Forward Projection Research Articles

Adjacent object distorts the organ-of-interest in filtered back-projection tomographic image reconstruction: 'ramp filter' or 'reconstruction' artifact revisited.

To simulate the artifact caused by an adjacent object on organ-of-interest during filtered back-projection (FBP) tomographic reconstruction (the so-called "ramp filter" artifact) and to demonstrate the extent to which an organ-of-interest is influenced by such adjacent hot spot or attenuating object. Two simulations are conducted with two simplified phantoms: "hot spot" and "cardiac." First one is used to visualize effect of hot spot on its periphery. Second one is used to evaluate effect of nearby object (liver) on left ventricle (LV) as organ-of-interest. To generate sinograms, forward projection process is done with and without modeling radiation attenuation. FBP using windowed ramp filter is done. All slices are analyzed by plotting intensity profiles. In tomographic slices, there is a hypo-intense halo around presumed edge of object when compared to background intensity, more noticeable in phantoms with less blurring. Even with ramp filter applied, no halo is visible in FBP without attenuation for cardiac phantom. In contrast, in slices with considering attenuation, zones with different levels of count suppression on both sides of object are visualized instead. The most prominent one is between liver and LV in simulation with higher-attenuating object and higher activity. A single hot spot with sufficient amount of blurring does not distort its surroundings. Hot spots and attenuating objects near organ-of-interest, however, distort myocardial perfusion imaging. Artifactual defects are thus only created when attenuation is modeled during FBP, producing zones of count suppression between organ-of-interest and nearby object or hot spot.

Deep learning acceleration of iterative model-based light fluence correction for photoacoustic tomography

Photoacoustic tomography (PAT) is a promising imaging technique that can visualize the distribution of chromophores within biological tissue. However, the accuracy of PAT imaging is compromised by light fluence (LF), which hinders the quantification of light absorbers. Currently, model-based iterative methods are used for LF correction, but they require extensive computational resources due to repeated LF estimation based on differential light transport models. To improve LF correction efficiency, we propose to use Fourier neural operator (FNO), a neural network specially designed for estimating partial differential equations, to learn the forward projection of light transport in PAT. Trained using paired finite-element-based LF simulation data, our FNO model replaces the traditional computational heavy LF estimator during iterative correction, such that the correction procedure is considerably accelerated. Simulation and experimental results demonstrate that our method achieves comparable LF correction quality to traditional iterative methods while reducing the correction time by over 30 times.

Direct voxel classification from x-ray projections for 3D pore detection applied to laser sintered parts

X-ray computed tomography (XCT) is a validated and frequently used tool to verify part geometry and to perform a non-destructive inspection of additive manufacturing parts. However, the acquisition of a large number of x-ray projections generates long inspection times. This conflicts with a high throughput of the production process and hinders the integration of XCT as an in-line quality control procedure for low-end parts. In this paper, we propose a method to obtain three-dimensional (3D) information of the internal pores from a limited view or limited angle scan. The method combines a forward projection model of a cone-beam x-ray system and a deep learning neural network to directly classify each individual voxel, based on the x-ray projections in order to avoid the reconstruction and segmentation step. Accompanying reconstruction artifacts for limited view and limited angle XCT scans are thereby reduced, while preserving 3D information of the pores, defects or inclusions inside the material. The method is validated on real x-ray projections of polymer laser sintered industrial parts and shows a significant reduction in the required x-ray projections, hence acquisition time.

Evaluation of Spectral X-Ray Imaging for Panoramic Dental Images Based on a Simulation Framework.

