This PDF file contains the front matter associated with SPIE Proceedings Volume 12925, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Photon counting detectors (PCDs) with energy discrimination capabilities enable spectral imaging and provide higher contrast-to-noise ratio (CNR) than energy integrating detectors (EIDs). Previous work has shown that optimal weighting strategies can further improve CNR. However, a systematic comparison of different weighting strategies and their maximum achievable CNR has not been investigated. In this work, we derive the analytical expressions of the maximum achievable CNR using pre-log optimal weights, post-log optimal weights, and optimal virtual monoenergetic (VM) weights for realistic PCD energy responses and imaging tasks. We compare the CNR performance of these weighting strategies using simulated phantoms and a physical phantom scan for low-contrast tasks. Our results show that for spectral tasks, pre-log optimal weights, post-log optimal weights, and optimal VM weights achieve approximately equal maximum CNR (to within 2% of each other) and outperform total count (TC) images.

The spectral inconsistency across detector pixels is still a major challenge that could directly lead to obvious artifacts in reconstructed CT images and severe inaccuracies in material discrimination. This work attempts to develop a novel approach of modeling the energy threshold induced spectral inconsistency for photon-counting CT (PCCT) to adaptively characterize and correct for spectral effects. This proposed method is based on the general model expression of photon counts, adds variables to the integration term to represent spectral distortion and furthermore takes a threshold-related nonuniformity bias factor into account at integration bounds. In this work, we assume that the variables associated with spectral distortion are mainly related to the system geometry and the detector material, there is only threshold-related nonuniformity bias factor to change with the energy-threshold, regardless of some non-ideal effects like pileup and charge sharing. After simplification, we solve the linear equations to get the threshold-related nonuniformity bias at different thresholds, and the corrected photon counts can be calculated together with the spectral distortion variables have been worked out at the first reference threshold. Experimental results showed that the ring and band artifacts caused by spectral inconsistency were significantly reduced, the quality of reconstructed images is visibly improve. And the universality and robustness of our method are also demonstrated preliminary.

Photon-counting CT (PCCT) and energy-integrating CT (EICT) system offer differing image quality attributes, but it is uncertain if such attributes manifest themselves differently across phantom vs patients. This study investigated if image quality assessment with a multi-sized phantom can predict patient population trends in PCCT and EICT. Under IRB approval, 56 patients underwent clinical CT on a PCCT and four EICT scanners from two vendors (A, B). Additionally, the Mercury Phantom 3.0 (GAMMEX Sun Nuclear), representing different adult population sizes, was imaged with the same conditions and scanners. Modulation Transfer Function 10% frequency (MTF-f₁₀), Global Noise Index (GNI), and Noise Power Spectrum average frequency (NPS-f_av), peak frequency (NPS-f_peak), and 10% frequency (NPS-f₁₀) were evaluated in both patients and Mercury Phantom (MP). The related percentage differences between PCCT and EICT images were assessed. Median differences between patients imaged with PCCT and EICT-A were 33.1%, 160.2%, 42.3%, -14.3%, and 25.2% for MTF-f₁₀, GNI, NPS-f_av, NPS-f_peak, and NPS-f₁₀, respectively; differences between patients in PCCT and EICT-B were 11.8%, 197.6%, 35.7%, -14.0%, and 23.2%. For the same metrics, the median differences between MP imaged with PCCT and EICT-A were 256.3%, 105.5%, 48.7%, 141.8%, and 50.1%, respectively; differences between MP scanned with PCCT and EICT-B were 247.8%, 193.5%, 48.5%, 127%, and 47.5%. MP showed different values but similar trends as patient population when comparing PCCT and EICT image quality. Magnitude differences can facilitate adjustment factors to phantom data to improve representation of clinical image quality.

X-ray computed tomography (CT) based on photon counting detectors (PCD) extends standard CT by counting detected photons in multiple energy bins. PCD data can be used to increase the contrast-to-noise ratio (CNR), increase spatial resolution, reduce radiation dose, reduce injected contrast dose, and compute a material decomposition using a specified set of basis materials.1 Current commercial and prototype clinical photon counting CT systems utilize PCD-CT reconstruction methods that either reconstruct from each spectral bin separately, or first create an estimate of a material sinogram using a specified set of basis materials and then reconstruct from these material sinograms. However, existing methods are not able to utilize simultaneously and in a modular fashion both the measured spectral information and advanced prior models in order to produce a material decomposition. We describe an efficient, modular framework for PCD-based CT reconstruction and material decomposition using on Multi-Agent Consensus Equilibrium (MACE). Our method employs a detector proximal map or agent that uses PCD measurements to update an estimate of the pathlength sinogram. We also create a prior agent in the form of a sinogram denoiser that enforces both physical and empirical knowledge about the material-decomposed sinogram. The sinogram reconstruction is computed using the MACE algorithm, which finds an equilibrium solution between the two agents, and the final image is reconstructed from the estimated sinogram. Importantly, the modularity of our method allows the two agents to be designed, implemented, and optimized independently. Our results on simulated data show a substantial (450%) CNR boost vs conventional maximum likelihood reconstruction when applied to a phantom used to evaluate low contrast detectability.

