This PDF file contains the front matter associated with SPIE Proceedings Volume 10948, including the Title Page, Copyright information, Table of Contents, Author and Conference Committee lists.

Contrast-enhanced spectral mammography (CESM) is being implemented to overcome the limitations of conventional mammography where tumor visualization is obstructed by overlapping glandular tissue. CESM exploits the spectral properties of a contrast agent by subtracting two images one obtained above and other below the K-edge energy. The most common approach requires dual-exposure where two images are obtained with differ- ent incident spectra. However, this comes at the expense of increased patient dose and susceptibility to motion artifacts. We propose the use of photon counting spectral detectors to simultaneously obtain multiple images with single-exposure. This is demonstrated using a wide area CdTe Medipix3RX detector to acquire images of iodine contrast agent in an anthropomorphic breast imaging phantom. The electronic thresholds in the detector replace the traditional physical filters. Our results show single-exposure CESM for the detection of iodine with concentrations as low as 2.5 mg/mL of a 10 mm diameter target in a 5 cm thick heterogeneous background. These results demonstrate the viability of photon counting detectors for low dose contrast-enhanced mammography.

A numerical study of the wide angle x-ray scatter (WAXS) energy integrated signals (EISs) from breast biopsies was conducted. A benign biopsy was chosen as fibroglandular (fib) tissue whereas biopsies with cancer were approximated as consisting of fib tissue and a cluster of epithelial cells. The grouping of cells represented the malignant portion. The EISs due to scatter were computed for biopsies of thickness d_bio = 2 mm, 5 mm, 10 mm and 20 mm. For the malignant biopsies, the fractional volumes of cells (ν_cell) studied ranged from 0.01 to 0.1. Incident 2 mm diam beams of 30 kV, 50 kV, 80 kV, and 140 kV were considered. The tube current and exposure time were 3 mA and 1 minute, respectively. The WAXS signals were computed by adding the signals from annular detectors subtending scattering angles θ = 2°, 3°, , 23° and solid angles Ωθ = 2.0 x π x [cos(θ - ∆θ/2) - cos(θ + ∆θ/2)] x cos θ where ∆θ = 1°. Let the EIS due to a malignant biopsy be EIS^m_s and that of a benign one, EIS^b_s. With these signals, values of SNR were computed. The 30 kV beam provided SNRs < 5 for the lowest entrance exposure, X = 0.0015 C/kg. For biopsies with d_bio = 2, 5, 10, 20 mm and ν_cell = 0.1, the SNRs were 28.0, 38.2, 43.0, and 40.0. The findings suggest that there is potential to use WAXS EISs to diagnose malignancy in breast biopsies.

Digital breast tomosynthesis is rapidly replacing digital mammography as the basic x-ray technique for evaluation of the breasts. However, the sparse sampling and limited angular range gives rise to different artifacts, which manufacturers try to solve in several ways. In this study we propose an extension of the Learned Primal- Dual algorithm for digital breast tomosynthesis. The Learned Primal-Dual algorithm is a deep neural network consisting of several ‘reconstruction blocks’, which take in raw sinogram data as the initial input, perform a forward and a backward pass by taking projections and back-projections, and use a convolutional neural network to produce an intermediate reconstruction result which is then improved further by the successive reconstruction block. We extend the architecture by providing breast thickness measurements as a mask to the neural network and allow it to learn how to use this thickness mask. We have trained the algorithm on digital phantoms and the corresponding noise-free/noisy projections, and then tested the algorithm on digital phantoms for varying level of noise. Reconstruction performance of the algorithms was compared visually, using MSE loss and Structural Similarity Index. Results indicate that the proposed algorithm outperforms the baseline iterative reconstruction algorithm in terms of reconstruction quality for both breast edges and internal structures and is robust to noise.

Purpose: The purpose of this study is to apply targeted Parametric Imaging on aneurysms to quantitatively investigate contrast flow changes at pre-, post-treatment and follow-up with outcome scoring. Methods: The angiograms for 50 patients were acquired, 25 treated with coil embolization and 25 treated using a flow diverter. API was performed by synthesizing the time density curve (TDC) at every pixel. Based on the TDCs, we calculated various parameters for the quantitative characterization of contrast flow through the vascular network and aneurysms and displayed them using color encoded maps. The parameters included were : Time to Peak (TTP), Mean Transit Time (MTT), Time of Arrival (TTA), Peak Height (PH) and Area Under the Curve (AUC). Two Regions of Interest (ROI) were manually marked over the aneurysm dome and the main artery. Average aneurysm parameter values were normalized to those values recorded in the main artery and recorded pre-/post-treatment and follow-up and compared to Raymond Roy scores and flow diverter stent scoring. Results: The normalized mean values were as follows (pre and post treatment): TTP (1.09+/-0.14, 1.55+/-1.36), MTT (1.07+/-0.23, 1.27+/-0.42), TTA (0.14+/-0.15, 0.26+/-0.23), PH (1.2+/-0.54, 0.95+/-0.83) and AUC (1.29+/-0.69, 1.44+/- 1.92). The neural network gave a validation accuracy of 0.8036 with a loss of 0.0927. A receiver operating characteristic curve with an AUC of 0.866 was obtained. Conclusions: API can quantitatively describe the flow in the aneurysm for initial investigation of the radiomics of intracranial aneurysms. It also shows a clear demarcation between pre and post treatment. Statistical modelling and a machine learning network is used to prove the success of our model.

This study modeled a framework for virtual human liver phantoms, focusing primarily on the intricate vascular networks that comprise the liver. Large vasculature was segmented from clinical liver perfusion images to ascertain a general starting point for the vascular networks of the liver that would be common among a healthy population. Clinical imaging methods cannot currently resolve the vast majority of the vasculature of the liver, and at the limiting resolution, modeling techniques continued the structure of the existing vasculature according to empirically known properties of blood vessel formation. Such advances in virtual phantom modeling enable simulation work in CT liver imaging, as clinical CT liver imaging is not ideally performed without contrast and multi-phasic acquisitions taking place over the course of the contrast's perfusion. The total amount of contrast in each organ in the body as a function of time is known from prior work, and the complete vascular network of the liver allows this information to be translated into an organ-specific contrast-concentration as a function of time. The ability to simulate this physiology is necessary for liver perfusion imaging, as pathologies typically impede or otherwise alter healthy perfusion patterns. The perfusion simulated here was in good agreement with known patterns of perfusion. Thus, virtual clinical trials can be performed with a dynamic model of the liver containing a fully integrated and realistic vascular network.

Purpose. Investigate synthetic mammography approaches for carbon nanotube (CNT)-enabled stationary digital breast tomosynthesis (sDBT). Methods. Projection images of breast-mimicking phantoms containing soft-tissue masses and microcalcification clusters collected by sDBT were used to develop weighted-intensity forward-projection algorithms that generated a synthetic mammogram from the reconstructed 3D image space. Reconstruction was accomplished by an adapted fan-volume modification of the simultaneous iterative reconstruction technique. Detectability indices were used to quantify mass and calcification visibility. The image processing chain was then applied to projection views collected by sDBT on women with “suspicious” breast lesions detected by standard screening 2D digital mammography. Results. Quantifying detectability allowed correlation between the visibility of clinically-important image features and the order of the polynomial weighting function used during forward projection. The range of weighted functions exists between the extremes of an average-intensity projection (zero-order) and maximum-intensity projection (infinite-order), with lower order weights emphasizing soft-tissue features and higher-order weights emphasizing calcifications. Application of these algorithms to patient images collected by sDBT, coupled with dense-artifact reduction and background equalization steps, produced synthetic mammograms on which additional post-processing approaches can be explored, with the actual mammogram providing a reference for comparison. Conclusions. An image-processing chain with adjustable weighting during forward projection, dense-artifact reduction, and background equalization can yield a range of sDBT-based synthetic mammograms that display clinically-important features differently, potentially improving the ability to appreciate the association of masses and calcifications.

On-board 4D cone-beam CT (CBCT) imaging using a linear accelerator (LINAC) is recently favored scanning protocol in image guided radiotherapy, but it raises the problems of excessive radiation dose. Alternatively, the 4D digital tomosynthesis (DTS) has been introduced for small-sized target localization, such as pancreas, prostate, or partial breast scan, which does not require a full 3D information. However, conventional filtered back-projection (FBP) reconstruction method produces severe aliasing artifacts due to sparsely sampled projections measured in each respiration phase within a limited angle range. This effect is even more severe when the LINAC gantry sweep speed is too fast to sufficiently cover the respiratory gating phase. Previous studies on total-variation (TV) minimization-based reconstruction framework would be an alternative approach to solve this problem, however, it presents the loss of sharpness during the iterations. In this study, we adopted an adaptively-weighted TV (AwTV) scheme which penalizes the images after the TV optimization. We introduced a look-up table which contains all possible weighting parameters during each iteration step. As a result, the proposed AwTV method provided better image quality compared to the conventional FBP and adaptive steepest descent-projection onto convex set (ASD-POCS) frameworks, showing higher structural similarities (SSIM) by factor of 1.12 compared to FBP and less root-mean-square error (RMSEs) by factor of 1.06 compared to ASD-POCS. The horizontal line profile of the spherical target inserted in the moving phantom showed that the images from FBP and ASD-POCS provided severe aliasing artifact and smoothed pixel intensities, but proposed AwTV scheme reduces the aliasing artifact while maintaining the object’s sharpness. In conclusion, the proposed AwTV method has a potential for low-dose and faster 4D-DTS imaging, which indicates an alternative option to 4D-CBCT for small region target localization.

We propose a new metal artifact correction method for the X-ray digital tomosynthesis by accurately detecting metal in the projection data. We combined 3D region growing for growing a few points in the metal to other projection angles and 2D region growing for growing the points further in order not to force the user to set the starting points at each projection angle. We compared the proposed method with the conventional FBP. In the phantom experiment with a mimicked artificial joint using the proposed method, the metal artifacts around the metal object were reduced. At the distance within 5 mm from the metal object, the root mean square errors evaluated with the conventional and proposed methods were 2700 and 200, respectively, and the root mean square errors improvement of more than 90% was demonstrated. When the distance from the metal object was shorter, the metal artifact became more significant in the conventional method, and the effectiveness of the proposed metal artifact correction was higher.

Aim: The impact of x-ray system parameters on detectability of specific (clinical) signals can be studied with simulation platforms if these tools are sufficiently accurate and realistic. This work describes the steps taken to verify and confirm the accuracy of a local platform developed for the use in virtual clinical trials of breast tomosynthesis. The (gold standard) reference data will be made available to the community. Materials and methods: Our simulation platform simulates specific targets, including microcalcifications into existing 2D FFDM and DBT background images, a method called partial simulation. There are three steps: (1) creation of a voxel model or 3D analytical object to be inserted into the ‘For Processing’ projections; (2) generation of a realistic object template for the geometry under study and the relevant resolution, scatter and noise properties; (3) insertion of the target into the projections and DBT reconstruction plus image processing. Three objects were simulated as part of the verification: a small high contrast 0.5 mm aluminum (Al) sphere in a poly(methyl methacrylate) (PMMA) stack, a 0.2 mm thick Al sheet in a PMMA stack and a 0.8 mm steel edge. For the small Al sphere, the peak contrast, the signal difference to noise ratio (SDNR), the profile in the (in plane) xy-direction and the artifact spread function (ASF) were compared to results from real acquisitions. Contrast and SDNR were compared to data from a real 0.2 mm Al sheet. Sharpness modelling was verified by comparing the modulation transfer function (MTF) calculated from real and simulated edges. The study was performed for a Siemens Inspiration DBT system. Results: Comparing peak contrast and SDNR for both sphere and sheet showed good agreement (<5% error) in 2D FFDM and DBT. The similarity of the pixel value profiles through the sphere and the sheet in the xy-direction and the ASF for real and simulated Al spheres confirmed accurate geometric modelling. Absolute and relative average deviation between MTF measured from real and simulated edge in the front-back and left-right directions show a good correlation for frequencies up to the Nyquist frequency for 2D FFDM and DBT mode. Real and simulated objects could not be differentiated visually. Conclusion: The close correspondence between simulated and real objects, both visually and quantitatively, indicates that this simulation framework is a strong candidate for use in virtual clinical studies employing 2D FFDM and DBT.

In digital breast tomosynthesis (DBT), projection images are acquired as the x-ray tube rotates in the plane of the chest wall. We constructed a prototype next-generation tomosynthesis (NGT) system that has an additional component of tube motion in the perpendicular direction (i.e., posteroanterior motion). Our previous work demonstrated the advantages of the NGT system using the Defrise phantom. The reconstruction shows higher contrast and fewer blurring artifacts. To expand upon that work, this paper analyzes how image quality can be further improved by customizing the motion path of the x-ray tube based on the object being imaged. In simulations, phantoms are created with realistic 3D breast outlines based on an established model of the breast under compression. The phantoms are given an internal structure similar to a Defrise phantom. Two tissue types (fibroglandular and adipose) are arranged in a square-wave pattern. The reconstruction is analyzed as a binary classification task using thresholding to segment the two tissue types. At various thresholds, the classification of each voxel in the reconstruction is compared against the phantom, and receiver operating characteristic (ROC) curves are calculated. It is shown that the area under the ROC curve (AUC) is dependent on the x-ray tube trajectory. The trajectory that maximizes AUC differs between phantoms. In conclusion, this paper demonstrates that the acquisition geometry in DBT should be personalized to the object being imaged in order to optimize the image quality.

In this work, we investigated and measured the noise in Digital Breast Tomosynthesis (DBT) slices considering the back-projection (BP) algorithm for image reconstruction. First, we presented our open-source DBT reconstruction toolbox and validated with a freely available virtual clinical trials (VCT) software, comparing our results with the reconstruction toolbox available at the Food and Drug Administration's (FDA) repository. A virtual anthropomorphic breast phantom was generated in the VCT environment and noise-free DBT projections were simulated. Slices were reconstructed by both toolboxes and objective metrics were measured to evaluate the performance of our in-house reconstruction software. For the noise analysis, commercial DBT systems from two vendors were used to obtain x-ray projections of a uniform polymethyl methacrylate (PMMA) physical phantom. One system featured an indirect thallium activated cesium iodide (CsI(TI)) scintillator detector and the other a direct amorphous selenium (a-Se) detector. Our in-house software was used to reconstruct raw projections into tomographic slices, and the mean pixel value, noise variance, signal-to-noise ratio (SNR) and the normalized noise power spectrum (NNPS) were measured. In addition, we investigated the adequacy of a heteroskedastic Gaussian model, with an affine variance function, to describe the noise in the reconstruction domain. The measurements show that the variance and SNR from reconstructed slices report similar spatial and signal dependency from previously reported in the projection domain. NNPS showed that the reconstruction process correlates the noise of the DBT slices in the case of projections degraded with almost uncorrelated noise.