Modern photon counting detectors allow the calculation of virtual monoenergetic or material decomposed X-ray images but are not yet used for dental panoramic radiography systems. To assess the diagnostic potential and image quality of photon counting detectors in dental panoramic radiography, ethics approval from the local ethics committee was obtained for this retrospective study. Conventional CT scans of the head and neck region were segmented into bone and soft tissue. The resulting datasets were used to calculate panoramic equivalent thickness bone and soft tissue images by forward projection, using a geometry like that of conventional panoramic radiographic systems. The panoramic equivalent thickness images were utilized to generate synthetic conventional panoramic radiographs and panoramic virtual monoenergetic radiographs at various energies. The conventional, two virtual monoenergetic images at 40keV and 60keV, and material-separated bone and soft tissue panoramic equivalent thickness X-ray images simulated from 17 head CTs were evaluated in a reader study involving three experienced radiologists regarding their diagnostic value and image quality. Compared to conventional panoramic radiographs, the material-separated bone panoramic equivalent thickness image exhibits a higher image quality and diagnostic value in assessing the bone structure and details such as teeth or root canals . Panoramic virtual monoenergetic radiographs do not show a significant advantage over conventional panoramic radiographs. The conducted reader study shows the potential of spectral X-ray imaging for dental panoramic imaging to improve the diagnostic value and image quality.

PARALLELPROJ-an open-source framework for fast calculation of projections in tomography.

In this article, we introduce parallelproj, a novel open-source framework designed for efficient parallel computation of projections in tomography leveraging either multiple CPU cores or GPUs. This framework efficiently implements forward and back projection functions for both sinogram and listmode data, utilizing Joseph's method, which is further extended to encompass time-of-flight (TOF) PET projections. Our evaluation involves a series of tests focusing on PET image reconstruction using data sourced from a state-of-the-art clinical PET/CT system. We thoroughly benchmark the performance of the projectors in non-TOF and TOF, sinogram, and listmode employing multi CPU-cores, hybrid CPU/GPU, and exclusive GPU mode. Moreover, we also investigate the timing of non-TOF sinogram projections calculated in STIR (Software for Tomographic Image Reconstruction) which recently integrated parallelproj as one of its projection backends. Our results indicate that the exclusive GPU mode provides acceleration factors between 25 and 68 relative to the multi-CPU-core mode. Furthermore, we demonstrate that OSEM listmode reconstruction of state-of-the-art real-world PET data sets is achievable within a few seconds using a single consumer GPU.

Simulation Study: Data-Driven Material Decomposition in Industrial X-ray Computed Tomography

Material-resolving computed tomography is a powerful and well-proven tool for various clinical applications. For industrial scan setups and materials, several problems, such as K-edge absence and beam hardening, prevent the direct transfer of these methods. This work applies dual-energy computed tomography methods for material decomposition to simulated phantoms composed of industry-relevant materials such as magnesium, aluminium and iron, as well as some commonly used alloys like Al–Si and Ti64. Challenges and limitations for multi-material decomposition are discussed in the context of X-ray absorption physics, which provides spectral information that can be ambiguous. A deep learning model, derived from a clinical use case and based on the popular U-Net, was utilised in this study. For various reasons outlined below, the training dataset was simulated, whereby phantom shapes and material properties were sampled arbitrarily. The detector signal is computed by a forward projector followed by Beer–Lambert law integration. Our trained model could predict two-material systems with different elements, achieving a relative error of approximately 1% through simulated data. For the discrimination of the element titanium and its alloy Ti64, which were also simulated, the relative error increased to 5% due to their similar X-ray absorption coefficients. To access authentic CT data, the model underwent testing using a 10c euro coin composed of an alloy known as Nordic gold. The model detected copper as the main constituent correctly, but the relative fraction, which should be 89%, was predicted to be ≈70%.

Exploring the Influence of Initial Estimates on Iterative Maximum Likelihood Expectation-maximization in Tomographic Reconstruction.