Many methods have been developed to reduce radiation dose in computed tomography (CT) scans without sacrificing image quality. Recently, deep learning-based methods have shown promising results on the task of CT image denoising. However, they remain difficult to interpret, and thus safety concerns have been raised. In this work we develop a method to reconstruct the invariances of arbitrary denoising methods with an approach inspired by the optimization schemes commonly used to generate adversarial examples. We apply our method to one proof-of-principle algorithm as well as to two previously proposed denoising networks and show that it can successfully reconstruct their invariances.

Motion artifacts are among the most important factors degrading the diagnostic performance of x-ray CT images, in particular for photon-counting CT where these artifacts can degrade the higher spatial resolution and quantitative imaging capabilities. The purpose of this simulation study is to evaluate the capability of deep neural networks to correct for motion artifacts in spectral photon-counting cardiac CT, by generating motion-corrected virtual monoenergetic images at a range of different keVs. We used CatSim to generate synthetic training data by simulating motion-corrupted and motion-free CT imaging (100kVp, 1s rotation) of the dynamic XCAT phantom including heart and respiratory motion. In total 2160 image pairs were generated. We trained two different neural networks for the task of estimating motion-artifact-reduced images from motion-corrupted images: one based on UNet and one based on a Wasserstein generative adversarial network with a gradient penalty (WGAN-GP). To make these networks applicable to virtual monoenergetic images at different energies, we trained them with 40keV and 70keV monoenergetic images as inputs and used a loss function with two terms: 1) L1-loss on soft tissue and bone basis images and 2) perceptual loss on 70keV monoenergetic images. Our results show that the motion artifacts from 40keV to 100keV are reduced substantially. In conclusion, these results demonstrate the potential of image-domain deep neural networks to correct for motion artifacts in spectral cardiac CT images.

Deep learning-based image reconstruction and noise reduction (DLIR) methods have been increasingly deployed in clinical CT. Accurate assessment of their data uncertainty properties is essential to understand the stability of DLIR in response to noise. In this work, we aim to evaluate the data uncertainty of a DLIR method using real patient data and a virtual imaging trial framework and compare it with filtered-backprojection (FBP) and iterative reconstruction (IR). The ensemble of noise realizations was generated by using a realistic projection domain noise insertion technique. The impact of varying dose levels and denoising strengths were investigated for a ResNet-based deep convolutional neural network (DCNN) model trained using patient images. On the uncertainty maps, DCNN shows more detailed structures than IR although its bias map has less structural dependency, which implies that DCNN is more sensitive to small changes in the input. Both visual examples and histogram analysis demonstrated that hotspots of uncertainty in DCNN may be associated with a higher chance of distortion from the truth than IR, but it may also correspond to a better detection performance for some of the small structures.

This work introduces a conditional generative model using the denoising diffusion probabilistic model learning strategy to generate iodine maps from single-kV contrast-enhanced CT data. The purpose is to enhance the functionality of single-kV CT scanners, expanding their clinical applications. The model was trained using a clinical dataset of 17,284 paired images from 151 subjects. The testing dataset had 903 image slices from 12 subjects independent of the training set. The model's performance was assessed using quantitative metrics such as RMSE and SSIM, with median values of 0.58 mg/ml and 0.979, respectively. Bland-Altman analysis confirmed the consistency between DECT and the proposed method.

Photon counting detector CT (PCD-CT) offers potential advantages over traditional energy-integrating detector CT (EID-CT), such as enhanced spatial resolution, contrast-to-noise ratio, and spectral imaging capabilities. A crucial characteristic influencing its performance is “dead time.” This metric, introduced in the literature, describes the minimum time required between two photon events for accurate counting statistics. However, determining the dead time of a PCD detector proves challenging because of its dependence on multiple technical factors. This challenge is exacerbated by pulse pileup correction schemes developed by various detector manufacturers and academic researchers. Consequently, a direct experimental method to measure dead time becomes increasingly important in practice. In this study, we present such a few experimental methods to estimate the “effective dead time” of a PCD detector.