The statistics of photon counting detectors (PCDs) differ in several aspects from the statistics of energy integrating detectors (EIDs). Particularly, the effect of crosstalk in a PCD involves a 0/1 decision: a photon may be counted in a neighboring pixel or not whereas in an EID the neighboring pixel may just receive a fraction of the signal. Another interesting effect is that, especially for high counting thresholds, there exists a zone at the edge of the pixel where absorbed x-ray energy will not produce any signal. This may lead to a modulation transfer function (MTF) exceeding the theoretical limit given by the nominal pixel aperture. This fact has also been observed in measurements. Goal of this work is to present a simple but comprehensive description of PCD detectors in the low flux limit capable of including all relevant effects. The model presented is based on a Monte Carlo simulation of the x-ray energy deposition in the detector and a simple model of the charge cloud propagation. A reformulation of the probability generating function formalism allows calculating all relevant quantities like mean signal values or covariances between thresholds and/or neighboring pixels or the MTF and DQE as a function of input photon energy directly from the Monte Carlo simulation.

Neuro-endovascular image-guided interventions (EIGIs) are aided by use of detectors with improved spatial resolution. A new detector is capable of switching between standard resolution, flat-panel detector (FPD) zoom and high-definition (Hi-Def) zoom modes, with 194 and 76 μm pixels respectively. The relative performance of the two zoom modes to image specific, high-resolution objects was quantitatively investigated. Detector DQEs were measured for both zoom modes and used to determine the Relative Object Detectability (ROD), which compares two imaging detectors’ relative abilities to image a simulated object by integrating the DQE of one detector with the square of the Fourier transform of the simulated object function and dividing the result by a similar integral for a second reference detector. Initial evaluations used a pre-whitened matched filter (PWMF) ideal-observer model. Comparisons were also made using the generalized-ROD (G-ROD) which uses the generalized-DQE (GDQE) that includes the effects of clinically relevant parameters such as magnification, focal-spot size, and scatter and the generalized-measured-ROD (GM-ROD) which uses the square of the Fourier transform of the actual images of the object acquired with the detectors. Each of the metrics demonstrated improved performance of the Hi-Def zoom mode over the standard FPD mode in imaging a wide array of objects such as stents, wires, and spheres. Of particular note is the greater performance of the Hi-Def when considering the high spatial frequencies necessary for visualizing fine image details of a pipeline stent. These initial investigations demonstrate the great potential of the Hi-Def zoom mode during neuro-interventions.

Complementary metal-oxide-semiconductors (CMOS) flat panel detectors (FPD) have steadily gained acceptance into medical imaging applications1-15. Selecting the proper detector technology for the imaging task requires optimization to balance the cost and the image quality. To facilitate this, fundamental detector performance of CMOS and a-Si panels were evaluated using the following quantitative imaging metrics: X-ray sensitivity, Noise Equivalent Dose (NED,) Noise Power Spectrum (NPS), Modulation Transfer Function (MTF), and Detective Quantum Efficiency (DQE). Imaging task measurements involved high-contrast and low-contrast resolution assessment. Varex FPDs evaluated for this study included: CMOS 3131 (150 μm pixel), a-Si 3030X (194 μm pixel), a-Si XRpad2 3025 (100 μm) and CMOS 2020 (100 μm pixel). Performance comparisons were organized by pixel size: large pixels, 150 μm CMOS and 194 μm a-Si, and small pixels, 100 μm in a-Si and CMOS technology. The results showed high dose DQE of the a-Si 3030X was about 10% higher than the CMOS 3131 between 0 - 1.8 cycles/mm, while beyond 1.8 cycles/mm, the CMOS performed better. The 3030X low dose DQE was higher than the 3131 between 0-1.3 cycles/mm, while the CMOS performance was higher beyond 1.3 cycles/mm. The high dose DQE of 100 μm a-Si was higher than the 100 μm CMOS for all frequencies. However, the low dose DQE of 100 μm CMOS was higher beyond 0.6 cycles/mm, while the 100 μm a-Si pixel had higher DQE only between 0 – 0.6 cycles/mm. Large pixel image quality (IQ) assessment favored a-Si pixel with 7% higher Contrast-to-Noise-Ratio (CNR) results for both high and low contrast-detail at 500 nGy. Small pixel CNR favored CMOS with ~38% better high contrast-detail and 12% greater low contrast-detail at ~500 nGy. Through these measurements that combine imaging metrics and image quality, we demonstrated a practical method for selecting the appropriate detector technology based on the requirements of the imaging applications.

In proton therapy treatment, proton residual energy after transmission through the treatment target may be determined by measuring sub-relativistic transmitted proton time-of-flight velocity and hence the residual energy. We have begun developing this method by conducting proton beam tests using Large Area Picosecond Photon Detectors (LAPPDs) which we have been developing for High Energy and Nuclear Physics Applications. LAPPDs are 20cm x 20cm area Micro Channel Plate Photomultiplier Tubes (MCP-PMTs) with millimeter-scale spatial resolution, good quantum efficiency and outstanding timing resolution of ≤70 picoseconds rms for single photoelectrons. We have constructed a time-of-flight telescope using a pair of LAPPDs at 10 cm separation, and have carried out our first tests of this telescope at the Massachusetts General Hospital's Francis Burr Proton Therapy Center. Treatment protons are sub-relativistic, so precise timing resolution can be combined with paired imaging detectors in a compact configuration while still yielding high accuracy in proton residual energy measurements through proton velocity determination from nearly monoenergetic protons. This can be done either for proton bunches or for individual protons. Tests were performed both in "ionization mode" using only the Microchannel Plates to detect the proton bunch structure and also in "photodetection mode" using nanosecond-decay-time quenched plastic scintillators to excite the photocathode within each of the paired LAPPDs. Data acquisition was performed using a remotely operated oscilloscope in our first beam test, and using 5Gsps DRS4 Evaluation Board waveform digitizers in our second test, in each case reading out both ends of single microstrips from among the 30 within an LAPPD. First results for this method and future plans are presented.

In 1963, Shaw applied Fourier analysis to the zero-frequency DQE and developed the frequency-dependent DQE or DQE(k) and made it clear that DQE(k) is applicable to every frequency level within the system bandwidth, including the zero frequency. Over time, especially after entering the modern era of digital x-ray imaging, the experimental measurement methods of DQE(k) (particularly the measurements of the NPS which is an important element in DQE(k)) have evolved, and some measurement methods may generate nonphysical NPS and DQE results at k=0. As a result, an experimental DQE(k) curve is often cut off at certain low frequency above zero. This work presents a new experimental method to deal with two challenges: severe NPS(k=0) underestimation due to polynomial-based background detrending; severe NPS(k=0) overestimation due to the presence of faint but non-negligible system drift. Based on a theoretical analysis of the impact of drift to the measured autocovariance function, the error introduced by drift can be isolated, and corresponding correction can be applied to NPS(k=0). Both numerical simulation with known ground truth and experimental studies demonstrated that the proposed method enables accurate DQE(k=0) measurement.

A deep learning Generative Adversarial Networks (GANs) were developed and validated to provide an accurate way of direct NPS estimation from a single patient CT scan. GANs were utilized to map a white noise input to a CT noise realization with correct CT noise correlations specific to a single local uniform ROI. To achieve this, a two-stage strategy was developed. In the pre-training stage, ensembles of 64x64 MBIR noise-only images of a quality assurance phantom were used as training samples to jointly train the generator and discriminator. They were fined-tuned using training samples from a single 101x101 ROI of an abdominal anthropomorphic phantom. Results from GANs and physical scans were compared in terms of its mean frequency and radial averaged NPS. This workflow was extended to a patient case where reference dose and 25% of reference dose CT scans were provided for fine-tuning. GANs generated noise-only image samples that are indistinguishable from physical measurement. The overall mean frequency discrepancy between NPS generated from GANs and those from physically acquired data was 0.2% for anthropomorphic phantom validation. The KL divergence for 1D radial averaged NPS profile of these two NPS acquisitions was 2.2×10^(-3). Statistical test indicates trained GANs generated equivalent NPS to physical scans. In clinical patient-specific NPS studies, it showed a distinction between the reference dose case and 25% of reference dose case. It was demonstrated the GANs characterized the properties of CT noise in terms of its mean frequency and 1D NPS profile shape.

This work examined the impact of the presampling Modulation Transfer Function (MTF) on detectability of lesion-like targets in digital mammography. Two needle CR plates (CR1 and CR2) with different MTF curves but identical detector response (sensitivity) were selected. The plates were characterized by MTF, normalized noise power spectrum (NNPS) and detective quantum efficiency (DQE). Three image quality phantoms were applied to study the impact of the difference in MTF: first, the CDMAM contrast-detail phantom to give gold thickness threshold (T); second, a 3D structured phantom with lesion models (calcifications and masses), evaluated via a 4-alternative forced-choice study to give threshold diameter (d_tr) and third, a detectability index (d') from a 50 mm PMMA flat field image and an 0.2 mm Al contrast square. MTF coefficient of variation was ~1%, averaged up to 5 mm^-1. At 5 mm^-1, a significant 24% reduction in MTF was observed. The lower MTF caused a 12% reduction in NNPS for CR2 compared to CR1 (at detector air kerma 117 μGy). At 5 mm^-1, there was a drop in DQE of 34% for CR2 compared to CR1. For the test objects, there was a trend to lower detectability for CR2 (lower MTF) for all but one parameter, however none of the changes were significant. The MTF is a sensitive and easily applied means of tracking changes in sharpness before these changes are uncovered using lesion simulating objects in test objects.

A next generation tomosynthesis (NGT) prototype was designed to investigate alternative scanning geometries for digital breast tomosynthesis (DBT). The NGT system uses a 2D plane as an address space for the x-ray source, and onedimensional linear detector motion to determine an acquisition geometry. This design provides myriad acquisition geometries for investigation. This system is also capable of magnification DBT. We performed image quality measurements to evaluate performance of the NGT system for both contact and magnification imaging in 2D and 3D. The modulation transfer function (MTF) was computed using the slanted-edge method to evaluate spatial resolution. The first zero of the MTF was observed to increase by a factor of the magnification. In-plane spatial resolution performance for 3D was measured using an in-house metric, and was found to be commensurate to the MTF. This metric uses a star pattern as an input object to produce the contrast transfer function (CTF). The 2D noise power spectra (NPS) were calculated to evaluate the degradation of image quality due to noise. 3D NPS were also calculated for various 3D image reconstructions. 3D renditions of the NPS show how the NGT can sample a broader range of frequencies in the Fourier domain than conventional DBT. The system’s lag was measured and found not to affect 3D image reconstructions significantly. A wax calcification phantom was constructed and imaged using the NGT system. The performance of this system has been evaluated and the results suggest that image quality is sufficient for clinical investigation.

Model-based reconstruction (MBR) algorithms in CT have demonstrated superior dose-image quality tradeoffs compared to traditional analytical methods. However, the nonlinear and data-dependent nature of these al- gorithms pose significant challenges for performance evaluation and parameter optimization. To address these challenges, this work presents an analysis framework for quantitative and predictive modeling of image proper- ties in general nonlinear MBR algorithms. We propose to characterize the reconstructed appearance of arbitrary stimuli by the generalized system response function that accounts for dependence on the imaging conditions, reconstruction parameters, object, and the stimulus itself (size, contrast, location). We estimate this nonlinear function using a multilayer perceptron neural network by providing input and output pairs that samples the range of imaging parameters of interest. The feasibility of this approach was demonstrated for predicting the appearance of a spiculated lesion reconstructed by a penalized-likelihood objective with a Huber penalty in a physical phantom as a function of its location and reconstruction parameters β and δ. The generalized system response functions predicted from the trained neural network show good agreement with those computed from mean reconstructions, proving the ability of the framework in mapping out the nonlinear function for combinations of imaging parameters not present in the training data. We demonstrated utility of the framework to achieve desirable (e.g., non-blocky) lesion appearance in arbitrary locations in the phantom without the need for performing actual reconstructions. The proposed prediction framework permits efficient and quantifiable performance evaluations to provide robust control and understanding of image properties for general classes of nonlinear MBR algorithms.

Star pattern test objects are used to evaluate the high-contrast performance of imaging systems. These objects were used to investigate alternative scanning geometries for a prototype next-generation tomosynthesis (NGT) system. The NGT system has 2D planar source motion and linear detector motion, and is capable of myriad acquisition geometries. We designed a virtual star pattern with a voxel size of 5𝜇m, and used it to evaluate the spatial resolution performance of the NGT system for three different acquisition geometries. The Open Virtual Clinical Trials (OpenVCT) framework was used to simulate virtual star patterns for acquisition geometries of the NGT system. Simulated x-ray projections of the virtual phantom were used to create super-sampled 3D image reconstructions. Using the same acquisition geometries on the NGT system, a physical star pattern was imaged to create experimental 3D image reconstructions. The simulated and physical data were compared qualitatively by visual inspection, and quantitatively using an in-house metric. This metric computes the Fourier transform radially for one quadrant of the star pattern to discern the limit of spatial resolution (LSR) and the existence of aliasing. The results exhibit the same characteristics in terms of super-resolution and Moiré patterns (arising from aliasing) with visual inspection. The simulated LSR for the 12 conditions analyzed are all within 3% of the physical data. Aliasing was determined to be present in the same simulated image reconstructions as the experimental complements. Super-resolution is observed for two of the three NGT acquisition geometries in the experimental and simulated images.

We have conducted virtual clinical trials (VCTs) of digital breast tomosynthesis (DBT) to evaluate the parameters that affect detectability of breast lesions. The OpenVCT framework was used to simulate the breast anatomy and imaging systems. We generated 36 anthropomorphic breast phantoms (700 ml volume, 6.33 cm compressed thickness), varying the number of simulated tissue compartments and their shape. We inserted 42 calcifications into each phantom with variable sizes of 1-3 voxels. DBT projections of phantoms with and without lesions were synthesized assuming a clinical acquisition geometry. We varied the detector element size (140 μm and 70 μm), the source motion (continuous and stepand- shoot), and the reconstructed voxel size (100 μm and 70 μm). The reconstructed images were cropped in the plane where the calcifications are located, with regions of interest (ROIs) centered on the lesion position. We also simulated virtual readers to evaluate the calcification detectability using multiple-reader, multiple-case method, using Barco’s Medical Virtual Image Chain (MeVIC) software. Human readers were simulated using channelized Hotelling observers with 15 Laguerre-Gauss channels. We used spreads of 22 and 31, and ROIs of 150×150 and 214×214 pixels for images reconstructed with pixel size of 100 μm and 70 μm, respectively. Reconstructed voxels of 70μm provided better overall calcification detection, especially for small calcifications. For one-voxel polycubes, the difference in AUC using five readers was 6.5% (0.713 and 0.667). The impact of calcification detection from most to least significant is: reconstruction voxel size, source motion, and detector element size, especially for small calcifications.

