diseases associated with connective tissue disease (CTD-ILD). Fifty-three consecutive adult patients with CTD-ILD underwent regular-dose chest CT (RDCT) at 110kVp followed by LDCT with tin-filtered 100kVp. RDCT was reconstructed with filtered back projection (FBP) and advanced modeled iterative reconstruction (ADMIRE); LDCT was reconstructed with ADMIRE. Image noise, streak artifact, image quality, and visualization of normal and abnormal CT features were evaluated and compared among RDCT-ADMIRE, RDCT-FBP, and LDCT-ADMIRE groups. The mean radiation dose of LDCT was reduced to 20% of RDCT. Objective image noise of RDCT-ADMIRE (38.08 ± 6.37 HU), LDCT-ADMIRE (51.68 ± 9.06 HU), and RDCT-FBP (62.09 ± 10.95 HU) increased progressively (p < 0.001 in any two pairs). RDCT-ADMIRE significantly improved subjective image noise, streak artifact, and overall image quality compared with RDCT-FBP and LDCT-ADMIRE (all p < 0.001), while no significant difference was noted between the latter two groups. All abnormal lung structures were better scored in RDCT-ADMIRE compared with those in RDCT-FBP (all p < 0.001). LDCT-ADMIRE was inferior to RDCT-FBP in visualizing peripheral bronchi and vessels as well as reticulation (all p < 0.001); other normal and abnormal structures were similar between the two groups. LDCT with tin filtration and third-generation IR was applicable in evaluating ILD lesions of CTD. Image quality was significantly improved after applying ADMIRE algorithm to CT protocols. • Optimization of CT radiation dose is a clinical concern in patients with connective tissue disease. • Spectral shaping and third-generation iterative reconstruction emerge as promising techniques in reducing radiation dose and acquiring desired image quality of CTD-ILD patients. • The third-generation iterative reconstruction algorithm can optimize visualization of ILD patterns in low-dose CT.