Current photon-counting detectors are limited to a pixel size of 0.3mm to 1mm, as decreasing the pixel size further generally introduces degraded dose efficiency and energy resolution from excessive charge sharing. In this work, we present experimental measurements of the first photon-counting detector prototype designed to leverage the charge sharing to estimate the photon interaction position, where simulations indicate a theoretical resolution of around 1μm using a similar geometry. The goal of the measurements is to validate our Monte-Carlo simulation for further development. DAC sweeps are performed with an x-ray beam at specified locations on the sensor front, with the beam at 20keV and 35keV, as well as with different sensor biases with the beam at 35keV. The experimental data are then compared to a Monte Carlo simulation combined with a charge transport model. In this first prototype wire bonds are used, and as such only a few channels are connected. The experimental data agree generally well with the simulated data with the beam close to the electrodes, with the simulated data diverging from the experiments with the beam further away from the electrodes. The induced charge cloud signal exhibits a fairly linear dependency on the beam position, indicating that any estimation techniques will yield more precise position when the photon interacts further away from the electrodes, rather than closer. With the experimental data and the simulations agreeing generally well, together with the same software previously indicating a resolution of around 1μm, we expect an ultra-high-resolution detector to be in reach, and are encouraged to continue development.

Energy-integrating detectors used in diagnostic CT consist of a scintillator that converts x-rays into visible light, and reflective septa that channel the produced visible light and reduce crosstalk between pixels. The reflective septa reduce the fill factor and dose efficiency of CT. It is desirable to reduce the size of the reflective septa, but for mechanical and optical reasons, such size reductions are not easily achieved. For this reason, we propose an alternative strategy: loading the septa with a high-Z material that can fluoresce with the incident x-rays and reemit characteristic x-rays, some of which would be absorbed in adjacent scintillator pixels. We model this approach using Monte Carlo simulations and show that approximately half of the area lost to reflective septa can be recovered with loading of Gd at a density of 1g/cc of Gd. We show that this can translate into higher resolution detectors: in the absence of an anti-scatter grid and assuming a reflective septa thickness of 0.1mm, dose efficiency can be better preserved even as the spatial resolution increases. When moving from a 1.0mm to 0.5mm pixel pitch, the expected reduction in dose efficiency is 26% without x-ray fluorescence loading. With x-ray fluorescence loading, the reduction in dose efficiency is only 4%.

Due to its cost-effectiveness and semiconductive properties, single-crystal Si is a promising photon-counting detector for CT imaging. However, its low Z value causes significant X-ray scattering, affecting resolution and efficiency. High Z metal foils, introduced to absorb these scattered X-rays, reduce detectable photon percentages and radiation dose efficiency. We introduce a hybrid edge-on detector design replacing metal foils with high Z scintillators and Si-based thin-film photodiodes. Our experiments, using a prototype of paired Si and scintillator, demonstrate that scattered X-rays induce scintillation detectable by photodiodes, offering enhanced dose efficiency and signal detection from scintillator layers.

Despite the widespread use of cascaded linear models in x-ray imaging, no publicly available implementation exists as of now. Our goal was to implement the cascaded linear systems theory and create a flexible and publicly available code capable of modeling direct and indirect-conversion x-ray imaging detectors under different acquisition conditions.

The spatial resolution of photon counting detectors (PCDs) can be further improved if photons that contribute to charge sharing (leading to counts in two adjacent pixels) were detected and binned separately. These events are captured in a coincidence counting architecture, previously proposed for improving spectral imaging performance. Here, we examine the problem of using coincidence counts in a non-spectral application for the purpose of improving spatial resolution. With typical PCD parameters, the number of photons that lead to coincidence counts is 3 times fewer than photons that lead to a single count only. Effectively, the detector has an alternating aperture: channels of the detector alternate between being wide and narrow. An analytic reconstruction algorithm that simply normalizes against the air scan will be inefficient at low spatial frequencies because the narrow detector channels contribute equal weight to reconstruction but have worse statistical information. We propose decomposing the sinogram into the narrow and wide apertures components separately, zero-stuffing them, and applying frequency weights so that the detective quantum efficiency at low spatial frequencies is restored, while information content at high spatial frequencies can be accessed. The frequency weighted sinograms are then summed and backprojected. We test this approach in simulations with various detector models on numerical phantoms.

Dark-field imaging with x-rays is a novel interferometric technique that enables visualization of the alveolar structure. In the clinical context, the technique has been limited to radiographic applications and has not yet been implemented for computed tomography (CT) imaging. Initial studies have already demonstrated that dark-field imaging complements and improves conventional radiography of the thorax. Dark-field computed tomography, which is capable of yielding unobstructed 3D views, has only recently been brought to the human scale: With a first prototype system, we demonstrated the feasibility to implement a Talbot-Lau interferometer on a clinical CT system to perform dark-field imaging with clinical acquisition time and field of view. Until now, this prototype was limited to axial scans. In this work, we present our advancements in extending the setups capabilities to also support other modes of acquisition, namely surview and helical scans. The new capabilities of the updated dark-field CT scanner are demonstrated using an anthropomorphic thorax phantom.