To explore the influence of initial guess or estimate (uniform as "ones" and "zeros" vs. filtered back projection [FBP] image) as an input image for maximum likelihood expectation-maximization (MLEM) tomographic reconstruction algorithm and provide the curves of error or convergence for each of these three initial estimates. Two phantoms, created as digital images, were utilized: one was a simple noiseless object and the other was a more complicated, noise-degraded object of the section of lower thorax in a matrix of 256 × 256 pixels. Both underwent radon transform or forward projection process and the corresponding sinograms were generated. For filtering during tomographic image reconstruction, ramp and Butterworth filters, as high-pass and low-pass ones, were applied to images. The second phantom (lower thorax) was radon-transformed and the resulting sinogram was degraded by noise. As initial guess or estimate images, in addition to FBP tomographic image, two uniform images, one with all pixels having a value of 1 ("ones") and the other with all having zero ("zeros"), were created. The three initial estimates (FBP, ones, and zeros) were reconstructed with iterative MLEM tomographic reconstruction (with 1, 2, 4, 8, 16, 32, and 64 iterations). The difference between the object and the updated slice was calculated at the end of each iteration (as error matrix), and the mean squared error (MSE) was computed and plotted separately or in conjunction with the MSE curves of other initial estimates. All computations were implemented in MATLAB software. The results of ones and zeros seemed strikingly similar. The curves of uniform ones and uniform zeros were so close to each other that overlap near-perfectly. However, in the FBP slice as an initial estimate, the resulting tomographic slice was similar with a much higher extent to the object even after 1 or 2 iterations. The pattern of convergence for all three curves was roughly similar. The normalized MSE decreased sharply up to 5 iterations and then, after 10 iterations, the curves reached a plateau until 32 iterations. For the phantom of the lower thorax section with its noise-degraded sinogram, similar to the pattern observed for simple disk-shaped phantom, the curves (normalized MSE) fell sharply up to 10 iterations and then rapidly converged thereafter until 64 iterations. Similar results are observed when choosing different initial guesses or estimates (uniform vs. FBP) as the starting point, based on the error calculation using MSE. The algorithm converges almost similarly for all initial estimates. Therefore, selecting a uniform initial guess image can be an appropriate choice and may be preferred over an FBP image. Reducing the processing time can be a valid reason for this choice.

Intra-session reliability of isometric muscle strength of the bilateral standing press in female handball players.

Systematizing reliable protocols and procedures for strength assessment in handball has allowed for a more thorough kinetic analysis and increased precision in detecting training-induced changes in muscular strength. The evaluation of upper limb strength with a bilateral standing press (BSP) exercise in handball players approximates blocking actions in the defensive phase, pushing and fixation. The aim of this study was to analyze the relative and absolute reliability of intra-session comparisons in an isometric peak muscle strength protocol in a bilateral standing press (BSP) exercise among female handball players. Sixteen young female handball players at an international level, aged between 22 ± 4 years, with no prior experience in using functional electromechanical dynamometers (FEMD) participated in this study. The participants initiated the assessment of maximum isometric force in a bipedal stance with a forward projection of the dominant foot between 20 and 30 cm. The knees were kept semi-flexed, and the hip extension of the non-dominant limb ranged from 15 to 20°. Bilateral anterior push of the upper limb was performed with shoulder abduction and elbow flexion at 90°, maintaining a pronated hand position. Participants were instructed to exert three sets of maximal force for 5 s in BSP exercise. Relative reliability was assessed using the model intraclass correlation (ICC) and absolute reliability was assessed using the coefficient of variation (CV) and standard error of measurement (SEM). For this study, the parameters of maximum and mean muscle strength were considered. The results demonstrated high relative reliability (ICC 0.93-0.97) and absolute reliability (SEM 0.19-2.79) y (CV 4.78-9.03) for both mean force and peak force, with no significant differences between the sets (p > 0.05), indicating a negligible effect size (0.01-0.12). The mean and peak isometric muscle strength for the BSP exercise controlled with FEMD in female handball players exhibits high relative and absolute reliability between series.

GPU-accelerated polyenergetic forward projection for 9 MeV industrial CT system

Polyenergetic forward projection has great significance in inspecting hazardous materials, establishing optimal radiographic variables and investigating beam hardening effects. However, it is computationally intensive to perform polyenergetic forward-projection calculations for high-resolution phantoms. To address this issue, a rapid polyenergetic forward-projection algorithm is proposed for a 9 MeV industrial computed tomography (CT) system. The FLUktuierende KAskade (FLUKA) software package is used to generate the 9 MeV X-ray spectrum data. Two voxelised phantoms are used to model scanned objects, one being a multi-material cylinder and the other a single-material turbine blade. An incremental version of Siddon's algorithm is adopted to calculate the intersection lengths between the X-rays and the auxiliary phantoms. Three strategies are utilised to accelerate the calculation, in which: the intersection lengths do not vary with the energy bins and can be used repeatedly until all the energy bins are counted; a graphics processing unit (GPU) is used to accelerate the ray tracing algorithm by utilising a parallel computing technique; and faster memory access is achieved by binding the auxiliary phantoms to texture objects. The simulation results in this paper show that the GPU-based approach not only maintains the image precision but also gains significant speed-ups over the conventional central processing unit (CPU)-based Siddon method. Furthermore, beam hardening artefacts can clearly be seen from the profile curves of the reconstructed slices, indicating that this method is effective.