Image processing algorithms based on deep learning techniques are being developed for a wide range of medical applications. Processed medical images are typically evaluated with the same kind of image similarity metrics used for natural scenes, disregarding the medical task for which the images are intended. We propose a com- putational framework to estimate the clinical performance of image processing algorithms using virtual clinical trials. The proposed framework may provide an alternative method for regulatory evaluation of non-linear image processing algorithms. To illustrate this application of virtual clinical trials, we evaluated three algorithms to compute synthetic mammograms from digital breast tomosynthesis (DBT) scans based on convolutional neural networks previously used for denoising low dose computed tomography scans. The inputs to the networks were one or more noisy DBT projections, and the networks were trained to minimize the difference between the output and the corresponding high dose mammogram. DBT and mammography images simulated with the Monte Carlo code MC-GPU using realistic breast phantoms were used for network training and validation. The denoising algorithms were tested in a virtual clinical trial by generating 3000 synthetic mammograms from the public VICTRE dataset of simulated DBT scans. The detectability of a calcification cluster and a spiculated mass present in the images was calculated using an ensemble of 30 computational channelized Hotelling observers. The signal detectability results, which took into account anatomic and image reader variability, showed that the visibility of the mass was not affected by the post-processing algorithm, but that the resulting slight blurring of the images severely impacted the visibility of the calcification cluster. The evaluation of the algorithms using the pixel-based metrics peak signal to noise ratio and structural similarity in image patches was not able to predict the reduction in performance in the detectability of calcifications. These two metrics are computed over the whole image and do not consider any particular task, and might not be adequate to estimate the diagnostic performance of the post-processed images.

Cross-scatter is often the dominant scatter mode in modern dual source CT (DSCT). Like forward scatter (intra source-detector-pair scatter), which is present in all CT systems, cross-scatter (inter source-detector-pair scatter) leads to streak and cupping artifacts. Having recently developed the deep scatter estimation (DSE) to estimate forward scatter in single source CT, we now tested the performance of DSE in such a cross-scatter-dominated DSCT. Given only the total intensity in a projection as input, we trained a deep convolutional neural network to estimate the scatter distribution, which was then subtracted from the total intensity, to obtain scatter-corrected data. The projections used for training and testing were simulated using Monte Carlo methods. Our method estimates cross- and forward scatter simultaneously and in real-time, with a mean error of only 1.7 %. The error of the CT values is reduced from hundreds of HU to a few dozens of HU. Our method can compete with a measurement-based approach, but does not require any additional hardware.

The focal spot size of x-ray tubes as well as the pixel size and scintillator thickness limit the spatial resolution of projection images as they result in blurring and degradation of the system’s point spread function. Deblurring of those images has been a topic of research for several decades. However, it is not solved in general. In this manuscript the application of a convolutional neural network for the deblurring of x-ray projection images is presented and compared to a standard deblurrig technique. The advantages of the neural network in terms of image quality and applicability are demonstrated with simulations and measurements originating from table top and gantry based micro-CT systems.

Reducing the radiation dose of computed tomography (CT) and thereby decreasing the potential risk suffered by the patients is desirable in CT imaging. However, lower dose often results in additional noise and artifacts in reconstructed images that may negatively affect the clinical diagnoses. Recently, many image-domain denoising approaches based on deep learning have been proposed and obtained promising results. However, since reconstructed CT image values are not directly related to noise level, estimating noise level from CT images is not an easy task. In this work, we propose a count-domain denoising approach using a convolutional neural network (CNN) and a filter loss function. Compared with image-domain denoising methods, the proposed count-domain method can easily estimate the noise level in projections based on the measurement in each detector bin. Moreover, because each projection is ramp-filtered before being backprojected to the image-domain, we propose a filter loss function where the training loss is computed using the ramp filtered projection, rather than the original projection. Since the filter loss is closely related to the differences in the image-domain, it further improves the quality of reconstructed CT images.

Image reconstruction from line integrals is one of foundations in computed tomography (CT) for medical diagnosis and non-destructive detection purpose. To accurately recover the density function from measurements taken over straight lines, analytic-formula-based or optimization-based inversions have been discovered over the past several decades. Accurate image reconstruction can be achieved if the acquired dataset satisfies data sufficiency conditions and data consistency conditions. However, if these conditions are violated, accurate image reconstruction remains an intellectual challenge provided that significant a priori information about image object and/or physical process of data acquisition need to be incorporated. In this work, we show that a deep learning method based upon a brand new network architecture, termed intelligent CT neural network (iCT-Net), can be employed to discover accurate image reconstruction solutions from fully-truncated and sparsely-sampled line integrals without explicit incorporations of a priori information of either image object or data acquisition process. After a two-stage training, the trained iCT-Net was directly applied to real human subject data to demonstrate the generalizability of iCT-Net to experimental data.

We developed photoacoustic tomography to peer deep into biological tissue. Photoacoustic tomography (PAT) provides in vivo omniscale functional, metabolic, molecular, and histologic imaging across the scales of organelles through organisms. We also developed compressed ultrafast photography (CUP) to record 10 trillion frames per second, 10 orders of magnitude faster than commercially available camera technologies. CUP can tape the fastest phenomenon in the universe, namely, light propagation, and can be slowed down for slower phenomena such as combustion. PAT physically combines optical and ultrasonic waves. Conventional high-resolution optical imaging of scattering tissue is restricted to depths within the optical diffusion limit (~1 mm in the skin). Taking advantage of the fact that ultrasonic scattering is orders of magnitude weaker than optical scattering per unit path length, PAT beats this limit and provides deep penetration at high ultrasonic resolution and high optical contrast by sensing molecules. Broad applications include early-cancer detection and brain imaging. The annual conference on PAT has become the largest in SPIE’s 20,000-attendee Photonics West since 2010. CUP can image in 2D non-repeatable time-evolving events. CUP has a prominent advantage of measuring an x, y, t (x, y, spatial coordinates; t, time) scene with a single exposure, thereby allowing observation of transient events occurring on a time scale down to 100 femtoseconds, such as propagation of a light pulse. Further, akin to traditional photography, CUP is receive-only—avoiding specialized active illumination required by other single-shot ultrafast imagers. CUP can be coupled with front optics ranging from microscopes to telescopes for widespread applications in both fundamental and applied sciences.

Field-emission x-ray source arrays have been studied for both tomosynthesis and CT applications, however these arrays tend to have limited output. We propose the use of multi-source x-ray arrays using thermionic cathodes, contained within a single vacuum housing. A prototype 3-source x-ray array has been fabricated and tested, and the utility of multi-x-ray-source arrays has been demonstrated using physical simulations in both tomosynthesis and in cone beam CT. The prototype x-ray tube made use of a cylindrical molybdenum anode, machined to have 3 specific focal tracks. Grid-controlled cathode assemblies were fabricated and aligned to each focal tract, and the individual x-ray focal spots were evaluated with a star pattern at 35 kV and 40 mA. The 3-source assembly was used to physically simulate tomosynthesis imaging geometry, and tomosynthesis images of a lemon were obtained. Physical simulations using a cone beam breast CT scanner were also performed, by vertically moving the single x-ray source into 5 different locations – simulating 5 different source positions. A new geometry for cone beam CT imaging is proposed, where each source of a multi-x-ray source array is individually collimated to eliminate rays involving large cone angles. This geometry also allows three sources to be simultaneously pulsed onto a single flat panel detector, achieving better duty cycle and view sampling in cone beam CT. A reconstruction algorithm was written to accommodate the different source positions, and phantoms designed to demonstrate cone beam artifacts were imaged. The tomosynthesis images illustrate appropriate depth resolution in the test object. Analysis of the CT data demonstrate marked improvement compared to one source. We conclude that multi-source x-ray arrays using thermionic cathodes will have important applications in medical imaging, especially breast tomosynthesis and cone beam computed tomography.

State-of-the-art X-ray breast imaging (BI) modalities such as digital breast tomosynthesis (DBT), contrast enhanced spectral mammography (CESM) and breast CT (BCT) impose demanding requirements on digital X-ray detectors. This work studies the imaging performance of a GaAs two-threshold photon-counting detector (PCD) prototype for BI relevant X-ray spectra. The prototype has a 75 μm pixel size, two calibrated energy thresholds from 8 to 60 keV, 8 x 4 cm² area and a 0.5 mm thick GaAs sensor. The X-ray spectra used were 28 and 35 kVp with 2 mm Al filtration from a W-target tube emulating RQA-M2. The main imaging metrics probed include modulation transfer function (MTF) and detective quantum efficiency (DQE). Air kerma spanned three orders of magnitude from 370 nGy to 330 μGy. Furthermore, the detector’s linearity, lag and ghosting were also tested. For 28 kVp, the GaAs PCD exhibits 85% and 48% DQE for 0 and 5 lp/mm respectively, independent of the applied dose. MTF ranges from 98% to 53% for 1 and 6.667 lp/mm (Nyquist limit). Excellent linearity, zero lag and ghosting were observed. GaAs PCD technology is an ideal candidate for BI detector panels. Its stable temporal behavior, inherent zero readout noise and excellent DQE independent of the applied X-ray dose, improve on the combined advantages of current CsI/CMOS and a-Se/a-Si BI detectors. In addition, the multiple energy thresholds can enable spectral single-shot methods without motion blur.

Cost effective direct conversion detectors for different photon energies that provide high performance are of fundamental importance in medical diagnosis. Conventional direct conversion detectors typically provide either large area devices with moderate performance at reduced costs or expensive high performance architectures with limited size. In order to investigate the feasibility of highly efficient detectors based on low-cost large area processable hybrid organic-inorganic materials, multilayered device architectures consisting of stacked conversion layers are investigated. For this purpose, models that describe the sensitivity and the detective quantum efficiency are extended to the proposed detector design. This enables to evaluate the performance of multilayered detectors based on scintillator-sensitized organic and polycrystalline perovskite materials. A sensitivity analysis based on various multilayered designs at different photon energies shows significantly higher performance of polycrystalline perovskite conversion layers compared to scintillator-sensitized organic materials. The evaluation of detective quantum efficiencies lead to limitations for the number of stacked layers and enables to deduce design rules based on optimal layout parameters. The comparison with conventional single layer detectors shows competitive performance of multilayered detectors compared to high quality single crystals for all investigated photon energies.

In the past decade, grating-based x-ray multi-contrast imaging has demonstrated potential advantages for breast imaging, including reduced anatomical noise, sharper tumor boundary and improved visibility of microcalcifications. However, most of the studies have been performed on benchtop-based systems. The experimental conditions including the dose, scanning time and system geometry may not meet clinical standards. Therefore, to evaluate true clinical benefits of grating-based multi-contrast breast imaging, in-vivo imaging should be performed, which requires a human-compatible system. The purpose of this paper is to report the development of a human-compatible prototype multi-contrast imaging system. In particular, this work focuses on several key challenges in building the prototype system. Regarding the challenge of patient safety, the mean glandular dose (MGD) and the scatter radiation were evaluated for the prototype system. Regarding the challenge of the limited field-of-view (FOV), the origin of the problem and corresponding technical solutions are presented. Finally, imaging results of several test phantoms are presented and strategies to improve the image quality are discussed.

Purpose: We develop and validate a model-based framework for artifact correction and image reconstruction to enable application of Cone-Beam CT (CBCT) in quantitative assessment of bone mineral density (BMD). Compared to conventional quantitative CT, this approach does not require a BMD calibration phantom in the field-of-view during an object scan. Methods: The quantitative CBCT (qCBCT) imaging framework combined fast Monte Carlo (MC) scatter estimation, accurate models of detector response, and polyenergetic Poisson likelihood (PolyPL, Elbakri et al 2003). The underlying object model assumed that the tissues were ideal mixtures of water and calcium carbonate (CaCO3). Accuracy and reproducibility of qCBCT was evaluated in benchtop test-retest studies emulating a compact extremity CBCT system (axis-detector distance=56 cm, 90 kVp x-ray beam, ~16 mGy central dose). Various arrangements of Ca inserts (50 – 500 mg/mL) were placed in water cylinders of ~11 cm to ~15 cm diameter and scanned at multiple positions inside the fieldof- view for a total of 20 configurations. In addition, a cadaveric ankle was imaged in five configurations (with and without Ca inserts and water bath). Coefficient of variation (CV) of BMD values across different experimental configurations was used to assess reproducibility under varying imaging conditions. The performance of the model-based qCBCT framework (MC + PolyPL) was compared to FDK with water beam hardening correction and MC scatter correction. Results: The PolyPL framework achieved accuracy of 20 mg/mL or better across all insert densities and experimental configurations. By comparison, the accuracy of the FDK-based BMD estimates deteriorated with higher mineralization, resulting in ~120 mg/mL error for a 500 mg/mL Ca insert. Additionally, the model-based approach mitigated residual streaks that were present in FDK reconstructions. The CV of both methods was ~15% at 50 mg/mL Ca and less than ~8% for higher density inserts, where the PolyPL framework achieved 20-25% lower CV than the FDK-based approach. Conclusion: Accurate and reproducible BMD measurements can be achieved in extremity CBCT, supporting clinical applications in quantitative monitoring of fracture risk, osteoporosis treatment, and early osteoarthritis.

Deep neural networks have attracted growing interests in medical image due to its success in computer vision tasks. One barrier for the application of deep neural networks is the need of large amounts of prior training pairs, which is not always feasible in clinical practice. Recently, the deep image prior framework shows that the convolutional neural network (CNN) can learn intrinsic structure information from the corrupted image. In this work, an iterative parametric reconstruction framework is proposed using deep neural network as constraint. The network does not need prior training pairs, but only the patient’s own CT image. The training is based on Logan plot derived from multi-bed-position dynamic positron emission tomography (PET) images using ⁶⁸Ga-PRGD2 tracer. We formulated the estimation of the slope of Logan plot as a constraint optimization problem and solved it using the alternating direction method of multipliers (ADMM) algorithm. Quantification results based on real patient dataset shows that the proposed parametric reconstruction method is better than the Gaussian denoising and non-local mean denoising methods.

Radon transform is widely used in physical and life sciences and one of its major applications is the X-ray computed tomography (X-ray CT), which is significant in modern health examination. The Radon inversion or image reconstruction is challenging due to the potentially defective radon projections. Conventionally, the reconstruction process contains several ad hoc stages to approximate the corresponding Radon inversion. Each of the stages is highly dependent on the results of the previous stage. In this paper, we propose a novel unified framework for Radon inversion via deep learning (DL). The Radon inversion can be approximated by the proposed framework with an end-to-end fashion instead of processing step-by-step with multiple stages. For simplicity, the proposed framework is short as iRadonMap (inverse Radon transform approximation). Specifically, we implement the iRadonMap as an appropriative neural network, of which the architecture can be divided into two segments. In the first segment, a learnable fully-connected filtering layer is used to filter the radon projections along the view-angle direction, which is followed by a learnable sinusoidal back-projection layer to transfer the filtered radon projections into an image. The second segment is a common neural network architecture to further improve the reconstruction performance in the image domain. The iRadonMap is overall optimized by training a large number of generic images from ImageNet database. To evaluate the performance of the iRadonMap, clinical patient data is used. Qualitative results show promising reconstruction performance of the iRadonMap.

