X-ray Image Reconstruction Research Articles

X-ray image reconstruction for continuous acquisitions with a generalized motion model

Continuous X-ray imaging is known to reduce mechanical vibrations and scan time compared to a step-and-shoot acquisition approach. However, motion during X-ray exposure leads to blurred projections and consequently to loss of spatial resolution and contrast in conventionally reconstructed images. Recent works that aim to reduce continuous motion blur focus only on rotational motion and often include linearization approximations, while many applications would benefit from a more generalized continuous acquisition strategy. In this paper, we propose a dedicated reconstruction technique for rotational, translational, and roto-translational motion, without requiring a linearization approximation. Through simulations and real experiments, we show that motion blur artifacts caused by roto-translational continuous scanning are substantially reduced, allowing for faster scanning while retaining image quality.

First look at neutron emission shape characteristics of ignition hotspots at the National Ignition Facility (invited)

The nuclear imaging system has been capturing neutron images of inertial confinement fusion (ICF) driven implosions for over a decade at the National Ignition Facility. This imaging system has evolved from one to three nearly orthogonal lines-of-sight, allowing for the study of three-dimensional shape characteristics of ignition shots. Limited-view tomography algorithms help visualize the burning hotspot in 3D and assess neutron source geometry using Legendre mode parameters. With its neutron, gamma-ray, and x-ray image reconstruction capabilities, NIS has provided critical insight into mechanisms that have limited implosion performance, such as fill tube diameter for ignition-type targets. This comprehensive diagnostic suite opens a window into the shape characteristics of ignition shots and how symmetry affects ICF implosion performance. In more recent ignition shots, neutron yields have visibly increased. Analyzing the shape and size of the reconstructed neutron source has shown an expansion of the burn volume, which is indicative of more efficient alpha heating during the implosion process.

Grating-based spatial harmonic frequency X-ray imaging for qualitative and quantitative studies of granular and liquids materials

Grating-based spatial harmonic frequency X-ray imaging method allows the individual reconstruction of attenuation, scattering and phase-contrast X-ray images with a single exposure of the samples. In this work, the influence of the attenuation grating features (visibility and projected period) and sample features (thickness and grains sizes) on the reconstructed images were experimentally explored. Results showed that fringe visibility higher than 15% are fundamental for the proper image reconstruction. Even though reconstructed imagens are low spatial resolution (140–340 μm), they presented higher signal-to-noise ratio (SNR 85–700) than conventional images (SNR 14–39), due to the Fourier analyses procedure. Scattering could not resolve the individual grains structures, nevertheless, they revealed the presence of structures smaller than the image system resolution. Finally, it is presented that combining attenuation and scattering signals allows differentiation among the different grain sizes in granular materials and different liquids that appears similar in conventional images.

DLSIA: Deep Learning for Scientific Image Analysis.

DLSIA (Deep Learning for Scientific Image Analysis) is a Python-based machine learning library that empowers scientists and researchers across diverse scientific domains with a range of customizable convolutional neural network (CNN) architectures for a wide variety of tasks in image analysis to be used in downstream data processing. DLSIA features easy-to-use architectures, such as autoencoders, tunable U-Nets and parameter-lean mixed-scale dense networks (MSDNets). Additionally, this article introduces sparse mixed-scale networks (SMSNets), generated using random graphs, sparse connections and dilated convolutions connecting different length scales. For verification, several DLSIA-instantiated networks and training scripts are employed in multiple applications, including inpainting for X-ray scattering data using U-Nets and MSDNets, segmenting 3D fibers in X-ray tomographic reconstructions of concrete using an ensemble of SMSNets, and leveraging autoencoder latent spaces for data compression and clustering. As experimental data continue to grow in scale and complexity, DLSIA provides accessible CNN construction and abstracts CNN complexities, allowing scientists to tailor their machine learning approaches, accelerate discoveries, foster interdisciplinary collaboration and advance research in scientific image analysis.