Origin of paired extension-compression during rotational rifting: An early Paleogene example from the northeast Atlantic region and its implications

Abstract The formation of paired extension-compression (PEC) postulated by rotational kinematics of rift propagation is demonstrated by analogue models but rarely observed in nature. In our study of the early Paleogene continental rift in the northeast Atlantic, a PEC is proposed based on the northeastward propagation and the coeval compression at the rift tip. The propagation is deduced from tectono-magmatic trends, including along-axis development of magmatism, and migration of tectonic faulting inward and toward the rift tip. Where this propagation terminated, we documented coeval extension and compression in the form of a horst-and-graben system (H&G) and V-shaped anticline (VA), respectively. Given their structural characteristics and spatiotemporal relationships with the rift, their origin is best illustrated by a three-stage model: (1) Rifting initiated at the site of mantle upwelling and propagated northeastward in the Paleocene. (2) The rift tip was stalled by an elevated mafic-ultramafic body at the Barents Shelf, which led to forward projection of the rift’s driving force to create the H&G and the VA (PEC), dissipating the along-axis force component. (3) Domination of axis-perpendicular components then promoted orthogonal extension and sheared margin development. Our study suggests that PEC has a crucial role in both termination of propagation and rift-mode conversion.

Backward and forward Wasserstein projections in stochastic order

Motivated by applications of stochastic orders in statistics and economics, we study metric projections onto cones in the Wasserstein space of probability measures, defined by stochastic orders. Duality theorems for backward and forward projections are established under mild conditions, which lay a foundation for investigating sampling of measures in stochastic orders. To demonstrate applications of these theorems beyond practice, particular attention is given to mathematical properties of convex order and subharmonic order. While backward and forward cones possess distinct geometric properties, strong connections between backward and forward projections can be obtained in the convex order case. Compared with convex order, the study of subharmonic order is subtler, particularly the existence and regularity of optimal mappings. In all cases, Brenier-Strassen type polar factorization theorems are proved, thus providing a full picture of the decomposition of optimal couplings between probability measures given by deterministic contractions (resp. expansions) and stochastic couplings. The factorization of convex order projection completes the decomposition obtained by Gozlan and Juillet, which builds a connection with Caffarelli's contraction theorem. A further noteworthy addition to earlier results is the decomposition in the subharmonic order case where the optimal mappings are characterized by volume distortion properties. To our knowledge, this is the first time in this occasion such results are available in the literature.

S-Wave Accelerates Optimization-based Photoacoustic Image Reconstruction in vivo

Photoacoustic imaging has undergone rapid development in recent years. To simulate photoacoustic imaging on a computer, the most popular MATLAB toolbox currently used for the forward projection process is k-Wave. However, k-Wave suffers from significant computation time. Here we propose a straightforward simulation approach based on superposed Wave (s-Wave) to accelerate photoacoustic simulation. In this study, we consider the initial pressure distribution as a collection of individual pixels. By obtaining standard sensor data from a single pixel beforehand, we can easily manipulate the phase and amplitude of the sensor data for specific pixels using loop and multiplication operators. The effectiveness of this approach is validated through an optimization-based reconstruction algorithm. The results reveal significantly reduced computation time compared with k-Wave. Particularly in a sparse 3-D configuration, s-Wave exhibits a speed improvement >2000 times compared with k-Wave. In terms of optimization-based image reconstruction, in vivo imaging results reveal that using the s-Wave method yields images highly similar to those obtained using k-Wave, while reducing the reconstruction time by approximately 50 times. Proposed here is an accelerated optimization-based algorithm for photoacoustic image reconstruction, using the fast s-Wave forward projection simulation. Our method achieves substantial time savings, particularly in sparse system configurations. Future work will focus on further optimizing the algorithm and expanding its applicability to a broader range of photoacoustic imaging scenarios.