Iterative coordinate descent (ICD) is an optimization strategy for iterative reconstruction that is sometimes considered incompatible with parallel compute architectures such as graphics processing units (GPUs). We present a series of modifications that render ICD compatible with GPUs and demonstrate the code on a diagnostic, helical CT dataset. Our reference code is an open-source package, FreeCT ICD, which requires several hours for convergence. Three modifications are used. First, as with our reference code FreeCT ICD, the reconstruction is performed on a rotating coordinate grid, enabling the use of a stored system matrix. Second, every other voxel in the z-is updated direction simultaneously, and the sinogram data is shuffled to coalesce memory access. This increases the parallelism available to the GPU. Third, NS voxels in the xy-plane are updated simultaneously. This introduces possible crosstalk between updated voxels, but because the interaction between non-adjacent voxels is small, small values of NS still converge effectively. We find NS = 16 enables faster reconstruction via greater parallelism, and NS = 256 remains stable but has no additional computational benefit. When tested on a pediatric dataset of size 736x16x14000 reconstructed to a matrix size of 512x512x128 on a single GPU, our implementation of ICD can converge within 10 HU RMS in less than 5 minutes. This suggests that ICD could be competitive with simultaneous update algorithms on modern, parallel compute architectures.

Purpose: Our goal is to synthesize high quality and accurate Amyloid PET images with only ultra-low-dose PET images as input by using Generative Adversarial Network (GAN). Methods: 40 patients’ PET data was acquired with the injection of 330±30 MBq amyloid radiotracer 18F-florbetaben. The raw list mode PET data was reconstructed as the standard-dose ground truth and was randomly undersampled by a factor of 100 to reconstruct 1% low-dose PET scans. 32 volumes were used for training and the other 8 for testing. A 2D encoder-decoder network was implemented as the generator to synthesize a standard-dose image and a CNN based discriminator network was used to evaluate them. The two networks contested with each other to achieve accurate synthesis of standard-dose PET images with high visual quality from ultra-low-dose PET. Multi-slice input is used to reduce noise by providing the network with 2.5D information. Feature matching was applied to reduce the hallucinate structure. The image quality was evaluated by peak signal-to-noise ratio (PSNR), structural similarity (SSIM), mean square error (MSE), frequency domain blur measure (FBM) and edge blur measure (EBM) metrics. Results: The synthesized PET images showed remarkable improvement on all quality metric compared with the low-dose images. Comparing with the state-of-art method, adversarial learning is essential to ensure image quality and mitigate the blurring in the generated image. Multi-slice input reduced random noise and feature matching suppressed the hallucinate structure. Conclusion: Standard-dose Amyloid PET images can be synthesized from ultra-low-dose image by GAN. Applying adversarial learning, multi-slice input and feature matching technique are essential to ensure image quality.

Iterative reconstruction is a good match with the sparsely sampled limited angle data generated by breast tomosynthesis systems. However, it suffers from a specific artifact near the breast edge where it overestimates the x-ray path length, resulting in a considerable underestimation of the reconstructed linear attenuation coefficients. In this work, we present the application of a method that uses the measured 3D breast shape to reduce these artifacts in patient data, by including this information as an additional constraint in the image reconstruction process. A series of 50 patients undergoing breast tomosynthesis were additionally imaged with a pair of structured light cameras placed left and right of the mammography unit. These 3D surfaces were then aligned with the help of the backprojected breast outline from the x-ray data to form a single contour following the true breast shape. This was then further processed to generate a binary 3D mask set to 1 inside and to 0 outside the breast, and used as constraint in the reconstruction. Due to incomplete coverage and image artifacts, this mask was created successfully for only 19 out of 50 cases. Reconstructions were created with and without this constraint, and comparing attenuation profiles found that the artifact was almost completely corrected, bringing the reconstructed attenuation near the breast edge to the same level as the central region. Further visual inspection does show that higher quality optical 3D measurements and more precise alignment between optical and x-ray data are needed to avoid introducing new artifacts in the reconstruction.

Purpose: Cone-beam CT (CBCT) systems with a flat-panel detector (FPD) have advanced in a variety of specialty diagnostic imaging scenarios, with fluence modulation and multiple-gain detectors playing important roles in extending dynamic range and improving image quality. We present a penalized weighted least-squares (PWLS) reconstruction approach with a noise model that includes the effects of fluence modulation and electronic readout noise, and we show preliminary results that tests the concept with a CBCT head scanner prototype. Methods: Statistical weights in PWLS were modified using a realistic noise model for the FPD that considers factors such as system blur and spatially varying electronic noise in multiple-gain readout detectors (PWLS^e). A spatially varying gain term was then introduced in the calculation of statistical weights to account for the change in quantum noise due to fluence modulation (e.g. bowtie filter) (PWLS∗). The methods were tested in phantom experiments involving an elliptical phantom specially designed to stress dual-gain readout, and a water phantom and an anthropomorphic head phantom to quantify improvements in noise-resolution characteristics for the new PWLS methods (PWLS𝑒 and PWLS∗, and combined PWLS^∗e). The proposed methods were further tested using a high-quality, low-dose CBCT head scanner prototype in a clinical study involving patients with head injury. Results: Preliminary results show that the PWLS^e method demonstrated superior noise-resolution tradeoffs compared to conventional PWLS, with variance reduced by ~15-25% at matched resolution of 0.65 mm edge-spread-function (ESF) width. Clinical studies confirmed these findings, with variance reduced by ~15% in peripheral regions of the head without loss in spatial resolution, improving visual image quality in detection of peridural hemorrhage. A bowtie filter and polyenergetic gain correction improved image uniformity, and early results demonstrated that the proposed PWLS∗ method showed a ~40% reduction in variance compared to conventional PWLS when used with a bowtie filter. Conclusion: A more accurate noise model incorporated in PWLS statistical weights to account for fluence modulation and electronic readout noise reduces image noise and improves soft-tissue imaging performance in CBCT for clinical applications requiring a high degree of contrast resolution.

Dual Energy (DE) imaging has been widely used in digital radiography and fluoroscopy, as has dual energy CT for various medical applications. In this study, the imaging performance of a dynamic dual-layer a-Si flat panel detector (FPD) prototype was characterized for dual energy imaging tasks. Dual energy cone beam CT (DE CBCT) scans were acquired and used to perform material decomposition in the projection domain, followed by reconstruction to generate material specific and virtual monoenergetic (VM) images. The dual-layer FPD prototype was built on a Varex XRD 4343RF detector by adding a 200 μm thick CsI scintillator and a-Si panel of 150 μm pixel size on top as a low energy detector. A 1 mm copper filter was added as a middle layer to increase energy separation with the bottom layer as a high energy detector. The imaging performance, such as Modulation Transfer Function (MTF), Conversion Factor (CF), and Detector Quantum Efficiency (DQE) of both the top and bottom detector layers were characterized and compared with those of the standard single layer XRD4343 RF detector. Several tissue equivalent cylinders (solid water, liquid water, bone, acrylic, polyethylene, etc.) were placed on a rotating stand, and two separate 450-projection CBCT scans were performed under continuous 120 kV and 80 kV X-ray beams. After an empirical material decomposition calibration, water and bone images were generated for each projection, and their respective volumes were reconstructed using Varex’s CBCT Software Tools (CST 2.0). A VM image, which maximized the contrast-to-noise ratio of water to polyethylene, was generated based on the water and bone images. The MTF at 1.0 lp/mm from the low energy detector was 32% and 22% higher than the high energy detector and the standard detector, respectively; the DQE of both high and low energy detectors is much lower than that of the standard XRD 4343RF detector. The CNR of water to polyethylene from the VM image improved by 50% over that from the low energy image alone at 120 kV, and by 80% at 80 kV. This study demonstrates the feasibility of using a dual-layer FPD in applications such as DE CBCT for contrast enhancement and material decomposition. Further evaluations are underway.

An x-ray detector’s ability to produce high signal-to-noise ratio (SNR) images for a given exposure is described by the detective quantum efficiency (DQE) as a function of spatial frequency. Current mammography and radiography detectors have poor DQE performance at high frequencies due to noise aliasing when using a high- resolution converter layer. The Apodized-Aperture Pixel (AAP) design is novel detector design that increases high-frequency DQE by removing noise aliasing using smaller sensor elements (eg. 5 - 50 μm) than image pixel size (eg. 50 - 200 μm). The purpose of this work is to implement the AAP design on a selenium (Se) CMOS micro-sensor prototype with 7.8 × 7.8 μm size elements. Conventional (binned) and AAP images with 47 μm pixel size were synthesized and used to measure the modulation transfer function (MTF), normalized Wiener noise power spectrum (NNPS) and DQE. A micro-focus x-ray source (with a tungsten target) and a 60kV beam filtered with 2mm of aluminum was used to measure performance with DQEPro (DQEInstruments Inc., London, Canada) in a dynamic image acquisition mode at a high exposure level (9.7mR). The AAP design has 1.5x greater MTF near the image cut-off frequency (u_c = 10.6 cyc/mm) than conventional design. DQE near u_cwas 2.5x greater with the AAP design than conventional, and specimen imaging of a kidney stone shows greater SNR of fine-detail in the AAP image.

Photon-counting detector technology using directly converting high-Z sensor materials has recently gained popularity in medical imaging due to its capability to reduce patient dose, increase spatial resolution and provide single shot multienergy information. However, in medical imaging applications, the novel technology is not yet widely used due to technical challenges in manufacturing gap-less detectors with large areas. Here, a nearly gap-less, large-area, multi-energy photoncounting detector prototype is presented which was built with existing ASIC and sensor technology. It features an active area of 8x8 cm², a 1 mm thick cadmium telluride (CdTe) sensor, four independent energy thresholds and a pixel size of 150 μm. Single and multi-threshold imaging performance of the detector is evaluated by assessing various metrics relevant for conventional (polychromatic) and spectral imaging applications. A high detective quantum efficiency (DQE(0)=0.98), a low dark noise threshold (6.5 keV) and a high count rate capability (up to 3x10⁸ counts/s/mm²) indicate that the detector is optimally suited for conventional medical X-ray imaging, especially for low dose applications. Spectral performance was assessed by acquiring spectra from fluorescence samples, and the results show a high accuracy of energy peak positions (< 1 keV), precise energy resolution (within a few keV) and decent peak-to-background ratios. Spectral absorption measurements of water and iodinated contrast agent, as well as spectral X-ray radiographs of a human hand phantom, decomposed into bone and soft tissue basis images, demonstrate the multi-energy performance of the detector.

It has been reported and discussed that electrical current can be produced when an insulating material interacts with ionizing radiation. We have found that high-resolution images can be obtained from insulating materials if this current is guided by an electric field to the pixels of a TFT array. The charge production efficiency of insulators is much smaller than that of photoconductor materials such as selenium, silicon, or other conventional semiconductors. Nevertheless, when the intensity of the ionizing radiation is sufficiently high, a charge sensitive TFT imaging array with only dielectric material can produce high MTF images with contrast resolution proportional to the intensity of the radiation. The function of the dielectric in this new detector may be similar to that of an ionization chamber. Without the semiconductor charge generating material, the dielectric imaging detector does not exhibit charge generation fatigue or charge generation saturation. Prototype detectors have been tested using diagnostic x-ray beams with energy ranging from 25 kVp to 150 kVp, and therapeutic 2.5MV, 6MV, 10MV, and 15MV photon beams (with and without an electron built-up layer), electron beams, broad area proton beams, and proton pencil beams in the energy range of 150 MeV. High spatial resolution images up to the Nyquist frequency have been demonstrated. The physics, structure, and the imaging properties as well as the potential application of this detector will be presented and discussed.

Photon counting detectors are of increasing interest for clinical imaging. By measuring and digitally recording the deposited energy of each detected x-ray photon, these detectors can decrease the effect of electronic readout noise and Swank noise, potentially leading to improved image quality at decreased dose. The energy information of each photon can also be used to perform advanced imaging techniques such as material separation and k-edge imaging. Photon counting detectors employing crystalline silicon for the detector backplane are currently used for mammographic imaging and are under active investigation for fan-beam breast computed tomography (BCT). Our group has been exploring the possibility of creating monolithic, large-area photon counting detectors in order to perform cone-beam CT (CBCT) for BCT and radiation therapy (kV CBCT) applications. The detectors employ polycrystalline silicon (poly-Si) – a thin-film material that can be used to economically manufacture monolithic, large-area backplanes. In addition, poly-Si transistors have demonstrated good radiation damage resistance. The introduction of poly-Si-based photon counting detectors to BCT and kV CBCT would combine the benefits of photon counting with those of CBCT. However, one major challenge is designing circuits that are capable of handling the x-ray fluxes associated with these applications. A previously developed simulation methodology has been employed to model various levels of input flux in order to evaluate the count rate capabilities of poly-Si circuit designs. In this paper, the count rate capabilities of a promising poly-Si amplifier design under BCT and kV CBCT conditions are reported.

Material decomposition for imaging multiple contrast agents in a single acquisition has been made possible by spectral CT: a modality which incorporates multiple photon energy spectral sensitivities into a single data collection. This work presents an investigation of a new approach to spectral CT which does not rely on energy-discriminating detectors or multiple x-ray sources. Instead, a tiled pattern of K-edge filters are placed in front of the x-ray to create spatially encoded spectra data. For improved sampling, the spatial-spectral filter is moved continuously with respect to the source. A model-based material decomposition algorithm is adopted to directly reconstruct multiple material densities from projection data that is sparse in each spectral channel. Physical effects associated with the x-ray focal spot size and motion blur for the moving filter are expected to impact overall performance. In this work, those physical effects are modeled and a performance analysis is conducted. Specifically, experiments are presented with simulated focal spot widths between 0.2 mm and 4.0 mm. Additionally, filter motion blur is simulated for a linear translation speeds between 50 mm/s and 450 mm/s. The performance differential between a 0.2 mm and a 1.0 mm focal spot is less than 15% suggesting feasibility of the approach with realistic x-ray tubes. Moreover, for reasonable filter actuation speeds, higher speeds are shown to decrease error (due to improved sampling) despite motion-based spectral blur.

Multi-energy CT (MECT) enabled by energy-resolved photon-counting-detector CT (PCD-CT) is promising for materialspecific imaging with multiple contrast agents. However, non-idealities of the PCD such as pulse pileup, K-edge escape, and charge sharing may degrade the spectral performance. To perform MECT, an alternative approach was proposed by extending a “Twin Beam” design to a dual-source CT scanner with energy-integrating-detector (EID) by operating one or both sources in the “Twin Beam” mode to acquire three (triple-beam configuration) or four (quadruple-beam configuration) distinct X-ray beam measurements. Previous computer simulation studies demonstrated that the image quality and dose efficiency of the triple-beam configuration were comparable to that in PCD-CT for a three-material decomposition task involving iodine, bismuth, and water. The purpose of this work is to experimentally validate the proposed triple-beam MECT technique in comparison with PCD-CT. To mimic the dual-source triple-beam acquisition, two separate scans, one at 80 kV and the other at 120 kV operated in the “Twin Beam” mode, were performed on a single-source CT scanner. Two potential clinical applications of MECT for multiple contrast agents were investigated: iodine/gadolinium for biphasic liver imaging and iodine/bismuth for small bowel imaging. The results indicate that the imaging performance of the EID-based MECT may be comparable to that on the current PCD-CT platform for both the iodine/gadolinium and the iodine/bismuth material decomposition tasks.