When implementing differential phase-contrast imaging with current CT detectors the limited pixel size forces one to scan an analyser-grating in front of the detector to resolve the interference pattern. This simulation study compares the approach of using an analyser-grating and the approach of using a high-resolution detector to directly resolve the interference pattern by applying the non-prewhitening observer on dose-matched simulated CT scan reconstructions of the two approaches. A phantom with two concentric cylinders is used and 1000 CT scans are generated with the inner cylinder present and not present. The non-prewhitening observer is applied to the samples and the ROC curve is extracted, together with the results of a 2AFC test, the CNR, and the detectability. The high-resolution approach shows a 20% increase in the phase AUC compared to the grating-based approach, with a similar increase in the 2AFC test score. The image CNR of the phase shows an increase of 134%, with the detectability increasing by 138% when compared to the grating based approach. An ultra-high-resolution detector, capable of directly resolving the interference pattern of differential phase-contrast imaging could change medical imaging CT as we know it. The implementation in a clinical CT would be simpler, could lower cost, and increase the dose-efficiency due to the obviation of the G2-grating, and at the same time provide an additional two diagnostic signals.

Specimen radiography is critical to cancer diagnostics, but conventional transmission X-ray imaging often lacks the contrast to conclusively differentiate healthy from cancerous tissue. X-ray diffraction (XRD), however, has been shown to be sensitive to structural changes in tissue that are correlated with cancer progression. We use a home-built XRD imaging system to study glioblastoma in brain tissue and ductal carcinoma in situ (DCIS) in breast tissue, and identify both commonalities and unique aspects of cancer structural changes in the brain and breast. These findings demonstrate the potential for XRD imaging as a new tool for use in both research and diagnostic medicine.

In preclinical studies, micro-CT is frequently employed to yield valuable anatomical insights. However, there has been an increasing need for micro-CT in extracting functional measurements such as with perfusion imaging. Perfusion imaging plays a crucial role in understanding and quantifying tissue vascular properties. This paper focuses on our development of preclinical micro-photon counting (PC)CT perfusion imaging and demonstrates quantification of perfusion metrics in a controlled fluid flow phantom experiment. For this study, we utilized a novel bioprinted perfusion phantom and a dedicated preclinical photon-counting CT (PCCT) system to estimate perfusion maps at different flow rates. A continuous water flow through the phantom was maintained by a peristaltic pump. PCCT imaging was performed during a delayed contrast injection of clinical iodinated contrast agent. Imaging was repeated under 3 different flow rates: 4, 6, and 8 mL/min. Our results demonstrate successful visualization and quantification of flow parameters by employing gamma variate curves fit to voxel measurements of temporal PCCT reconstructions as well as decomposed iodine material maps, enabling the calculation of volumetric maps for mean transit time, blood volume index, and blood flow index. Furthermore, we showcase the application of this technique in quantifying in vivo perfusion characteristics in a head and neck cancer model in a mouse. Through our research, we aim to highlight the potential of preclinical micro-PCCT perfusion imaging in advancing our understanding of tissue perfusion dynamics and its potential applications in studying various pathologies and cardiovascular conditions.

The prevalence of kidney stones has significantly risen among the elderly population in recent decades, with some countries experiencing rates approaching 15%. Due to high recurrence rates, analyzing stone composition is essential for recurrence prevention. Fourier transform infrared spectroscopy (FT-IR) and conventional x-ray diffractometry (XRD) are used for this purpose, however, both methods are labor-intensive and require skilled operators, as stones need careful dissection and grinding. Many kidney stones have heterogeneous compositions, and even with spectroscopic analyses, error rates remain high. Despite improvements, challenges in stone sampling and tiny material amounts can still lead to inaccuracies. This prompts the question of how to better manage patients and potentially enhance recurrence prevention strategies despite the shift toward spectroscopic techniques in analytical labs. In this work, we demonstrate the potential of an in-house grating-based high-resolution x-ray phase-contrast μ-CT imaging system in characterizing kidney stones and their compositional analysis. Employing the reported compact benchtop x-ray phase-contrast μ-CT imaging system facilitates stone identification and analysis using additional quantitative data—phase and scattering—obtained from the emerging x-ray phase-contrast imaging system. Having the three sets of images—transmission, phase, and scattering—could potentially pave the way for an accurate and faster stone compositional analysis with clinical values in studying the pathophysiological mechanisms of kidney stone disease that could help improve the recurrence prevention rates.

X-ray “phase-contrast” imaging shows promise for improved image contrast with lower dose, circumventing many disadvantages of traditional diagnostic “absorption-contrast” imaging. However, various challenges must be addressed before it can be realized clinically. One of the most clinically viable phase-contrast imaging techniques is spectral propagation-based imaging (PBI). PBI has the unique advantage of not requiring any specialized optical equipment, and spectral imaging can provide the multiple images necessary to distinguish phase and absorptive properties in a single acquisition. Here, we present a method to quantify the Cramér-Rao lower bound (CRLB) with spectral PBI. We accounted for the spatial dependence of phase effects using a two-dimensional detector and finite difference approximation. We applied this to the task of basis material decomposition with varying propagation distances (0 to 300mm). We modeled two monochromatic sources (24 and 34keV) and an object comprising a 400μm sphere of soft tissue and a 100μm sphere of either bone or adipose tissue. Output CRLBs were used to estimate basis material signal-to-noise ratio (SNR). Without phase contrast, bone SNR was higher than adipose SNR. Both materials’ SNR increased greatly with propagation distance, with the most rapid improvements from 0 to 20mm and a slight decrease in slope above 100mm. This model demonstrates the utility of our spectral PBI CRLB framework and the benefit of phase contrast in tasks with both good and poor absorption. This method can be used to optimize spectral PBI acquisition parameters and has potential to be extended to clinical-scale imaging.