Refraction error analysis in stereo vision for system parameters optimization

The traditional stereo vision system with an air-based imaging model is unsuitable for multi-layer imaging conditions, where refraction effects can introduce significant errors in the measurement outcomes. Although numerous theoretical studies have developed refraction correction algorithms, it may be more beneficial and desirable to reduce the error by optimizing the system parameters in some cases. To address this dilemma, in this study, we first build a three-layer stereo vision model by forward projection and backward reconstruction. Then, the impact of various factors on the errors of reconstructed coordinates in the conditions of air-glass-air and air-glass-water is demonstrated through simulations. Real experiments are further carried out to verify the validity of the simulation. Finally, a practical guide for conducting experiments and tests is presented to assist in the design of more effective system parameters.

Adjustment of acquisition arc in cardiac malposition during myocardial perfusion SPECT imaging: computer simulation based on deterministic modeling

ObjectivesTo simulate cardiac malpositions, leftward and rightward shift and dextrocardia, and also to compare distribution of activity of septal and lateral walls of left ventricle acquired in standard acquisition arc and after relevant adjustment. MethodsIn this study, digital phantoms with cardiac malpositions are designed and procedure of acquisition of scan in standard arc (from right anterior oblique to left posterior oblique) and adjusted acquisition arc is simulated. The three situations of malposition including leftward and rightward shift and dextrocardia are considered. For all types, acquisition is conducted in standard and then adjusted arcs (from anterior to posterior and also from right to left for leftward and rightward shifts, respectively, and for dextrocardia, from left anterior oblique to right posterior oblique). All obtained projections are reconstructed using the algorithm of filtered back projection. During forward projection to obtain sinograms, radiation attenuation is also modeled by incorporation of a simplified transmission map to emission map. The resulting tomographic slices of the LV (septum, apex, and lateral wall) are presented visually and are compared by plotting intensity profiles of the walls. Finally, normalized error images are also computed. All the computations are performed in MATLAB software package. ResultsIn transverse slice, septum and lateral wall are attenuated progressively from apex, which is closer to the camera, to the base in similar fashion. In tomographic slices of standard acquisition arc, the septum shows remarkably higher activity compared to lateral wall. However, after adjustment, both seems equally intense and progressively being attenuated from apex to base, similar to that found in phantom with normally positioned heart. Likewise, for the phantom with rightward shift, when the scanning was done in standard arc, the septum is more intense than the lateral wall. And similarly, adjustment of the arc renders both walls equally intense. In dextrocardia, level of attenuation of basal parts of septum and lateral wall is higher in 360° arc compared to adjusted 180° arc. ConclusionAdjustment of acquisition arc exerts perceptible changes in distribution of activity over LV walls which are more compatible with normally positioned heart.

A dataset of ecological quality indexes of terrestrial ecosystems in China from 2000 to 2018

The deteriorating ecological conditions have attracted widespread attention, spurring the development of quantitative analysis methods for ecological quality, improving the concept of ecological quality, and thus facilitating the formulation of future ecological environmental protection policies. Using satellite remote sensing and ecosystem process model data, we constructed a conceptual framework for ecological quality based on a pixel-level historical baseline of current ecosystem functions and biodiversity habitats through data quality control in forward normalization, correlation analysis, and projection pursuit algorithm based on particle swarm optimization, resulting in dataset of ecological quality indexes of terrestrial ecosystems in China from 2000 to 2018 with a spatial resolution of 1 km. By sharing this dataset, we aim to provide a data foundation for the advancement of ecological assessment theories and methods, as well as the management and decision-making of macro ecosystems.