Iodine K-edge CT imaging utilizes the sudden increase in the attenuation coefficient of iodine when the x-ray energy exceeds the K-shell binding energy of iodine. Early works on K-edge CT used multiple K-edge filters to generate different quasi-monoenergetic spectra with mean energies that straddled the iodine K-edge, and then multiple projections acquired with these spectra were processed to enhance the sensitivity of imaging iodine. Recent developments in energy-resolving photon counting detector (PCD) technology offer the potential for single-shot K-edge CT imaging. However, the performance of PCD-based iodine K-edge CT is often limited by the relatively low energy of the iodine K-edge (33.2 keV) compared with the mean energy of a polychromatic spectrum used in CT. This work explored the potential of introducing an iodine beam filter to PCD-based iodine K-edge CT to improve its imaging performance. To optimize the beam filtration condition, a realistic energy response function of an experimental PCD system was used when calculating the Cramér-Rao Lower Bounds (CRLBs) of three-material (iodine, bone, and water) decomposition estimators for each filtration condition. Experimental studies with a benchtop PCD CT system were performed to confirm the CRLB results. Both theoretical and experimental results demonstrated that by using an optimized iodine filter, quantitative accuracy of material basis images was improved. Compared with a commercial dual-energy-CT system, the optimized experimental K-edge CT system effectively reduced residual iodine signal in the bone basis image and reduced residual bone signal in the water-basis image.

We propose a two-dimensional (2D) contrast-enhanced dual-energy (DE) approach for functional x-ray imaging of respiratory disease. With this approach, non-radioactive xenon is used to provide contrast between ventilated regions of the lung and unventilated regions of the lung (i.e. ventilation defects); DE subtraction is used to suppress rib structures from 2D thoracic images. We modeled theoretically the signal-to-noise ratio (SNR) and area under the receiver operating characteristic curve (AUC) of a human observer for a defect present vs. defect absent binary classification task under signal-known-exactly/background-known-exactly conditions. Our model accounted for the size of ventilation defects, contrast of ventilation defects, quantum noise, finite spatial resolution, x-ray attenuation and observer efficiency. We modeled spherical defects with diameters up to 2.5 cm, and contrast and noise levels relevant for imaging of children, adolescents, adult males and adult females. Quantum noise and spatial resolution properties were calculated assuming an ideal energy-integrating x-ray detector. All calculations were performed assuming low-energy and high-energy applied tube voltages of 70 kV and 140 kV, respectively, with 2 mm of added copper filtration on the high-energy spectrum, and a total entrance exposure of 18 mR, which is typical for anterior-posterior thoracic imaging procedures. Our analysis shows that an AUC of 0.85 can be achieved for defect diameters as small as 1.1 cm, 1.2 cm, 1.3 cm and 1.4 cm for children ages 2 to 8, adolescents ages 9 to 14, adult males and adult females, respectively. Our results suggest that the DE approach proposed here warrants further investigation as a low-dose, low-cost alternative to existing approaches for functional imaging of respiratory disease.

Dual-energy (DE) technology is useful in chest radiography because it can separate anatomical structures such as bone and soft tissue. The standard log subtraction (SLS), simple smoothing of the high-energy image (SSH), anti-correlated noise reduction (ACNR), and a general linear noise reduction algorithm (GLNR) are used as conventional DE techniques to separate bone and soft tissue. However, conventional DE techniques cannot accurately decompose the anatomical structures because these techniques are based on the assumptions that X-ray imaging is a linear relationship. This relationship can cause quantum noise as well as anatomical loss of normal tissue and difficulty in detecting lesions. In this study, we propose a non-linear DE technique which requires a step to calculate the coefficient in advance using a calibration phantom. The calibration phantom composed to aluminum and PMMA material to calculate non-linear coefficients using the quadratic fitting model for soft tissue and bone. The results demonstrated that a non-linear DE technique showed the higher contrast-to-noise ratio (CNR), signal to noise ratio (SNR) and figure of merit (FOM) at 60 /70 kVp and 130 kVp. In addition, it showed better performance and image quality than conventional DE technique in terms of material decomposition capability. In conclusion, a non-linear DE technique is expected to increase the diagnostic accuracy in chest radiography.

In the context of x-ray CT, data completion is the process of augmenting truncated projection data to avoid artifacts during reconstruction. Data completion is commonly employed in dual-source CT where physical or hardware constraints limit the field of view covered by one of the two imaging chains. Practically, data completion is accomplished by extrapolating missing data based on the imaging chain with the full field of view, including some reweighting to approximate any spectral differences. While this approach works well in clinical applications, there are applications which would benefit from improved spectral estimation over the full field of view, including model-based iterative reconstruction, contrastenhanced abdominal imaging of large patients, and combined temporal and spectral imaging. Additionally, robust spectral data completion methods could provide an alternative to interior tomography for dose management in cardiac and spectral CT applications. To illustrate challenges with and potential machine-learning (ML) solutions for the spectral data completion problem, we present two realistic simulation experiments. A circular, cone-beam experiment disambiguates three contrast materials with dual-energy data and uses a generative network to inject 3D geometric information into a 2D, image-domain completion problem. A second clinical MDCT experiment uses a sophisticated variational network based on the split Bregman method and is structured to integrate directly into existing analytical reconstruction pipelines. While further work is required to establish performance limits and expectations, the results of both experiments strongly recommend the use of ML in spectral data completion problems.

The purpose of this study was to evaluate the feasibility of spectral mammography using the dual-energy method to noninvasively distinguish between type I (calcium oxalate, CO) and type II (calcium hydroxyapatite, HA) microcalcifications. Two types of microcalcifications are difficult to distinguish due to a similar linear attenuation coefficient. In order to improve the detection efficiency of microcalcifications, we used the photon counting detector with energy discrimination capability and microcalcifications were classified into optimal energy bins. Two energy bins were used to obtain dualenergy images. In this study, photon counting spectral mammography system was simulated using Geant4 Application for Tomographic Emission (GATE) simulation tools. The thickness of the breast phantom was 3 cm and microcalcifications of various sizes ranging from 130-550 μm were embedded into the breast phantom. Microcalcifications were classified as being calcium hydroxyapatite or calcium oxalate based on score calculation with the dual-energy images. According to the results, the measured CNR of calcium hydroxyapatite (HA) was higher than that of the calcium oxalate (CO) in conventional single-energy image. In addition, two types of microcalcifications were distinguished using dual-energy analysis method. This classification represents better performance with a high energy of 50 kVp and an energy threshold of 30 keV. These results indicate that the classification performance was improved when the difference in the low energy image and high energy image was used. This study demonstrated the feasibility of photon counting spectral mammography for classification of breast microcalcifications. We expect that dual-energy method can reduce the frequency of biopsy and discriminate microcalcifications in mammography. These results are expected to potentially improve the efficiency of early breast cancer diagnosis.

Contrast-enhanced digital mammography using spectroscopic x-ray detectors may improve image quality relative to existing contrast-enhanced breast imaging approaches. We present a framework for theoretical modelling of signal and noise in contrast-enhanced spectral mammography (CESM) and apply our framework to systems that use a spectroscopic amorphous selenium (a-Se) field-Shaping multi-Well Avalanche Detector (SWAD) which uses avalanche gain to overcome the low conversion gain of a-Se. We modelled an approach that uses an a-Se SWAD with 100x100 μm² detector elements, a converter thickness of 300 μm, an avalanche gain of ten, a 10-keV electronic noise floor and two energy bins. We modelled the influence of quantum efficiency, conversion gain, avalanche gain, characteristic emission, electronic noise, energy thresholding and image subtraction on the modulation transfer function (MTF), noise power spectrum (NPS) and iodine contrast. We investigated the choice of energy thresholds for the task of visualizing iodine signals. Our analysis demonstrates that reabsorption of characteristic photons yields energy-bin-dependent MTFs. As a result, spectral subtraction of low-energy and high-energy images enhances high spatial frequencies relative to low spatial frequencies. This effect, combined with better noise performance when using the lowest possible threshold to separate low-energy photons from electronic noise, results in better imaging performance than when reabsorption is suppressed through thresholding. Our theoretical framework enables quantifying trade offs between contrast, spatial resolution and noise for analysis of novel approaches for CESM, and provides a theoretical platform for comparison of CESM with existing approaches.

Cone-beam breast computed tomography (CBCT) would be a promising modality in screening and diagnosis of breast, providing complete 3-dimensional images with little painful compression of breast during the imaging compared to conventional mammography and tomosynthesis. To date, all CBCT systems including a commercial one by Koning have been utilizing a typical filament-based x-ray tube. However, the filament-based x-ray tube even in a grid type has strict limitation in time resolution, of longer than few milliseconds, with a limited dose rate to cause a large motion blur in CBCT projection images. Micro-calcifications of less than 1 mm in early breast cancer could be hardly distinguished by using conventional CBCT systems. We tried to solve this problem by adopting a fast digital x-ray tube based on carbon nanotube (CNT) field emitters. We, for the first time, developed a rotational anode x-ray tube with CNT emitters for advanced CBCTs. The x-ray tube consisted of CNT paste-emitters and a rotating anode made of W/Re target, and was fully vacuum-sealed with a glass envelope. Ultra-short x-ray pulses of less than sub-ms with a moderate high current of more than 200 mA and a focal spot of ~0.3 in nominal value was successfully obtained. We performed preliminary studies on CBCT imaging using the digital x-ray tube and achieved 300 projection images for 10 s, great reducing motion blurs in the images. It is expected that the CNT digital x-ray tube developed improves CBCT imaging greatly and then promotes CBCT modality in breast screening and diagnosis.

The objective of our study was to evaluate the feasibility of using silver sulfide nanoparticles (Ag₂S-NP) as a contrast agent for photon-counting mammography. The efficacies of Ag₂S-NP and iodine contrast were evaluated using a contrastembedded gradient phantom. The phantom was constructed using tissue-equivalent materials and varied continuously in composition from 100% glandular tissue to 100% adipose tissue. Each contrast agent was prepared at eight different concentrations: 1, 2, 5, 10, 15, 20, 25, and 30 mg/mL. Tubes of contrast agent were inserted into holes bored through the phantom in the direction of varying glandularity. Various images of the phantom were acquired by altering the acquisition parameters (kV, mAs, and high bin fraction). A range of beam energies from 26 kV to 40 kV was tested in this study. Our results demonstrate that for a given contrast agent, the contrast-to-noise ratio (CNR) is linearly proportional to concentration, and its magnitude is dependent on the kV of the spectrum. At mammographic energies, the Ag₂S-NP contrast increases with increasing kV and increasing solution concentration. Comparatively, the iodine signal becomes detectable only when the kV of the image is above iodine’s K-edge (33.2 keV). This indicates that the optimal energy for imaging iodine may exceed the clinical mammographic energy range. In summary, we have demonstrated the feasibility of using Ag₂S-NP as a contrast agent for breast imaging. Preliminary results from spectral photon-counting mammography indicate that Ag₂S-NP contrast has a significantly higher signal than iodine, especially when imaging at lower energies.

A previously developed image modification method used to simulate mammographic images as if acquired with different detectors and at different dose levels was extended to account for changes in the acquisition technique factors. The method was validated using two 3D printed realistic breast phantoms denoted ‘Anna’ and ‘Barbara’ of thickness 34 mm and 50 mm, respectively. A complete imaging system characterization of a commercial digital mammography system was performed. Using this system, three images from each phantom were acquired with Mo/Mo 26 kV and Rh/Rh 29 kV spectra, respectively, at two dose levels each, denoted the original images. To validate this method, additional images at a lower dose and at lower and higher tube voltages, denoted the target images, were also acquired. The original images were modified to simulate the target images, resulting in the simulated images. The signal in the original images was changed taking into account the target images conditions. Additional noise was also added to match that in the target images. The power spectra of the target and simulated images for each acquisition factor were found to match to within an average difference of 4% and 5% for the ‘Anna’ and ‘Barbara’ phantoms, respectively. Also, average structural similarity indices of 0.999 and 0.975 respectively, were obtained, meaning that all images are very similar. The method was found to accurately reflect acquisition with different techniques, as tested with 3D printed breast phantoms of different thicknesses, making it possible to use this method for future image quality evaluation studies.

Purpose: We optimize scan orbits and acquisition protocols for 3D imaging of the weight-bearing spine on a twin-robotic x-ray system (Multitom Rax). An advanced Cone-Beam CT (CBCT) simulation framework is used for systematic optimization and evaluation of protocols in terms of scatter, noise, imaging dose, and task-based performance in 3D image reconstructions. Methods: The x-ray system uses two robotic arms to move an x-ray source and a 43×43 cm² flat-panel detector around an upright patient. We investigate two classes of candidate scan orbits, both with the same source-axis distance of 690 mm: circular scans with variable axis-detector distance (ADD, air gap) ranging from 400 to 800 mm, and elliptical scans, where the ADD smoothly changes between the anterior-posterior view (ADDAP) and the lateral view (ADDLAT). The study involved elliptical orbits with fixed ADDAP of 400 mm and variable ADDLAT, ranging 400 to 800 mm. Scans of human lumbar spine were simulated using a framework that included accelerated Monte Carlo scatter estimation and realistic models of the x-ray source and detector. In the current work, x-ray fluence was held constant across all imaging configurations, corresponding to 0.5 mAs/frame. Performance of circular and elliptical orbits was compared in terms of scatter and scatter-to-primary ratio (SPR) in projections, and contrast, noise, contrast-to-noise ratio (CNR), and truncation (field of view, FOV) in 3D image reconstructions. Results: The highest mean SPR was found in lateral views, ranging from ~5 at ADD of 300 mm to ~1.2 at ADD of 800 mm. Elliptical scans enabled image acquisition with reduced lateral SPR and almost constant SPR across projection angles. The improvement in contrast across the investigated range of air gaps (due to reduction in scatter) was ~2.3x for circular orbits and ~1.9x for elliptical orbits. The increase in noise associated with increased ADD was more pronounced for circular scans (~2x) compared to elliptical scans (~1.5x). The circular orbit with the best CNR performance (ADD=600 mm) yielded ~10% better CNR than the best elliptical orbit (ADD_LAT=600 mm); however, the elliptical orbit increased FOV by ~16%. Conclusion: The flexible imaging geometry of the robotic x-ray system enables development of highly optimized scan orbits. Imaging of the weight-bearing spine could benefit from elliptical detector trajectories to achieve improved tradeoffs in scatter reduction, noise, and truncation.

The acquisition time of cone-beam CT (CBCT) systems is limited by different technical constraints. One important factor is the mechanical stability of the system components, especially when using C-arm or robotic systems. This leads to the fact that today’s acquisition protocols are performed at a system speed, where geometrical reproducibility can be guaranteed. However, from an application point of view faster acquisition times are useful since the time for breath-holding or being restraint in a static position has direct impact on patient comfort and image quality. Moreover, for certain applications, like imaging of extremities, a higher resolution might offer additional diagnostic value. In this work, we show that it is possible to intentionally exceed the conventional acquisition limits by accepting geometrical inaccuracies. To compensate deviations from the assumed scanning trajectory, a marker-free auto-focus method based on the gray-level histogram entropy was developed and evaluated. First experiments on a modified twin-robotic X-ray system (Multitom Rax, Siemens Healthcare GmbH, Erlangen, Germany) show that the acquisition time could be reduced from 14 s down to 9 s, while maintaining the same high-level image quality. In addition to that, by using optimized acquisition protocols, ultra-high-resolution imaging techniques become accessible.