The quantification of breast density, as an evaluation of the fibro-glandular tissue content, has a key role in assessing breast cancer risk and cancer masking in mammography, which are of paramount importance for establishing screening recommendations. Digital breast tomosynthesis (DBT) is a promising imaging technology for breast cancer screening, which allows a pseudo-3D representation of the breast and therefore has the potential to improve detection of breast cancer compared to conventional 2D mammography. We propose a regularization method for DBT reconstruction based on the accretion of adipose and fibro-glandular tissues to provide an accurate breast density estimation. Inspired by the phenomenon of planetary accretion, we established a correspondence between the fraction of each tissue type with the number of particles with attractive interactions, which allows for the redistribution of the tissues, allowing their better identification at each voxel and improving breast density quantification. Our method is combined with a polychromatic reconstruction algorithm with material decomposition. Furthermore, our work offers a non-learning method for DBT glandularity estimation, which does not rely on an accurate recovery of each tissue localization. We evaluated our method using 51 3D digital phantoms based on segmented breast CT patient scans and we found that the error of the reconstructed glandularity was +0.4% on average, ranging between −9.6% and +5.2%, for glandularity values ranging between 0.09 and 0.55; while the relative error was +4.7% on average, ranging between −45.6% and +38.5%.

There are two main acquisition modes for acquiring digital breast tomosynthesis (DBT) projection data: continuous mode and step-and-shoot mode. This work characterizes a new x-ray tube with flying focal spot (FFS) technology, designed to compensate for the continuous focus motion during exposure. The image sharpness of an established wide-angle DBT system with moving focus during exposure (the current standard) and a newly introduced FFS x-ray tube was assessed using the modulation transfer function (MTF). The spatial frequency for the 25% MTF value (f_25%) was measured at 2, 4 and 6cm above the table. The impact on the visibility of calcification-like test objects was investigated with the CDMAM phantom that was positioned at various heights above the table. The threshold gold thickness (T_t) was calculated for the 0.1 mm discs. For the moving focus system, f_25% measured at 2, 4 and 6cm in the 0° DBT projection fell by 35%, 49% and 59% in the tube-travel direction compared to figures of 7.6, 6.8, and 6.6 mm^-1 for the FFS system. In the frontback direction, f_25% was 7.4, 7.3 and 7.2 mm^-1 at 2, 4 and 6cm, respectively for the moving focus unit while for the FFS system, these figures were 7.1, 6.7 and 5.9 mm^-1. There was no significant difference in f_25% for the front-back and tube-travel directions for the FFS unit (p<0.04). The T_t values at 4cm and 6cm above the table in DBT mode improved by 33% and 52%, respectively for the FFS system versus the moving focus system. The improvement in T_t increased as object height above the table increased. The FFS system has a reduced DBT scan time compared to the moving focus system (4.8s vs 21.7s) which is expected to reduce patient motion artefacts. The FFS system has in DBT mode an improved isotropic resolution, improved small detail detectability and faster total scan times. Therefore, this new DBT system has the potential for increased microcalcification detection and patient throughput compared to its predecessors.

Wide–angle digital breast tomosynthesis (DBT) is well known to offer benefits in mass perceptibility compared to narrow–angle DBT due to reduced anatomical overlap. Regarding the perceptibility of micro–calcifications the situation is somehow inverted. On the one hand this can be related to effects during data acquisition and their impact on the system MTF. On the other hand there is a wider spread of calcifications in depth direction in narrow–angle DBT, which distributes calcifications over more slices. This is equivalent to an inherent thicker slice for high spatial frequencies. In this work we want to assume an equivalent quality of raw data and only focus on the effects of different acquisition angles in the reconstruction. We propose an algorithm which creates so–called hybrid thick DBT slices and optimizes the visualization of calcifications while preserving the high mass perceptibility of thin wide–angle DBT slices. The algorithm is purely based on filtered backprojection (FBP) and can be implemented in an efficient manner. For validation simulation studies using the VICTRE (FDA) pipeline are performed. Our results indicate that hybrid thick–slices in wide-angle DBT enable to successfully solve the contrarian imaging tasks of high mass and high calcification perception within one imaging setup.