The study of optical properties of single soot aggregate using three-dimension soft X-ray tomographic reconstruction

The linear depolarization ratio (LDR) is a key parameter used in remote sensing to distinguish particle types. The LDR of the soot fractal model is uneasy to match the measured values. In this paper we used three-dimensional (3D) synchrotron soft X-ray tomography (SXT) for the 3D reconstruction of the toluene-burning soot particles and used discrete dipole approximation to calculate optical properties. The obtained results showed the 3D reconstructed soot particles can efficiently reproduce the measured LDR. The results of this study will encourage us to develop a new morphology model, completely different from the traditional fractal model, for bare soot based on 3D reconstruction. This study is helpful for the understanding and explanation of LDR of bare soot particles.

Sparse-view synchrotron X-ray tomographic reconstruction with learning-based sinogram synthesis.

Synchrotron radiation can be used as a light source in X-ray microscopy to acquire a high-resolution image of a microscale object for tomography. However, numerous projections must be captured for a high-quality tomographic image to be reconstructed; thus, image acquisition is time consuming. Such dense imaging is not only expensive and time consuming but also results in the target receiving a large dose of radiation. To resolve these problems, sparse acquisition techniques have been proposed; however, the generated images often have many artefacts and are noisy. In this study, a deep-learning-based approach is proposed for the tomographic reconstruction of sparse-view projections that are acquired with a synchrotron light source; this approach proceeds as follows. A convolutional neural network (CNN) is used to first interpolate sparse X-ray projections and then synthesize a sufficiently large set of images to produce a sinogram. After the sinogram is constructed, a second CNN is used for error correction. In experiments, this method successfully produced high-quality tomography images from sparse-view projections for two data sets comprising Drosophila and mouse tomography images. However, the initial results for the smaller mouse data set were poor; therefore, transfer learning was used to apply the Drosophila model to the mouse data set, greatly improving the quality of the reconstructed sinogram. The method could be used to achieve high-quality tomography while reducing the radiation dose to imaging subjects and the imaging time and cost.

A 3D/2D registration method based on reconstruction of orthogonal-view Xray images

To establish a 3D/2D registration method for preoperative CT and intra-operative X-ray images in imageguided spine surgery. We propose a 3D/2D registration algorithm based on 3D image reconstruction. The algorithm performs 3D image reconstruction of 2D orthogonal view X-ray images, thus converting the problem into 3D/3D registration. By constructing an end-to-end framework that combines the two tasks of reconstruction and registration, the geodesic distance is measured in the 3D manifold space to complete the registration. We conducted experiments on the public dataset CTSpine1k. The tests on two test sets with different initial registration errors showed that for data with small initial errors, the proposed algorithm achieved a rotation estimation error of 0.115±0.095° and a translation estimation error of 0.144±0.124 mm; for data with larger initial errors, a rotation estimation error of 0.792±0.659° and a translation estimation error of 0.867±0.701 mm were achieved. The proposed method can achieve robust and accurate 3D/2D registration at a speed that meets real-time requirements to improve the performance of spine surgery navigation.

X-ray Tomographic Reconstruction for Understanding Lignin-Induced Changes in the Microstructure and Properties of Polyurethane Insulation Foam

X-ray Tomographic Reconstruction for Understanding Lignin-Induced Changes in the Microstructure and Properties of Polyurethane Insulation Foam

Seeing the whole picture: Methods for getting the most from micro X-ray computed tomography of TRISO nuclear fuel particles

Tristructural isotropic (TRISO) coated fuel particles are a nuclear fuel form under extensive study for use in advanced nuclear reactor concepts. TRISO fuels are subjected to high temperature neutron irradiations and then examined to assess their performance by determining fission product retention and studying morphological changes. Micro X-ray computed tomography is one method of nondestructively studying the effects of TRISO performance. This work addresses the need for image processing to remove X-ray tomographic reconstruction artifacts that prevent the study of TRISO features, as the TRISO particles’ high Z kernel can introduce metal artifacts that degrade the image quality in the surrounding low Z coating layers. These metal artifacts were reduced by imaging the TRISO particles with both high- and low-energy X-rays and applying a mask to the radiographs obtained with low-energy X-rays to digitally remove the dense fuel kernel region. These masked radiographs were then used to produce a tomographic reconstruction which was combined with the tomographic reconstruction of the high-energy data. This enabled the relatively-low-density TRISO buffer layer to be examined in more detail, providing information on irradiation induced dimensional changes of the coatings. This methodology, which helps see the full picture of a TRISO particle, is not limited to nuclear fuels but can be applied to systems that contain highly attenuating material surrounded by less dense materials.