This paper discusses a novel strategy for estimating projection matrices using the contours of anatomical structures tracked on x-ray projection images. It establishes an auto-calibration routine which calculates the geometrical projection parameters from unknown patient geometry based on iterative reconstruction and registration. By introducing the uncertainty of calibration parameters in registration, we achieve a robust correction for broad types of patient motion. As our method does not rely on consistency between projection data and tomography reconstruction it is robust towards truncation, noise and other typical artifacts of CBCT reconstruction. We evaluated our proposed method of auto-calibration on digital reconstructed radiographs (DRRs) of a CT headscan. In a standard dental CBCT setup our approach shows an average recovery of volume sharpness of 83:67% for different motion types.

Purpose: Interventional cone-beam CT (CBCT) is used for 3D guidance in interventional radiology (IR) procedures in the abdomen, with extended presence in trans-arterial chemoembolization (TACE) interventions for liver cancer. Image quality in this scenario is challenged by complex motion of soft-tissue abdominal structures, and by long acquisition times. We propose an image-based approach to estimate complex deformable motion through a combination of locally rigid motion trajectories. Methods: Deformable motion is estimated by minimizing a multi-region autofocus cost function. Motion is considered locally rigid for each region of interest (ROI) and the deformable motion field is obtained through spatial spline-based interpolation of the local trajectories. The multi-component cost function includes two regularization terms; one to penalize abrupt temporal transitions, and another to penalize abrupt spatial changes in the trajectory. Performance of deformable motion compensation was assessed in simulation studies with a digital abdomen phantom featuring a motion-induced deformable liver in static surrounding anatomy. Spherical inserts (4 – 12 mm diameter, -100 – 100 HU contrast) were placed in the liver. Image quality was evaluated by structural similarity (SSIM) with the static image as reference. Results: Motion compensated liver images showed better delineation of structure boundaries and recovery of distorted spherical shapes compared to their motion-corrupted counterparts. Consistent increase in SSIM was observed after motion compensation for the range of motion amplitudes studied (4 mm to 10 mm), showing 11% and 26% greater SSIM for 4 mm and 10 mm motion, respectively. Conclusion: The results indicate feasibility of image-based deformable motion compensation in soft-tissue abdominal CBCT imaging.

Edge illumination X-ray phase-contrast tomography (EIXPCT) is a emerging imaging technology in which partially opaque gratings are utilized with laboratory-based X-ray sources to estimate the distribution of the complex-valued refractive index. Spatial resolution in EIXPCT is mainly determined by the grating period of a sample mask, but can be significantly improved by a dithering technique in which multiple projection images are required per tomographic view angle as the object is moved over sub-pixel distances. Drawbacks of dithering include increased data acquisition times and radiation doses. Motivated by the flexibility in data acquisition designs enabled by a recently developed joint reconstruction (JR) method, a novel partial-dithering strategy for data acquisition is proposed. In this strategy, dithering is implemented at only a subset of the tomographic view angles. This results in spatial resolution that is comparable to that of the conventional full-dithering strategy where dithering is performed at every view angle, but the acquisition time is substantially decreased. The effect of dithering parameters on image resolution is explored.

The sensitivity factor of a grating-based x-ray differential phase contrast (DPC) imaging system determines how much fringe shift can be observed for a given refraction angle. It is commonly believed that increasing the sensitivity factor will improve the signal-to-noise ratio (SNR) of the phase signal. However, this may not always be the case if the intrinsic phase wrapping effect is taken into consideration. In this work, a theoretical derivation is provided to quantify relationship between the sensitivity and SNR for a given refraction angle, exposure level, and grating based x-ray DPC system. The theoretical derivation shows that the expected phase signal is not always proportional to the sensitivity factor and may even decrease when the sensitivity factor becomes too large. The noise variance of the signal is not always solely dependent on the exposure level and fringe visibility but may become signal-dependent under certain circumstances. As a result, SNR of the phase signal does not always increase with higher sensitivity. Numerical simulation studies were performed to validate the theoretical models. Results show that when the fringe visibility and exposure level are fixed, there exists an optimal sensitivity factor which maximizes the SNR for a given refraction angle; further increase of the sensitivity factor may decrease the SNR.

The design of an X-ray phase contrast tomography system for intra-operative specimen imaging based on edge illumination is presented. The use of edge illumination makes possible working with large focus, polychromatic X-ray sources reducing acquisition times of tomography scans down to values compatible with clinical use, while maintaining phase sensitivity in a compact device. The results collected so far show that application of this technology to breast conservation surgery has great potential to reduce re-operations, thus saving additional costs for healthcare services and stress for patients.

Standard histopathological examination is the gold standard for many disease diagnoses although the technique is limited with no full 3D volume analysis possible. Three dimensional X-ray Phase-Contrast Imaging(PCI) methods have been under constant and fast developments in the recent decades due to their superior performance for imaging low density objects and their ability to provide complementary information compared to attenuation based imaging. Despite the progresses, X-ray Phase Contrast Tomography still encounters remaining challenges to overcome on its way to become a routine non-invasive technique allowing the 3D assessment of tissue architecture in laboratory set-ups. Speckle Based Imaging (SBI) forms a new class of X-ray PCI techniques, sensitive to the first derivative of the phase. The set-up involved and the simplicity of implementation provide many advantages to SBI such as having no field of view and no resolution limitation in addition to have low requirements on the beam coherences. These advantages make SBI a good candidate for the transfer on conventional sources. In this work, we present preliminary results obtained on a conventional μCT and their comparison with data acquired at the European Synchrotron. We used a new phase retrieval algorithm based on optical energy conservation. We applied the method on both phantoms and biological samples in order to evaluate its quantitativeness for a transfer. A comparison to previously available speckle tracking algorithms is also performed. We demonstrate that the combination of the phase retrieval method with a standard μCT can achieve high resolution and high contrast within a few minutes, with a comparable image quality to the results obtained using synchrotron light.

X-ray phase differences are a thousand times greater than attenuation differences, but phase imaging has found limited clinical use due to requirements on x-ray coherence which may not be easily translated to clinical practice. Instead, this work employs a conventional source to create structured illumination with a simple wire mesh. The system simultaneously collects phase, attenuation, and scatter information. X-ray coherent scatter allows differentiation between tissue types with potentially much higher contrast than conventional radiography. Coherent-scatter images are collected with simple 1D slot-scanning and an angular shield to select signatures of interest from a relatively large region.

CMOS photon detectors (CPDs) are recently proposed photon sensing devices utilizing the latest CMOS image sensor (CIS) technologies [1]. CPDs are non-electron-multiplying devices, whose pixels have a fully depleted photo diode and have a readout noise of sub electron RMS even at room temperature. Using a 15μm pixel CPD test device coupled to a CsI(Tl) scintillator, we successfully obtained photon-count X-ray images. A Hamamatsu Photonics scintillator with 150μm CsI(Tl) layer coupled to fiber optic plate (FOP) of 3mm thickness is diced in dry condition and directly glued to the sensor surface. X-ray photons are injected from an X-ray tube with accelerating voltage of 30kV and 45kV using W target. Each X-ray photon creates a scintillation light spot in the captured images, where the injected position and photon energy are determined by integrating multiple pixel outputs at that spot. X-ray energy distributions were obtained at both 30kV and 45kV with reasonable differences. Modulated transfer function (MTF) of over 0.7 at 10LP/mm was achieved by mapping injected positions at 30kV. Photon-count images for slanted-edge MTF measurements as well as 10LP/mm of X-ray test chart were achieved. Those photon-count images were compared with conventional energy integrating images obtained with the same sensing device. Both image types confirmed superior resolution with photon-counting. This indirect X-ray photon counting technique using CPD has a potential of getting critical X-ray information for medical applications by achieving accurate injected positions of X-ray photons and their energies simultaneously.

Photon counting detectors are interesting candidates for next-generation clinical computed tomography scanners, promising improved contrast-to-noise ratio, spatial resolution and energy information compared to conventional energy-integrating detectors. Most attention is focused on cadmium telluride (CdTe) (or CZT) detectors, but silicon (Si) has been proposed as an alternative. We present detector simulation models fitted to published spectral response data for CdTe and Si, and use linear-systems theory to evaluate the spatial-frequency dependent DQE for lesion quantification and detection. Our fitted spectral response is consistent with Gaussian charge clouds with σ = 20.5 µm independent of energy for CdTe, and σ = 17 µm at 60 keV with an energy dependence of E^0.54 for Si. For a silicon strip detector with 0.5 × 0.5 mm² pixels separated by a 1D grid of 20 µm tungsten foils, the zero-frequency DQE for iodine detection is 0.43 for 30 mm detector absorption length and 0.46 for 60 mm detector absorption length. For iodine quantification in a water-iodine decomposition, the DQE is 0.26 for 30 mm and 0.27 for 60 mm Si. Compared to this detector, the DQE of a 1.6 mm thick CdTe detector with 0.225 mm pixels and two energy bins is 11-36 % higher for water and iodine detection but 28-51 % lower for material quantification. The predicted performance of Si is competitive with CdTe, suggesting that further consideration is warranted.

Photon-counting silicon strip detectors are attracting interest for use in next generation CT scanners. For silicon detectors, a low noise floor is necessary to obtain a good dose efficiency. A low noise floor can be achieved by having a filter with a long shaping time in the readout electronics. This also increases the pulse length, resulting in a long deadtime and thereby a degraded count-rate performance. However, as the flux typically varies greatly during a CT scan, a high count-rate capability is not required for all projection lines. It would therefore be desirable to use more than one shaping time within a single scan. To evaluate the potential benefit of using more than one shaping time, it is of interest to characterize the relation between the shaping time, the noise, and the resulting pulse shape. In this work we present noise and pulse shape measurements on a photon-counting detector with two different shaping times along with a complementary simulation model of the readout electronics. We show that increasing the shaping time from 28.1 ns to 39.4 ns decreases the noise and increases the signal-to-noise ratio (SNR) with 6.5% at low count rates and we also present pulse shapes for each shaping time as measured at a synchrotron source. Our results demonstrate that the shaping time plays an important role in optimizing the dose efficiency in a photon-counting x-ray detector.

Theoretical modeling of the performance of x-ray imaging detectors enables understanding relationships between the physics of x-ray detection and x-ray image quality, and enables theoretical optimization of novel x-ray imaging techniques and technologies. We present an overview of a framework for theoretical modeling of the frequency-dependent signal and noise properties of single-photon-counting (SPC) and energy-resolving x-ray imaging detectors. We show that the energy-response function, large-area gain, modulation transfer function (MTF), noise power spectrum (NPS) (including spatio-energetic noise correlations) and detective quantum efficiency (DQE) of SPC and energy-resolving x-ray imaging detectors are related through the probability density function (PDF) describing the number electron-hole (e-h) pairs collected in detector elements following individual x-ray interactions. We demonstrate how a PDF-transfer approach can be used to model analytically the MTF and NPS, including spatio-energetic noise correlations, of SPC and energy-resolving x-ray imaging detectors. Our approach enables modeling the combined effects of stochastic conversion gain, electronic noise, characteristic emission, characteristic reabsorption, coulomb repulsion and diffusion of e-h pairs and energy thresholding on the MTF and NPS. We present applications of this framework to (1) analysis of the frequency-dependent DQE of SPC systems that use cadmium telluride (CdTe) x-ray converters, and (2) analysis of spatio-energetic noise correlations in CdTe energy-resolving x-ray detectors. The developed framework provides a platform for theoretical optimization of next-generation SPC and energy-resolving x-ray imaging detectors.

Spectral CT with photon-counting detectors has demonstrated potential for improved material decomposition but is challenged by nonideal effects such as pulse pileup. The purpose of this study was to investigate the performance of Neural Network (NN) material decomposition under varying pulse-pileup conditions. We hypothesize that the NN can compensate for pileup effects as it learns the relationship between the spectral measurements and basis material thicknesses through calibration. Photon-counting experiments were performed to: (1) investigate the optimal NN architecture across varying pileup conditions, (2) quantify the performance of NN material decomposition across varying pileup conditions and (3) demonstrate material decomposition of photon-counting CT data across varying pileup conditions. The NN was trained with log-normalized spectral transmission measurements through known thicknesses of basis materials (PMMA and aluminum) at five flux levels. The trained NN was then applied to photoncounting transmission measurements through Teflon, Delrin, and Neoprene to estimate the basis material thicknesses. The trained neural network was also applied to photon-counting CT data of a rod phantom. The optimal NN configuration remained generally consistent across the studied flux levels, thus a NN configuration with five hidden-layer nodes was selected for the subsequent analysis. For the test material slabs, overall decomposition error decreased with flux for Teflon and Delrin, while increasing with flux for Neoprene. The CT experiments showed lowest material decomposition error for the lower flux condition, but similar error for the two higher flux conditions. Pulse-pileup decreased the variation in material decomposition estimates. The effects of pileup on material decomposition error varied across different materials. These preliminary results suggest that NN material decomposition may account for pulse-pileup effects in photon-counting CT.

Cerebral CT angiography (CTA) is widely used for the diagnosis of various cerebrovascular diseases, including strokes, vasculitis, aneurysms, and etc.2–4 For the diagnosis of ischemic strokes, the availability of high quality CTA images not only helps in identifying the presence/location of large vessel occlusion but also facilitates the assessment of collateral blood supply. As another example, accurate rendering of the superficial temporal arteries is valuable in identifying vessel inflammations induced by giant cell arteritis.5 While CTA is an established clinical gold standard for imaging large cerebral arteries and veins,1 an important challenge that currently remains for MDCT-based CTA is its limited performance in imaging small perforating arteries with a diameter below 0.5 mm.4 As a consequence, the relativley invasive artery biopsy procedure remains the current clinical gold standard for the diagnosis of giant cell arteritis.6 The use of indirect conversion energy integrating detectors puts intrinsic limit on the spatial resolution of MDCT, both in-plane and along the z direction. Severe partial volume averaging effect (PVE) and the preferential weighting of high energy photons7 are among major reasons for the relatively poor performance of MDCT-based CTA for imaging iodinated small vessels. Photon counting detector-based CT (PCD-CT) offers potential technological solutions to these challenges MDCT systems face for CTA. When compared to MDCT, the direct conversion design of PCD reduces limitations on both in-plane and through-plane spatial resolution, and the inherent equal weighting of high and low energy photons of PCD-CT systems offers an improvement in the CNR of iodinated vessels. The purpose of this work was to theoretically and experimentally study the potential impacts of the PCD-CT technology to an important component of CTA image package: the maximum intensity projection (MIP) image. MIP is a simple 3D image visualization method to display CTA data sets. Based on source images alone, it can be very challenging to evaluate occlusion conditions since most vessels extend to different z positions. In comparison, a MIP image that extracted information from a much longer z range can provide clearer evidence for an occlusion; in addition, it can effectively enhance the visibility of small collateral vessels. This work first derived the statistical properties of the MIP image, then analyzed how each of the benefits of PCD (improved z resolution; reduced noise autocovariance along z) propagates from the source CT images to the final MIP image. Finally, experiments were performed using a benchtop PCD-CT system and an anthropomorphic CTA phantom to showcase the significantly improved visibility of small perforating arteries.