Real Time 4D-Cone Beam CT via Single-Angle Projection with Dual Attention Mechanism

The hypothesis is that dual attention integrated Residual Network (DA-ResNet) together with a Principal Component Analysis (PCA) model accurately maps the measured on-treatment CBCT projections from different phases to each phase's 3D-CBCT real timely. Finally, it achieves real time 4D-CBCT. The DA-ResNet contains a CBAM (Convolutional Block Attention Module) and a SPA (Spatial Pyramid Structure). Hundreds of forward projections from one view of patient 4D-CT were simulated to train DA-ResNet. The extracted motion features were expressed as PCA coefficients. To test DA-ResNet performance, several single measured on-treatment projection is sent into it and the network output several groups of PCA coefficients real timely. Each group of PCA coefficient stands for the corresponding deformation field between the 0% phase and the predicted phase that the PCA label stands for. In labels such as Mean Absolute Error (MAE), R square, Normalized Cross Correlation (NCC), Structural Similarity Index Measure (SSIM) were used to evaluate DA-ResNet performance with other state-of-the-art networks such as CNN/Unet/ResNet. DA-ResNet outperforms other popular networks on all of the quantification labels, with MAE (14.41 ± 8.85), R square (0.95 ± 0.07) in PCA coefficient prediction, and NCC (0.9981 ± 0.0006), SSIM (0.98 ± 0.04) in image reconstruction. Compared with conventional ResNet, DA-ResNet reduced MAE by 79%. Meanwhile R square rises from 0.88 to 0.95, NCC rises from 0.996 to 0.998, SSIM rises from 0.94 to 0.98. DA-ResNet offers a solution for real time on-treatment 4D-CBCT image guidance by using single 2D on-line projections. It offers better motion prediction and image reconstruction accuracy compared with traditional deep learning networks (e.g., CNN/Unet/ResNet). The initial results are promising. The prospect of clinical application needs further clinical verification.

Deep learning-based fluorescence image correction for high spatial resolution precise dosimetry

Objective. While radiation-excited fluorescence imaging has great potential to measure absolute 2D dose distributions with high spatial resolution, the fluorescence images are contaminated by noise or artifacts due to Cherenkov light, scattered light or background noise. This study developed a novel deep learning-based model to correct the fluorescence images for accurate dosimetric application. Approach. 181 single-aperture static photon beams were delivered to an acrylic tank containing quinine hemisulfate water solution. The emitted radiation-exited optical signals were detected by a complementary metal-oxide semiconductor camera to acquire fluorescence images with 0.3 × 0.3 mm2 pixel size. 2D labels of projected dose distributions were obtained by applying forward projection calculation of the 3D dose distributions calculated by a clinical treatment planning system. To calibrate the projected dose distributions for Cherenkov angular dependency, a novel empirical Cherenkov emission calibration method was performed. Total 400-epoch supervised learning was applied to a convolutional neural network (CNN) model to predict the projected dose distributions from fluorescence images, gantry, and collimator angles. Accuracy of the calculated projected dose distributions was evaluated with that of uncorrected or conventional methods by using a few quantitative evaluation metrics. Main results. The projected dose distributions corrected by the empirical Cherenkov emission calibration represented more accurate noise-free images than the uncalibrated distributions. The proposed CNN model provided accurate projected dose distributions. The mean absolute error of the projected dose distributions was improved from 2.02 to 0.766 mm·Gy by the CNN model correction. Moreover, the CNN correction provided higher gamma index passing rates for three different threshold criteria than the conventional methods. Significance. The deep learning-based method improves the accuracy of dose distribution measurements. This technique will also be applied to optical signal denoising or Cherenkov light discrimination in other imaging modalities. This method will provide an accurate dose verification tool with high spatial resolution.

Refraction correction for deep-water three-dimensional visual measurement based on multi-objective optimization.

Refraction-induced errors affect the accuracy of three-dimensional visual measurements in deepwater environments. In this study, a binocular camera refractive imaging model was established, and a calibration method for the refraction parameters was proposed for high-accuracy shape and deformation measurements in deep-water environments. First, an initial estimate of the refractive axis was obtained using a three-dimensional calibration target. Then, the errors in the distance between the spatial point pairs and the reprojection errors are taken as the dual optimization objectives, and the Non-dominated Sorting Genetic Algorithm II is applied to optimize the refraction parameters. To efficiently calculate the reprojection error, an improved numerical computation method is proposed to accelerate the calculation of the analytical forward projection. Underwater experiments were conducted to verify the method's effectiveness. The results showed that the average error of the absolute position of the reconstructed points was less than 1.1 mm and the average error of the displacement was less than 0.04 mm. This study provides a sound solution for accurate three-dimensional visual measurement in deep-water environments.