X-ray Tomography Reconstruction Accelerated on FPGA through High-Level Synthesis Tools.

Model-Based Iterative Reconstruction (MBIR) algorithms iteratively use expensive computational operators of forward and backward projections. The irregular memory access pattern of these operators makes them a memory-bound application. Their computation time must be reduced to meet clinical routine constraints. This paper proposes a hardware accelerator architecture based on Field Programmable Gate Arrays (FPGAs) through high-level language, as an alternative to GPU architecture. This acceleration is based on an offline memory access analysis to address the main bottleneck of the algorithm and maximize the data reuse rate. The offline analysis allows for the tuning of the architecture parameters so that they converge to an optimal solution. Then, the Berkeley Roofline model guides our optimization steps by iteratively analyzing the design performance. Our design flow significantly improved the algorithm's computational intensity and overcame the memory bottleneck. Thus, our architecture takes advantage of the FPGA local memory to achieve significant memory bandwidth and efficiently harness the pipeline without stalling the computation. Furthermore, we present the scaling-up strategy from mid-range FPGA to high-end FPGA and any concerns of portability. We used two Intel FPGA devices to implement the algorithm, and then we compared the results with our GPU implementation in terms of speedup and energy efficiency. Our experimental results show that our design has achieved better computational throughput than the works on FPGA architectures reported in the literature.

A novel iterative algorithm to improve segmentations with deep convolutional neural networks trained with synthetic X-ray computed tomography data (i.S.Sy.Da.T.A)

We propose a novel iterative segmentation algorithm (i.S.Sy.Da.T.A: Iterative Segmentation Synthetic Data Training Algorithm) employing Deep Convolutional Neural Networks and synthetic training data for X-ray tomographic reconstructions of complex microstructures. In our method, we reinforce the synthetic training data with experimental XCT datasets that were automatically segmented in the previous iteration. This strategy produces better segmentations in successive iterations. We test our algorithm with experimental XCT reconstructions of a 6-phase Al-Si Matrix Composite reinforced with ceramic fibers and particles. We perform the analysis in 3D with a special network architecture that demonstrates good generalization with synthetic training data. We show that our iterative algorithm returns better segmentations compared to the standard single training approach. More specifically, phases possessing similar attenuation coefficients can be better segmented: for Al2O3 fibers, SiC particles, and Intermetallics, we see an increase of the Dice score with respect to the classic approach: from 0.49 to 0.54, from 0.66 to 0.72, and from 0.55 to 0.66 respectively. Furthermore, the overall Dice score increases from 0.77 to 0.79. The methods presented in this work are also applicable to other materials and imaging techniques.

CBCT Images to an STL Model: Exploring the "Critical Factors" to Binarization Thresholds in STL Data Creation.

In-house fabrication of three-dimensional (3D) models for medical use has become easier in recent years. Cone beam computed tomography (CBCT) images are increasingly used as source data for fabricating osseous 3D models. The creation of a 3D CAD model begins with the segmentation of hard and soft tissues of the DICOM images and the creation of an STL model; however, it can be difficult to determine the binarization threshold in CBCT images. In this study, how the different CBCT scanning and imaging conditions of two different CBCT scanners affect the determination of the binarization threshold was evaluated. The key to efficient STL creation through voxel intensity distribution analysis was then explored. It was found that determination of the binarization threshold is easy for image datasets with a large number of voxels, sharp peak shapes, and narrow intensity distributions. Although the intensity distribution of voxels varied greatly among the image datasets, it was difficult to find correlations between different X-ray tube currents or image reconstruction filters that explained the differences. The objective observation of voxel intensity distribution may contribute to the determination of the binarization threshold for 3D model creation.

Machine learning for detection of 3D features using sparse x-ray tomographic reconstruction.