Because total variation (TV) is non-differentiable, iterative reconstruction using a TV penalty comes with technical difficulties. To avoid these, it is popular to use a smooth approximation of TV instead, which is defined using a single parameter, herein called δ, with the convention that the approximation is better when δ is smaller. To our knowledge, it is not known how well image reconstruction with this approximation can approach a converged non-smooth TV-regularized result. In this work, we study this particular question in the context of X-ray computed tomography (CT). Experimental results are reported with real CT data of a head phantom and supplemented with a theoretical analysis. To address our question, we make proficient use of a generalized iterative soft-thresholding algorithm that allows us to handle TV and its smooth approximation in the same framework. Our results support the following conclusions. First, images reconstructed using the smooth approximation of TV appears to smoothly converge towards the TV result as δ tends to zero. Second, the value of δ does not need to be overly small to obtain a result that is essentially equivalent to TV, implying that numerical instabilities can be avoided. Last, though it is smooth, the convergence with δ is not particularly fast, as the mean absolute pixel difference decreases only as √δ in our experiments. Altogether, we conclude that the approximation is a theoretically valid way to approximate the non-smooth TV penalty for CT, opening the door to safe utilization of a wide variety of optimization algorithms.

Noise simulation methods for computed tomography (CT) scans are powerful tools for assessing image quality at a range of doses without compromising patient care. Current state of the art methods to simulate lower-dose images from standard-dose images insert Poisson or Gaussian noise in the raw projection data; however, these methods are not always feasible. The objective of this work was to develop an efficient tool to insert realistic, spatially correlated, locally varying noise to CT images in the image-domain utilizing information from the image to estimate the local noise power spectrum (NPS) and variance map. In this approach, normally distributed noise is filtered using the inverse Fourier transform of the square root of the estimated NPS to generate noise with the appropriate spatial correlation. The noise is element-wise multiplied by the standard deviation map to produce locally varying noise and is added to the noiseless or high-dose image. Results comparing the insertion of noise in the projection-domain versus the proposed insertion of noise in the image-domain demonstrate excellent agreement. While this image-domain method will never replace projection-domain methods, it shows promise as an alternative for tasks where projection-domain methods are not practical, such as the case for conducting large-scale studies utilizing hundreds of noise realizations or when the raw data is not available.

The Living Heart Model (LHM) was developed as part of the Living Heart Project by Dassault Systemes to provide a numerical finite element (FE) model of the human heart that accurately reproduces the normal cardiac physiology. We previously incorporated the LHM into the 4D extended cardiac-torso (XCAT) phantom for imaging research, rigidly transforming the model to fit it within different anatomies. This captured the variation in the overall size, position, and orientation of the heart but did not capture more subtle geometrical changes. Anatomic measurements of normal heart structures can show standard deviation variations of upwards of 25-30%. In this work, we investigate the use of Hyperelastic Warping to non-rigidly fit the LHM to new anatomies based on 4D CT data from anatomically diverse, normal patients. For each patient target, the mid-diastolic frame from the CT (heart is most relaxed) was segmented to define the cardiac chambers. The geometry of the LHM was then altered to match the targets using Hyperelastic Warping to register the LHM mesh, in its relaxed state, to each segmented dataset. The altered meshes were imported back into the FE software to simulate cardiac motion from the new geometries to incorporate into the XCAT phantom. By preserving the underlying LHM architecture, our work shows that Hyperelastic Warping allows for efficient revision of the LHM geometry. This method can produce a diverse collection of heart models, with added interior variability, to incorporate into the XCAT phantom to investigate 4D imaging methods used to diagnose and treat cardiac disease.

Volume-of-interest (VOI) imaging is a strategy in computed tomography (CT) that restricts x-ray fluence to particular anatomical targets via dynamic beam modulation. This permits dose reduction while retaining image quality within the VOI. VOI-CT implementation has been challenged historically by a lack of hardware solutions for tailoring the incident fluence to the patient and anatomical site as well as challenges involving interior tomography reconstruction of truncated projection data. In this work, we propose a general VOI-CT imaging framework using multiple aperture devices (MADs), an emerging beam filtration scheme based on two binary x-ray filters. Location of VOI is prescribed using two scout views at anterior-posterior (AP) and lateral perspectives. Based on a calibration of achievable fluence field patterns, MAD motion trajectories designed using an optimization objective that seeks to maximize the relative fluence in the VOI subject to minimum fluence constraints. A modified penalized-likelihood method is developed for reconstruction of heavily truncated data using the full-field scout views to help solve the interior tomography problem. Physical experiments were con- ducted to show the feasibility of non-centered and elliptical VOI in two applications - spine and lung imaging. Improved dose utilization and retained image quality is validated with respect to standard full-field protocols. Compared with full-field scans at reference dose, the MAD-VOI scans reduced total dose by 80% while retaining image quality. We observe that the contrast-to-noise ratio is 30% higher compared with low-dose full-field scans at the same dose usage.

In our previous study we successfully built a novel sheet-based dynamic beam attenuator (sbDBA) for fluence field modulation in X-ray computed tomography (CT) and performed a first-time experimental validation. In this work, we focus on the optimization of the DBA transmission properties for a given object. In clinical routine, CT scanners must cope with various attenuation properties differing from patient to patient. Typically, the attenuation of patients is high in the center of the fan beam, decreasing towards the periphery. The attenuation profiles of an object can also change for different X-ray tube positions. These variations cause unfavorable imbalances of image quality in the reconstructed object. Typically, the peripheral region, which is generally of minor diagnostically relevance, has relatively lower noise than the central region because the rays contributing to the peripheral region are less attenuated. This imbalance can be reduced by using beam-shaping prefilters, e.g. bowtie filters, attenuating the propagated intensity towards the periphery in a predefined, static profile. Bowtie filters, however, are not capable of dynamically adapting their attenuation to the attenuation profile of the patient. This can be accomplished by using dynamic beam attenuators (DBA) where the fan beam intensity can dynamically be modulated on a view-by-view basis, reducing noise inhomogeneities and enabling region-of-interest (ROI) imaging. Different scenarios (no attenuator, tube current modulation, conventional bowtie filter, the sbDBA and an ideal DBA) are compared in terms of image quality. The optimized sbDBA with tube current modulation (TCM) not only reduces the total radiation dose but also allows for spatial selection of intensity as required for ROI imaging.

In this work, a deep neural network architecture was developed and trained to reconstruct volumetric CT images from two-view radiograph scout localizers. In clinical CT exams, each patient will receive a two-view scout scan to generate both lateral (LAT) and anterior-posterior (AP) radiographs to help CT technologist to prescribe scanning parameters. After that, patients go through CT scans to generate CT images for clinical diagnosis. Therefore, for each patient, we will have two-view radiographs as input data set and the corresponding CT images as output to form our training data set. In this work, more than 1.1 million diagnostic CT images and their corresponding projection data from 4214 clinically indicated CT studies were retrospectively collected. The dataset was used to train a deep neural network which inputs the AP and LAT projections and outputs a volumetric CT localizer. Once the model was trained, 3D localizers were reconstructed for a validation cohort and results were analyzed and compared with the standard MDCT images. In particular, we were interested in the use of 3D localizers for the purpose of optimizing tube current modulation schemes, therefore we compared water equivalent diameters (Dw), radiologic paths and radiation dose distributions. The quantitative evaluation yields the following results: The mean±SD percent difference in Dw was 0.6±4.7% in 3D localizers compared to the Dw measured from the conventional CT reconstructions. 3D localizers showed excellent agreement in radiologic path measurements. Gamma analysis of radiation dose distributions indicated a 97.3%, 97.3% and 98.2% of voxels with passing gamma index for anatomical regions in the chest, abdomen and pelvis respectively. These results demonstrate the great success of the developed deep learning reconstruction method to generate volumetric scout CT image volumes.

Tomographic imaging using a penetrating wave, such as X-ray, light and microwave, is a fundamental approach to generate cross-sectional views of internal anatomy in a living subject or interrogate material composition of an object and plays an important role in modern science. To obtain an image free of aliasing artifacts, a sufficiently dense angular sampling that satisfies the Shannon-Nyquist criterion is required. In the past two decades, image reconstruction strategy with sparse sampling has been investigated extensively using approaches such as compressed-sensing. This type of approach is, however, ad hoc in nature as it encourages certain form of images. Recent advancement in deep learning provides an enabling tool to transform the way that an image is constructed. Along this line, Zhu et al¹ presented a data-driven supervised learning framework to relate the sensor and image domain data and applied the method to magnetic resonance imaging (MRI). Here we investigate a deep learning strategy of tomographic X-ray imaging in the limit of a single-view projection data input. For the first time, we introduce the concept of dimension transformation in image feature domain to facilitate volumetric imaging by using a single or multiple 2D projections. The mechanism here is fundamentally different from the traditional approaches in that the image formation is driven by prior knowledge casted in the deep learning model. This work pushes the boundary of tomographic imaging to the single-view limit and opens new opportunities for numerous practical applications, such as image guided interventions and security inspections. It may also revolutionize the hardware design of future tomographic imaging systems

We propose a learning method to generate corrected CBCT (CCBCT) images with the goal of improving the image quality and clinical utility of on-board CBCT. The proposed method integrated a residual block concept into a cyclegenerative adversarial network (cycle-GAN) framework, which is named as Res-cycle GAN in this study. Compared with a GAN, a cycle-GAN includes an inverse transformation from CBCT to CT images, which could further constrain the learning model. A fully convolution neural network (FCN) with residual block is used in generator to enable end-toend transformation. A FCN is used in discriminator to discriminate from planning CT (ground truth) and correction CBCT (CCBCT) generated by the generator. This proposed algorithm was evaluated using 12 sets of patient data with CBCT and CT images. The mean absolute error (MAE), peak signal-to-noise ratio (PSNR), normalized cross correlation (NCC) indexes and spatial non-uniformity (SNU) in the selected regions of interests (ROIs) were used to quantify the correction accuracy of the proposed algorithm. Overall, the MAE, PSNR, NCC and SNU were 20.8±3.4 HU, 32. 8±1.5 dB, 0.986±0.004 and 1.7±3.6%. We have developed a novel deep learning-based method to generate CCBCT with a high image quality. The proposed method increases on-board CBCT image quality, making it comparable to that of the planning CT. With further evaluation and clinical implementation, this method could lead to quantitative adaptive radiotherapy.

Conventional Cone-Beam Computed Tomography (CBCT) acquisition suffers from motion blurring artifacts at the region of the thorax, and consequently, it may result in inaccuracy in localizing the target of treatment and verifying delivered dose in radiation therapy. Although 4D-CBCT reconstruction technology is available to alleviate the motion blurring artifacts with the strategy of projection sorting followed by independent reconstruction, under-sampling streaking artifacts and noise are observed in the set of 4D-CBCT images due to relatively fewer projections and large angular spacing in each phase. Aiming at improving the overall quality of 4D-CBCT images, we explored the performance of the deep learning model on 4D-CBCT images, which has been paid little attention before. Inspired by the high correlation among the 4D-CBCT images at different phases, we incorporated a prior image reconstructed from full-sampled projections beforehand into a lightweight structured convolutional neural network (CNN) as one input channel. The prior image used in the CNN model can guide the final output image to restore detailed features in the testing process, so it is referred to as Prior-guided CNN. Both simulation and real data experiments have been carried out to verify the effectiveness of our CNN model. Experimental results demonstrate the effectiveness of the proposed CNN regarding artifact suppression and preservation of anatomical structures. Quantitative evaluations also indicate that 33.3% and 21.2% increases in terms of Structural Similarity Index (SSIM) have been achieved by our model when comparing with gated reconstruction and images tested on CNN without prior knowledge, respectively.

The purpose of this study was to develop a robust, automated multi-organ segmentation model for clinical adult and pediatric CT and implement the model as part of a patient-specific safety and quality monitoring system. 3D convolutional neural network (Unet) models were setup to segment 30 different organs and structures at the diagnostic image resolution. For each organ, 200 manually-labeled cases were used to train the network, fitting it to different clinical imaging resolutions and contrast enhancement stages. The dataset was randomly shuffled, and divided with 6/2/2 train/validation/test set split. The model was deployed to automatically segment 1200 clinical CT images as a demonstration of the utility of the method. Each case was made into a patient-specific phantom based on the segmentation masks, with unsegmented organs and structures filled in by deforming a template XCAT phantom of similar anatomy. The organ doses were then estimated using a validated scanner-specific MC-GPU package using the actual scan information. The segmented organ information was likewise used to assess contrast, noise, and detectability index within each organ. The neural network segmentation model showed dice similarity coefficients (DSC) above 0.85 for the majority of organs. Notably, the lungs and liver showed a DSC of 0.95 and 0.94, respectively. The segmentation results produced patient-specific dose and quality values across the tested 1200 patients with representative the histogram distributions. The measurements were compared in global-to-organ (e.g. CTDvol vs. liver dose) and organ-to-organ (e.g. liver dose vs. spleen dose) manner. The global-to-organ measurements (liver dose vs. CTDIvol: 𝑅 = 0.62; liver vs. global d’: 𝑅 = 0.78; liver vs. global noise: 𝑅 = 0.55) were less correlated compared to the organ-to-organ measurements (liver vs. spleen dose: 𝑅 = 0.93; liver vs. spleen d’: 𝑅 = 0.82; liver vs. spleen noise: 𝑅 = 0.78). This variation of measurement is more prominent for iterative reconstruction kernel compared to the filtered back projection kernel (liver vs. global noise: 𝑅𝐼𝑅 = 0.47 vs. 𝑅𝐹𝐵𝑃 = 0.75; liver vs. global d’: 𝑅𝐼𝑅 = 0.74 vs. 𝑅𝐹𝐵𝑃 = 0.83). The results can help derive meaningful relationships between image quality, organ doses, and patient attributes.

Deep learning methods have shown great promise in tackling challenging medical imaging tasks. Within the field of x-ray CT, deep learning for image denoising is of interest because of the fundamental link between ionizing radiation dose and diagnostic image quality, the limited availability of clinical projection data, and the computational expense of iterative reconstruction methods. Here, we work with 3D, temporal CT data (4D, cardiac CT), where redundancies in spatial sampling necessitate careful control of imaging dose. Specifically, using custom extensions to the Tensorflow and Keras machine learning packages, we construct and train a 4D, convolutional neural network (CNN) to denoise helical, cardiac CT data acquired in a mouse model of atherosclerosis. With the objective of accelerating iterative reconstruction, we train the CNN to map undersampled algebraic reconstructions of the 4D data to fully-sampled and regularized iterative reconstructions under mean-squared-error, perceptual loss, and low rank cost terms. Using phantom data for quantitative validation, we verify that the CNN robustly denoises static potions of the image without compromising temporal fidelity and that the CNN performs similarly to regularized, iterative reconstruction with the split Bregman method (CNN temporal RMSE: 142 HU; iterative temporal RMSE: 136 HU). Using in vivo validation and testing data excluded from CNN training, we verify that the CNN generalizes well, approximately reproducing the noise power spectrum of the iteratively reconstructed data (noise std. in water vial near heart, CNN: 62-73 HU, depending on cardiac phase; iterative: 94-100 HU), without degradation of spatial resolution (axial MTF, 10% cutoff, CNN: 2.69 lp/mm; iterative: 2.63 lp/mm). Overall, the results presented in this work represent a positive step toward realizing the promises of deep learning methods in medical imaging.