Virtual clinical trials (VCTs)and physical phantom studies can be helpful in the evaluation of new breast imaging technology.^{1, 2} These methods are usually designed to provide an objective assessment of image quality by evaluating task-based performance from an observer (e.g., human or model observer). Previous VCTs and phantom studies typically use a detection task (often a signal-known exactly task) to assess performance. However, real clinical studies are usually more complicated than the simple task of lesion detection. Most breast imaging clinical studies reported in the literature involve the observer both detecting whether a suspicious lesion is present, as well as rating how likely it is that the lesion is malignant. In fact, the decision variable commonly used in clinical breast imaging studies to form a receiver operating characteristic (ROC) curve is the probability of malignancy (POM). Designing VCTs and physical phantom studies that accurately model clinical studies is challenging because it is difficult to model malignant and benign lesions. One lesion feature that that is commonly used to discriminate malignant and benign lesions is the border of the lesion, with malignant lesions typically portraying spiculations. To more accurately model clinical breast imaging studies, we explore the use of a classification task instead of a detection task, that evaluates the reader’s performance in discriminating between spiculated and lobular (non-spiculated) masses. To assess this classification task, we use a convolutional neural network (CNN) model trained to differentiate spiculated and non-spiculated masses using Monte Carlo simulated images. The basis for using CNN observer models is that a previous study showed that the CNN based framework can be trained to approximate the ideal observer.³ In addition to the mass classification task, we also examine this CNN model observer in a more traditional detection task. This study evaluates the effect of dose and resolution and shows that different conclusions might result depending on whether the VCT model evaluates a mass detection task or a mass classification task.

Contrast-enhanced mammography (CEM) offers a promising alternative to address the limitations of digital mammography, particularly in cases of dense breast tissue, which compromises the performance of non-contrast x-ray imaging modalities. CEM uses iodinated contrast material to enhance cancer detection in denser breasts and provides critical functional information about suspicious findings. However, the process of combining images acquired with different x-ray energy spectra in CEM can introduce artifacts, challenging interpretation and confidence in CEM images. This study presents novel approaches to improve CEM image quality. First, deep learning (DL)-based algorithms for scatter correction in both low-energy and high-energy images are proposed to enhance contrast-enhancement patterns and iodine quantification. Additionally, a unique deep learning network is introduced to predict pixel-by-pixel the compressed breast thickness, enabling the use of local thickness-based image subtraction-weighting maps throughout the breast area. Results in phantom cases demonstrate the effectiveness of the scatter correction models in predicting the scatter signal, even in cases with the anti-scatter grid present. The thickness map model accurately estimates the local thickness, particularly in the constant thickness area of the breast. Comparison with clinical data revealed good agreement between estimated thickness maps and ground truth, with minor discrepancies attributed to alignment issues. Furthermore, the study explored the combined use of scatter correction and thickness-based weighting maps in creating recombined CEM images. This approach showed a marginal positive impact due to scatter correction, with larger improvements observed in the signal intensity homogeneity at the border of the breast. These advancements aim to enhance the CEM diagnostic accuracy, making it a valuable tool for breast cancer detection and evaluation, especially in cases with dense breast tissue.

Single-exposure, dual-energy x-ray angiography using photon-counting x-ray detectors is a potential alternative to kV-switching dual-energy angiography and digital subtraction angiography (DSA), but is unable to simultaneously suppress anatomic noise from soft tissue and bone. Triple-energy, photon-counting angiography that counts photons in three energy bins during a single x-ray exposure would overcome this limitation but high quantum noise levels compromise image quality. We extended anti-correlated noise reduction (ACNR) for dual-energy angiography to triple-energy photon-counting angiography and evaluated the resulting improvement in iodine signal-difference-to-noise ratio (SDNR). We implemented triple-energy photon-counting imaging of iodine using a cadmium-telluride, photon-counting x-ray detector with analog charge summing for charge-sharing correction. We imaged a phantom consisting of vessel-like structures with diameters of 3mm, 4mm and 5mm embedded in background clutter. The vessels were filled with an iodine solution containing 100mg/ml of iodine. We acquired images at 60kV, 80kV and 100kV using energy thresholds that theoretically maximized SDNR per root entrance air kerma without ACNR. For each tube voltage, we simulated three energy bins by acquiring two separate exposures using two different sets of energy thresholds. The triple-energy images with ACNR had SDNR that was approximately 7.5 times greater than those without. Further increases in ACNR are expected with optimization of tube voltage and energy thresholds in the presence of ACNR. Future work will focus on optimization and frequency-dependent image-quality assessment.