In many inertial confinement fusion (ICF) experiments, the neutron yield and other parameters cannot be completely accounted for with one and two dimensional models. This discrepancy suggests that there are three dimensional effects that may be significant. Sources of these effects include defects in the shells and defects in shell interfaces, the fill tube of the capsule, and the joint feature in double shell targets. Due to their ability to penetrate materials, x rays are used to capture the internal structure of objects. Methods such as computational tomography use x-ray radiographs from hundreds of projections, in order to reconstruct a three dimensional model of the object. In experimental environments, such as the National Ignition Facility and Omega-60, the availability of these views is scarce, and in many cases only consists of a single line of sight. Mathematical reconstruction of a 3D object from sparse views is an ill-posed inverse problem. These types of problems are typically solved by utilizing prior information. Neural networks have been used for the task of 3D reconstruction as they are capable of encoding and leveraging this prior information. We utilize half a dozen, different convolutional neural networks to produce different 3D representations of ICF implosions from the experimental data. Deep supervision is utilized to train a neural network to produce high-resolution reconstructions. These representations are used to track 3D features of the capsules, such as the ablator, inner shell, and the joint between shell hemispheres. Machine learning, supplemented by different priors, is a promising method for 3D reconstructions in ICF and x-ray radiography, in general.

Fast algorithms based on Empirical Interpolation Methods for selecting best projections in Sparse-View X-ray Computed Tomography using a priori information

To ensure the safety of critical parts in mass production, non-destructive techniques are increasingly considered. Industrial Computed Tomography (CT) is a unique tool to inspect a part integrity and dimensional conformity. However, it usually requires many radiographies (also called projections) to reconstruct a 3D CT image accurately. To reconstruct an object from a reduced number of projections has become one of the most crucial challenges in X-ray CT. Iterative reconstruction algorithms have relieved robotic CT systems from usual circular or helical trajectories, allowing them to focus on the most informative and relevant projections. In this paper, we propose an approach to select the best projections for X-ray tomographic reconstructions when a priori information on the inspected object is available. The presented algorithms are based on Empirical Interpolation Methods and show a good reconstruction quality from a dozen of projections. Those methods have been selected for their flexibility, speed, and the perspectives they open up for sinogram inpainting.

Looking under the skin: multi-scale CT scanning of a peculiarly constructed cornett in the Rijksmuseum

Covered tightly by a thin leather skin, three early seventeenth-century cornetts from the collection of the Rijksmuseum were examined with the focus on their construction and manufacturing. One cornett of the three unexpectedly turned out to have a peculiar construction and to be made out of two sections of different wood species. The question arose whether this could be original or is the result of an extensive restoration.As the internal structure is not accessible for analysis and examination, multi-scale Computed Tomography (CT) scanning was employed to identify the exact regions of interest (ROI) and subsequently perform scans at a sufficiently high resolution in those areas. 3D images of the hollow spaces such as the tunnelling structure caused by the common furniture beetle (Anobium punctatum) criss-crossing the wood species could be computed from the 3D x-ray tomography reconstruction. This allowed to place the occurrence of the insect infestation after the joining of the two sections. Fine tool-marks, signs of construction and potential indications of earlier treatments could be visualized. These results were compared with the other two instruments of the same group and cross-referenced to instruments in other collections, in an attempt to answer questions about the instrument’s authenticity and originality. While the unusual construction out of two wood species might be the result of an extensive repair, another possible hypothesis—based on the combination of the results—is that this unique choice of original manufacturing was intentional, possibly to avoid splitting of the wood when inserting the mouthpiece or to counteract undesired vibrations when played.

Measurement of Transport Properties of Woody Biomass Feedstock Particles Before and After Pyrolysis by Numerical Analysis of X-Ray Tomographic Reconstructions

Lignocellulosic biomass has a complex, species-specific microstructure that governs heat and mass transport during conversion processes. A quantitative understanding of the evolution of pore size and structure is critical to optimize conversion processes for biofuel and bio-based chemical production. Further, improving our understanding of the microstructure of biochar coproduct will accelerate development of its myriad applications. This work quantitatively compares the microstructural features and the anisotropic permeabilities of two woody feedstocks, red oak and Douglas fir, using X-ray computed tomography (XCT) before and after the feedstocks are subjected to pyrolysis. Quantitative analysis of the three-dimensional (3D) reconstructions allows for direct calculations of void fractions, pore size distributions and tortuosity factors. Next, 3D images are imported into an immersed boundary based finite volume solver to simulate gas flow through the porous structure and to directly calculate the principal permeabilities along longitudinal, radial, and tangential directions. The permeabilities of native biomass are seen to differ by three to four orders of magnitude in the different principal directions, but we find that this anisotropy is substantially reduced in the biochar formed during pyrolysis. The quantitative transport properties reported here enhance the ability of pyrolysis simulations to account for feedstock-specific effects and thereby provide a useful touchstone for the biorefining community.