In modern multi-slice CT systems, bowtie shape wedge filter is widely used for optimizing patient received radiation dose distribution. Since wedge filter is usually fixed in the path of X-ray, the induced scatter radiation could give impact on image quality under certain cases. In order to compensate this extra scatter radiation and improve the image quality, we introduced a wedge scatter correction algorithm integrated in the raw data pre-processing workflow. After the algorithm is implemented on our latest CT systems, the improvement can be seen from both water phantom images and clinical patient images.

In computed tomography (CT), miscalibrated or imperfect detector elements produce stripe artifacts in the sinogram. The stripe artifacts in Radon space are responsible for concentric ring artifacts in the reconstructed images. To remove the ring artifacts from the images, there exist several methods. While most of the methods are performed in the image domain, a few of them are adopted in the projection or sinogram domain. In the latter methods, the sinogram is intended to be corrected before the image reconstruction. In this work, a novel optimization model is proposed to remove the ring artifacts in an iterative reconstruction procedure. To correct a sinogram, a new correcting vector is proposed to compensate for malfunctioning of detectors in the projection domain. Moreover, a novel directional total variation (DTV) based regularization is developed to penalize the ring artifacts in the image domain. The optimization problem is solved by using the alternating minimization scheme (AMS). In each iteration, the fidelity term along with the DTV-based regularization is first solved to find the image, and then the correcting coefficient vector is updated according to the obtained image. Because the sinogram and the image are simultaneously updated, the proposed method basically performs in both the image and sinogram domains. The proposed method is evaluated using real preclinical datasets containing strong ring artifacts , and the experimental results show that the proposed algorithm can efficiently suppress the ring artifacts.

Skin cancer is the most common cancer in the US; one in five Americans will develop it by the age of 70. Early diagnosis offers more favorable treatment options; currently available diagnostics, however, shows large readerdependence. The simulation approach (virtual trials) to development and validation offers advantages in terms of quantitative and objective assessment of system performance in the design of novel imaging methods, and validation of clinical trial designs prior to execution of real clinical trials. We have designed a pipeline for performing Virtual Clinical Trials of optical imaging of skin lesions. This pipeline includes modules to simulate healthy skin and subcutaneous tissue, to create skin lesions and insert those in the skin models, and to simulate the acquisition of optical images resulting in simulated/virtual images of skin. The last module of the pipeline performs virtual reading of simulated images, which is based on the clinical task-based performance analysis. The physical properties of the skin and lesions used in our simulations were selected to represent clinically plausible conditions. Skin lesion images were simulated assuming the ambient white light, and a linear camera model. We have utilized the standards for optimal data representation based upon the XML schema, adopted from the VCT pipeline developed for breast imaging research. This paper describes the principles used for designing the proposed VCT pipeline and presents preliminary simulation results, including visual and quantitative evaluation. Integration of pipeline modules and its validation is ongoing.

Full-field 2D digital mammography is used for breast cancer screening throughout the world. Digital breast tomosynthesis (DBT) is now widely available and has shown promise as a breast cancer screening modality. Rigorous evaluation and comparison studies must be conducted before considering the new modality for routine breast cancer screening. Conventional clinical trials involving human subjects are time consuming and expensive and are limited to commercially-available system designs. Alternatively, Virtual Clinical Trials (VCTs) can be used to conduct such studies using modelling tools by simulating clinically-realistic images and experimental system designs. The OPTIMAM image simulation toolbox contains a suite of tools that can used to simulate visually and clinically realistic images for VCTs. Recently, tools for simulating a wide-angle DBT system were added to the toolbox. In this paper, we present the simulation methodology and validation results for a wide-angle DBT system. The validation was performed by simulating images of standard test objects and comparing these with real images acquired using identical settings on the simulated real system. The comparison of the contrast-to-noise ratios, geometrical distortion (z-resolution) and image blurring for real and simulated images of test objects showed good agreement. This suggests that the images of a wide-angle DBT system produced using our simulation approach are comparable to real images.

This work presents and tests a new method of adapting mammography images to appear as if acquired using different radiographic factors by changing the signal and noise within the images. A Hologic Selenia Dimensions was used for the validation of the conversion method. Two set of images were acquired for the validation, one set was of 30, 45 and 60 mm thick polymethyl methacrylate (PMMA) acquired for a range of beam qualities while the other set was the same except for the addition of a contrast object of 2 and 4 mm thick pieces of PMMA. One set of images was then adapted to appear the same as a target image acquired with a higher or lower tube voltage. The normalized noise power spectra (NNPS) and standard deviation of the flat field images and the contrast-to-noise ratio (CNR) of the other images were calculated for the simulated and target images. The NNPS and standard deviation for the target and the simulated images were found to be within 2%. The CNRs of the target and simulated images were found to be within 4% of each other. The methodology successfully converted the images and can be used in observer studies to examine the impact of radiographic factors on lesion detection.

Diffusion weighted MRI (DW-MRI) depends on accurate quantification signal intensities that reflect directional apparent diffusion coefficients (ADC). Signal drift and fluctuations during imaging can cause systematic non-linearities that manifest as ADC changes if not corrected. Here, we present a case study on a large longitudinal dataset of typical diffusion tensor imaging. We investigate observed variation in the cerebral spinal fluid (CSF) regions of the brain, which should represent compartments with isotropic diffusivity. The study contains 3949 DW-MRI acquisitions of the human brain with 918 subjects and 542 with repeated scan sessions. We provide an analysis of the inter-scan, inter-session, and intra-session variation and an analysis of the associations with the applied diffusion gradient directions. We investigate a hypothesis that CSF models could be used in lieu of an interspersed minimally diffusion-weighted image (b0) correction. Variation in CSF signal is not largely attributable to within-scan dynamic anatomical changes (3.6%), but rather has substantial variation across scan sessions (10.6%) and increased variation across individuals (26.6%). Unfortunately, CSF intensity is not solely explained by a main drift model or a gradient model, but rather has statistically significant associations with both possible explanations. Further exploration is necessary for CSF drift to be used as an effective harmonization technique.

We have been working on the development of a Scatter Display System (SDS) for monitoring and displaying scatterrelated dose to staff members during fluoroscopic interventional procedures. We have considered various methods for such a display using augmented reality (AR) and computer-generated virtual reality (VR). The current work focuses on development of the VR SDS display, which shows the color-coded scattered dose distribution in a horizontal plane at a selected height above the floor in a top-down view of the interventional suite. Reported is the first development of the methodology for real-time functionality of this software via integration of controller area network (CAN) bus digital signals from the Canon C-Arm Biplane System. Importing the CAN bus information allows immediate selection of the appropriate pre-calculated scatter dose distribution consistent with the x-ray beam orientation and characteristics as well as selection of the proper gantry and table graphic for the display. The Python CAN interface module was used for streamlining the integration of the CAN bus interface. Development of real-time functionality for the SDS allows it to provide feedback to staff during clinical procedures for informed dose management; the SDS can work alongside the patient skin dose tracking system (DTS) for complete clinical monitoring of staff and patient dose.

Deformation-driven CBCT reconstruction techniques can generate accurate and high-quality CBCTs from deforming prior CTs using sparse-view cone-beam projections. The solved deformation-vector-fields (DVFs) also propagate tumor contours from prior CTs, which allows automatic localization of low-contrast liver tumors on CBCTs. To solve the DVFs, the deformation-driven techniques generate digitally-reconstructed-radiographs (DRRs) from the deformed image to compare with acquired cone-beam projections, and use their intensity mismatch as a metric to evaluate and optimize the DVFs. To boost the deformation accuracy at low-contrast liver tumor regions where limited intensity information exists, we incorporated biomechanical modeling into the deformation-driven CBCT reconstruction process. Biomechanical modeling solves the deformation on the basis of material geometric and elastic properties, enabling accurate deformation in a low-contrast context. Moreover, real clinical cone-beam projections contain amplified scatter and noise than DRRs. These degrading signals are complex, non-linear in nature and can reduce the accuracy of deformation-driven CBCT reconstruction. Conventional correction methods towards these signals like linear fitting lead to over-simplification and sub-optimal results. To address this issue, this study applied deep learning to derive an intensity mapping scheme between cone-beam projections and DRRs for cone-beam projection intensity correction prior to CBCT reconstructions. Evaluated by 10 liver imaging sets, the proposed technique achieved accurate liver CBCT reconstruction and localized the tumors to an accuracy of ~1 mm, with average DICE coefficient over 0.8. Incorporating biomechanical modeling and deep learning, the deformation-driven technique allows accurate liver CBCT reconstruction from sparse-view projections, and accurate deformation of low-contrast areas for automatic tumor localization.

For most offline calibration methods, a specific phantom scanning is often required before the following imaging tasks. It’s time-consuming and unable to be applied to the unstable systems. Some online approaches don’t have such drawbacks but their accuracy cannot fulfil requirements compared with the offline ones. This paper proposes an Epipolar geometry consistency based online geometric calibration for cone beam CT(CBCT). Four parameters: the detector skew, rotation axis, mid-plane and the source to detector distance are employed for geometrical modeling of CBCT in this paper. A cost function is built by exploiting the Epipolar geometry consistency among the projective views. By taking advantage of the simplex-simulated annealing algorithm(SIMPSA) algorithm to minimize the cost function, we can obtain the geometric parameters of our practical CBCT systems for image reconstruction. In simulation, different noise levels are added to projection images respectively and the experimental results show that the proposed method is insensitive to noise. What’s more, the accuracies of the parameters of the detector skew and rotation axis is slightly higher than that of the other two. In the practical situation, we scanned a thin bamboo stick and a Chinese parasol tree branch. Compared with the image reconstructed with the geometric parameters calculated by the conventional offline calibration approach, the ones reconstructed with the proposed method show no apparent difference and geometric artifacts, which shows that the accuracy of our method is comparable to that of the offline calibration ones.

This study proposes a method of using one test bolus to monitor peak bolus arrival time at two locations, the aorta and the knee, in CT angiography lower-extremity runoff scans. The resulting aortopopliteal transition time will facilitate determining appropriate CT scan parameters to match the bolus speed. The proposed method first monitors the test bolus peak at the aorta. When the contrast enhancement peak is measured, the table is moved to monitor the test bolus at the knee. Instead of cross-sectional images, the proposed method exploits projection (single view) scans for monitoring the bolus to reduce X-ray exposure and to enable real-time peak identification. The feasibility on scan timing of the proposed method was verified by simulations. The medium and high mean blood velocities used in this study were simulated by Monte Carlo methods. Blood velocity at each location inside the arteries were obtained by a three-segment blood velocity simulation. Table motion specifications of a clinical CT scanner were also simulated. Results shown that for medium aortopopliteal distance (690 mm) and medium blood velocity (65.8 mm/sec), the table arrived at the knee position 9.99 seconds ahead of the test bolus peak, which is enough time to monitor the bolus peak at the second location. For the most challenging case, i.e. shortest aortopopliteal distance (624 mm) and high blood velocity (179.5 mm/sec), the time difference between table and bolus peak arrival to the second location was 1.87 seconds, which allows a small window of monitor scans to detect the bolus peak.

By extending the conventional single-energy computed tomography (SECT) along the energy dimension, spectral CT achieves superior energy resolution and material distinguishability. However, for the state-of-the-art photon counting detector (PCD) based spectral CT, because the emitted photons with a fixed total number for one X-ray beam are divided into several different energy bins, the noise level is increased in each reconstructed channel image, and it further leads to an inaccurate material decomposition. To improve the reconstruction quality and decomposition accuracy, in this work, we first employ a refined locally linear transform to convert the structural similarity among two-dimensional (2D) spectral CT images to a spectral-dimension gradient sparsity. By combining the gradient sparsity in the spatial domain, a global three-dimensional (3D) gradient sparse representation is constructed and measured by L₁, L₀- and trace-norm, respectively. For each sparsity measurement, we propose the corresponding optimization model, develop the corresponding iterative algorithm, and verify the effectiveness and superiority with both simulated and real datasets.

Beating of the heart is a type of motion which is the most difficult to control during the cardiac CT scan and causes significant artifacts. Hearts have the least motion at systole and diastole phase which for an average heart happens at 45% and 75% phase respectively. However, in practice this is not guaranteed, so doctors sometimes reconstruct several phases, review all those images and then make diagnosis from the phase that has the least artifacts. The new method for automatic dynamic optimal phase reconstruction has image quality comparable to the manual phase selection but it also significantly reduces the exam time by omitting the review of unnecessary phases.

In sparse-view Computed Tomography (CT), only a small number of projection images are taken around the object, and sinogram interpolation method has a significant impact on final image quality. When the amount of sparsity - the amount of missing views in sinogram data – is not high, conventional interpolation methods have yielded good results. When the amount of sparsity is high, more advanced sinogram interpolation methods are needed. Recently, several deep learning (DL) based sinogram interpolation methods have been proposed. However, those DL-based methods have mostly tested so far on computer simulated sinogram data rather experimentally acquired sinogram data. In this study, we developed a sinogram interpolation method for sparse-view micro-CT based on the combination of U-Net and residual learning. We applied the method to sinogram data obtained from sparse-view micro-CT experiments, where the sparsity reached 90%. The interpolated sinogram by the DL neural network was fed to FBP algorithm for reconstruction. The result shows that both RMSE and SSIM of CT image are greatly improved. The experimental results demonstrate that this sinogram interpolation method produce significantly better results over standard linear interpolation methods when the sinogram data are extremely sparse.

For a very long time, low-dose computed tomography (CT) imaging techniques have been performed by either preprocessing the projection data or regularizing the iterative reconstruction. The conventional filtered backprojection (FBP) algorithm is rarely studied. In this work, we show that the intermediate data during FBP possess some fascinating properties and can be readily processed to reduce the noise and artifacts. The FBP algorithm can be technically decomposed into three steps: filtering, view-by-view backprojection and summing. The data after view-by-view backprojection is naturally a tensor, which is supposed to contain useful information for processing in higher dimensionality. We here introduce a sorting operation to the tensor along the angular direction based on the pixel intensity. The sorting for each point in the image plane is independent. Through the sorting operation, the structures of the object can be explicitly encoded into the tensor data and the artifacts can be automatically driven into the top and bottom slices of the tensor. The sorted tensor also provides high dimensional information and good low-rank properties. Therefore, any advanced processing methods can be applied. In the experiments, we demonstrate that under the proposed scheme, even the Gaussian smoothing can be used to remove the streaking artifacts in the ultra-low dose case, with nearly no compromising of the image resolution. It is noted that the scheme presented in this paper is a heuristic idea for developing new algorithms of low-dose CT imaging.

In this work, we inv