A new dental CBCT metal artifact reduction method based on a dual-domain processing framework

Objective. Cone beam computed tomography (CBCT) has been wildly used in clinical treatment of dental diseases. However, patients often have metallic implants in mouth, which will lead to severe metal artifacts in the reconstructed images. To reduce metal artifacts in dental CBCT images, which have a larger amount of data and a limited field of view compared to computed tomography images, a new dental CBCT metal artifact reduction method based on a projection correction and a convolutional neural network (CNN) based image post-processing model is proposed in this paper. Approach. The proposed method consists of three stages: (1) volume reconstruction and metal segmentation in the image domain, using the forward projection to get the metal masks in the projection domain; (2) linear interpolation in the projection domain and reconstruction to build a linear interpolation (LI) corrected volume; (3) take the LI corrected volume as prior and perform the prior based beam hardening correction in the projection domain, and (4) combine the constructed projection corrected volume and LI-volume slice-by-slice in the image domain by two concatenated U-Net based models (CNN1 and CNN2). Simulated and clinical dental CBCT cases are used to evaluate the proposed method. The normalized root means square difference (NRMSD) and the structural similarity index (SSIM) are used for the quantitative evaluation of the method. Main results. The proposed method outperforms the frequency domain fusion method (FS-MAR) and a state-of-art CNN based method on the simulated dataset and yields the best NRMSD and SSIM of 4.0196 and 0.9924, respectively. Visual results on both simulated and clinical images also illustrate that the proposed method can effectively reduce metal artifacts. Significance. This study demonstrated that the proposed dual-domain processing framework is suitable for metal artifact reduction in dental CBCT images.

A transformer-based dual-domain network for reconstructing FOV extended cone-beam CT images from truncated sinograms in radiation therapy

Background and objectiveCone-beam computed tomography (CBCT) is widely used in clinical radiotherapy, but its small field of view (sFOV) limits its application potential. In this study, a transformer-based dual-domain network (dual_swin), which combined image domain restoration and sinogram domain restoration, was proposed for the reconstruction of complete CBCT images with extended FOV from truncated sinograms. MethodsThe planning CT images with large FOV (LFOV) of 330 patients who received radiation therapy were collected. The synthetic CBCT (sCBCT) images with LFOV were generated from CT images by the trained cycleGAN network, and CBCT images with sFOV were obtained through forward projection, projection truncation, and filtered back projection (FBP), comprising the training and test data. The proposed dual_swin includes sinogram domain restoration, image domain restoration, and FBP layer, and the swin transformer blocks were used as the basic feature extraction module in the network to improve the global feature extraction ability. The proposed dual_swin was compared with the image domain method, the sinogram domain method, the U-Net based dual domain network (dual_Unet), and the traditional iterative reconstruction method based on prior image and conjugate gradient least-squares (CGLS) in the test of sCBCT images and clinical CBCT images. The HU accuracy and body contour accuracy of the predicted images by each method were evaluated. ResultsThe images generated using the CGLS method were fuzzy and obtained the lowest structural similarity (SSIM) among all methods in the test of sCBCT and clinical CBCT images. The predicted images by the image domain methods are quite different from the ground truth and have low accuracy on HU value and body contour. In comparison with image domain methods, sinogram domain methods improved the accuracy of HU value and body contour but introduced secondary artifacts and distorted bone tissue. The proposed dual_swin achieved the highest HU and contour accuracy with mean absolute error (MAE) of 23.0 HU, SSIM of 95.7%, dice similarity coefficient (DSC) of 99.6%, and Hausdorff distance (HD) of 4.1 mm in the test of sCBCT images. In the test of clinical patients, images that were predicted by dual_swin yielded MAE, SSIM, DSC, and HD of 38.2 HU, 91.7%, 99.0%, and 5.4 mm, respectively. The predicted images by the proposed dual_swin has significantly higher accuracy than the other methods (P < 0.05). ConclusionsThe proposed dual_swin can accurately reconstruct FOV extended CBCT images from the truncated sinogram to improve the application potential of CBCT images in radiotherapy.