Real-time brain perfusion would be highly beneficial to fully leverage interventional procedures such as endovascular thrombectomy in treating ischemic stroke. However, the existing standard of care is constrained due to the absence of intra-operative real-time perfusion imaging. We propose an intra-interventional perfusion with no C-arm rotation (IPEN)-based method to perform quantitative perfusion analysis using x-ray angiography (XA) images in the interventional suite. IPEN is built using tomographic principles, where the value of N homogeneous regions-of-interests (ROI) of known shape can be estimated using N x-ray beams from a single projection. The objective of IPEN for brain perfusion is to provide dynamic time-enhancement curve (TEC) and temporal perfusion indices (PI’s) for ROIs within the brain. The proposed algorithm was assessed using XCAT phantoms developed for dynamic CT perfusion. Eight ischemic stroke cases were simulated and for each of them intra-arterially enhanced XA images were acquired. Using these XA images IPEN was employed to estimate TEC and PI such as time-to-maximum (Tmax) and mean-transit-time (MTT) for ROIs defined. The accuracy of the estimated values was quantitatively evaluated using the normalized root-mean-square error weighted by ROI’s fraction-by-volume (NRMSE) and the linear correlation coefficient (R). The estimated TEC, and PI’s showed good agreement with ground truth values. The NRMSE values were 15.5% ± 1.2% for TEC, 14.5% ± 5.7% for Tmax, and 12.2% ± 5.1% for MTT. The R values (P<0.01) across eight cases ranged 0.82 to 0.95 for TEC, 0.70 to 0.99 for Tmax, and 0.67 to 0.96 for MTT.

Portable chest x-ray (PCXR) is commonly employed in intensive care units (ICU). CXR provides valuable information but is hindered by structural overlap. Digital chest tomosynthesis (DCT) removes overlap, yielding quasi-three- dimensional images. Spatially distributed x-ray source arrays eliminate complex mechanical motion and are well-suited for mobile systems. The purpose of this study is to design and optimize a compact and portable DCT system employing a linear field emission x-ray source array in lieu of a conventional moving source, balancing image quality with size and clinical practicality. A prototype stationary DCT system was used to test a variety of system configurations. Tomosynthesis scans of an anthropomorphic phantom were performed repeatedly to generate an image ensemble. Subsets were selected for reconstruction to investigate the impact of spatial distribution on image quality. Reconstruction slice images were evaluated by visualization of simulated lung nodules and presence of artifact to determine the optimal source configuration. Reconstruction slice images demonstrated out-of-plane artifact in configurations with low projection image density. Narrow angular span compromised the visualization of structures and low contrast features. Of the evaluated configurations, one was determined to best balance image quality with physical constraints, utilizing 15 projections covering an angular span of 23 degrees. Bedside 3D imaging is achievable by use of x-ray source arrays. A configuration was determined that generates good image quality in a size conducive to design of a mobile tomosynthesis system. Such a system should provide increased clinical utility compared to 2D radiography in the ICU.

Spectral radiography and fluoroscopy with multi-layer flat-panel detectors (FPD) is being actively investigated in a range of clinical applications. For applications involving contrast administration, maximal contrast resolution is achieved when overlaying background anatomy is completely removed. This calls for three-material decomposition (soft tissue, bone, and contrast) enabled by measurements in three energy channels. We have previously demonstrated the feasibility of such decomposition using a triple-layer detector. While algorithmic solutions can be adopted to mitigate noise in the material basis images, in this work, we seek to fundamentally improve the conditioning of the problem through optimized system design. Design parameters include source voltage, the thickness of the top two CsI scintillators, and the thickness of two copper interstitial filters. The design objective is to minimize noise in the basis image containing contrast, chosen as gadolinium in this work to improve separation from bone. The optimized design was compared with other designs with unoptimized scintillator thickness and/or without interstitial filtration. Results show that CsI thickness optimization and interstitial filtration can significantly reduce noise in the gadolinium image by 35.7% and 42.7% respectively within a lung ROI, which in turn boosts detectability of small vessels. Gadolinium and bone signals are separated in all cases. Visualization of coronary vessels is enabled by the combining optimized system design and regularization. Results from this work demonstrate that three-material decomposition can be significantly improved with system design optimization. Optimized designs obtained from this work can inform imaging techniques selection and triple-layer detector fabrication for spectral radiography.

Deep learning algorithms have been applied to train denoising models for radiographs. Since obtaining full-dose reference images and low-dose training images through repeated scanning is not preferable or accessible, researchers simulate low-dose images through noise insertion. Traditionally, only white/uncorrelated noise is inserted into training images. However, the noise of flat-panel energy-integrated detectors is spatially correlated. In this work, we investigated the impact of spatial correlation on deep learning-based denoising. We applied spatial noise kernels of different strengths to the noise during noise insertion. Then, we compared the performance of denoising models that are trained with either uncorrelated-noise images or correlated-noise images. When the noise characteristic (uncorrelated or correlated) of test images matches that of the training images, the model can output images that are closest to the high-quality reference image. This implies that spatial correlation should be taken into consideration during image simulation and denoising model training.