X-Ray Fine Structure of a Limb Solar Flare Revealed by Insight-HXMT, RHESSI and Fermi

Abstract We conduct a detailed analysis of an M1.3 limb flare occurring on 2017 July 3, which have the X-ray observations recorded by multiple hard X-ray telescopes, including Hard X-ray Modulation Telescope (Insight-HXMT), Ramaty High Energy Solar Spectroscopic Imager (RHESSI), and the Fermi Gamma-ray Space Telescope (Fermi). Joint analysis has also used the extreme ultraviolet (EUV) imaging data from the Atmospheric Imaging Assembly (AIA) aboard the Solar Dynamic Observatory. The hard X-ray spectral and imaging evolution suggest a lower corona source, and the non-thermal broken power law distribution has a rather low break energy ∼15 keV. The EUV imaging shows a rather stable plasma configuration before the hard X-ray peak phase, and accompanied by a filament eruption during the hard X-ray flare peak phase. Hard X-ray image reconstruction from RHESSI data only shows one foot point source. We also determined the DEM for the peak phase by SDO/AIA data. The integrated EM beyond 10 MK at foot point onset after the peak phase, while the >10 MK source around reconnection site began to fade. The evolution of EM and hard X-ray source supports lower corona plasma heating after non-thermal energy dissipation. The combination of hard X-ray spectra and images during the limb flare provides the understanding on the interchange of non-thermal and thermal energies, and relation between lower corona heating and the upper corona instability.

3D magnetic configuration of ferrimagnetic multilayers with competing interactions visualized by soft X-ray vector tomography

Full control of magnetic properties in exchange coupled systems requires a good understanding of 3D magnetic configuration with lateral and in-depth resolution. Here we show results from a soft X-ray tomographic reconstruction which allow determining, solely from the experimental data, a detailed description of the vector magnetic configuration of a ferrimagnetic Gd12Co88/Nd17Co83/Gd24Co76 trilayer with engineered competing anisotropy, exchange and magnetostatic interactions at different depths. The trilayer displays chevron patterns with a distorted closure structure. Near the top Gd24Co76 layer, local exchange springs with out-of-plane magnetization reversal, quasi-domains with ripple-like patterns and magnetic vortices and antivortices across the thickness are observed. The detailed analysis of the magnetic tomogram shows that the effective strength of the exchange spring at the NdCo/GdCo interface can be finely tuned by GdxCo1-x composition and anisotropy (determined by sample fabrication) and in-plane stripe orientation (adjustable), demonstrating the suitability of 3D magnetic visualization techniques in magnetic engineering research.

Advantage of Machine Learning over Maximum Likelihood in Limited-Angle Low-Photon X-Ray Tomography

Limited-angle X-ray tomography reconstruction is an ill-conditioned inverse problem in general. Especially when the projection angles are limited and the measurements are taken in a photon-limited condition, reconstructions from classical algorithms such as filtered backprojection may lose fidelity and acquire artifacts due to the missing-cone problem. To obtain satisfactory reconstruction results, prior assumptions, such as total variation minimization and nonlocal image similarity, are usually incorporated within the reconstruction algorithm. In this work, we introduce deep neural networks to determine and apply a prior distribution in the reconstruction process. Our neural networks learn the prior directly from synthetic training samples. The neural nets thus obtain a prior distribution that is specific to the class of objects we are interested in reconstructing. In particular, we used deep generative models with 3D convolutional layers and 3D attention layers which are trained on 3D synthetic integrated circuit (IC) data from a model dubbed CircuitFaker. We demonstrate that, when the projection angles and photon budgets are limited, the priors from our deep generative models can dramatically improve the IC reconstruction quality on synthetic data compared with maximum likelihood estimation. Training the deep generative models with synthetic IC data from CircuitFaker illustrates the capabilities of the learned prior from machine learning. We expect that if the process were reproduced with experimental data, the advantage of the machine learning would persist. The advantages of machine learning in limited angle X-ray tomography may further enable applications in low-photon nanoscale imaging.

Accelerated Alternating Minimization for X-ray Tomographic Reconstruction

Accelerated Alternating Minimization for X-ray Tomographic Reconstruction