Over the last 17 years Mayo Clinic has become a world leader in a field now known as point of care manufacturing. Using additive manufacturing we focus on five distinct areas. First to create diagnostic anatomic models for each surgical subspecialty from diagnostic imaging. Second to manufacture custom patient-specific sterilizable osteotomy cutting guides for ENT, OMFS, Orthopedics, and Orthopedic Oncology. Third to build simulators and phantoms using a combination of special effects and 3Dprinting. Fourth using 3D printers to create custom phantoms, phantom holders, and other custom medical devices such as pediatric airway devices, proton beam appliances, and custom jigs and fixtures for the department and hospital. Finally to transfer the digital twins into virtual and augmented reality environments for preoperative surgical planning and immersive educational tools. Mayo Clinic has scaled this endeavor to all three of its main campuses including Jacksonville Fl and Scottsdale AZ to complete the enterprise approach. In doing so we have been able to advance patient care locally as well as assist in building the national IT, regulatory, billing, RSAN 3D SIG, and quality control infrastructure needed to assure scaling across this and other countries.

Engineers often design products to work within available test fixtures. Test fixtures define the goal posts for device evaluation. It is important for test fixtures to accurately represent the critical conditions of operation and be supported with justification for regulatory review. This presentation explores the role of 3D printing and model design workflows in producing anatomically relevant text fixtures which can be used to guide, and more importantly accelerate the device development process. The Jacobs Institute is a one-of-a-kind, not-for-profit vascular medical technology innovation center. The Jacobs Institute’s mission is to accelerate the development of next-generation technologies in vascular medicine through collisions of physicians, engineers, entrepreneurs, and industry.

AM is already a widely adopted manufacturing process used to produce millions of medical devices and healthcare products every year. Common uses for AM include the printing of patient-specific surgical implants and instruments derived from imaging data and the manufacturing of metal implants and instruments with features that are impossible to fabricate using traditional subtractive manufacturing. In addition to reducing costs, patient-specific solutions—such as customized surgical plans and personalized implants—aim to improve surgical outcomes for patients and give surgeons more options and more flexibility in the OR. With advancement in technology, implants are 3D printed in various materials and at various manufacturing sites including at the point-of-care. 3D Systems collaborates with medical device manufacturers and health systems to develop personalized health solutions and is the leader in design, manufacturing and getting regulatory approvals for 3D printed patient-specific implants in various materials and technologies

Additive Manufacturing has not realized it’s full potential due to a number of factors. The regulatory environment for medical devices is geared towards conventional manufacturing techniques, making it challenging to certify 3D-printed devices. Additive manufacturing may still not be competitive when scaled up for industrial production, and the need for post-processing may negate some of the benefits. The promises and the challenges of additive manufacturing will be explored in the context of medical imaging device design.

Single photon emission computed tomography (SPECT) can enable the quantification of activity uptake in lesions and at-risk organs in α-particle-emitting radiopharmaceutical therapies (α-RPTs). However, this quantification is challenged by the extremely low detected photon counts, complicated isotope physics, and the image-degrading effects in α-RPT SPECT. Thus, strategies to optimize the SPECT system and protocol designs for the task of regional uptake quantification are much needed. Objectively performing this task-based optimization requires a reliable (accurate and precise) regional uptake quantification method. Conventional reconstruction-based quantification (RBQ) methods have been observed to be erroneous for α-RPT SPECT. Projection-domain quantification methods, which estimate regional uptake directly from SPECT projections, have demonstrated potential in providing reliable regional uptake estimates, but these methods assume constant uptake within the regions, an assumption that may not hold. To address these challenges, we propose Wiener INtegration Projection-Domain Quantification (WIN-PDQ), a Wiener-estimator-based projection-domain quantitative SPECT method. The method accounts for the heterogeneity within the regions of interest while estimating mean uptake. An early-stage evaluation of the method was conducted using 3D Monte Carlo-simulated SPECT of anthropomorphic phantoms with ²²³Ra uptake and lumpy-model-based intra-regional uptake heterogeneity. In this evaluation with phantoms of varying mean regional uptake and intra-regional uptake heterogeneity, the WIN-PDQ method yielded ensemble unbiased estimates and significantly outperformed both reconstruction-based and previously proposed projection-domain quantification methods in terms of normalized root ensemble mean squared error. In conclusion, based on these preliminary findings, the proposed WIN-PDQ method is showing potential for estimating mean regional uptake in α-RPTs and towards enabling the objective task-based optimization of SPECT system and protocol designs.

Dual-energy CT (DECT) plays a vital role in quantitative imaging applications for its capability of material decomposition. Matrix inversion-based material decomposition suffers from severe degradation of signal-to-noise ratio (SNR) on the resultant images. Iterative decomposition methods perform noise suppression through regularization but the artificial image prior cannot characterize all inherent features of noiseless material-specific images. Supervised-learning material decomposition methods train a neural network to fully depict the mapping relationship between DECT images and decomposed images. However, large-size paired DECT and noiseless material-specific images are not always available in clinical practice, hindering the practice of supervised-learning methods. In this work, we propose a practical unsupervised-learning framework for DECT material decomposition with sinogram fidelity and the preliminary results show the feasibility and potential of the proposed method in quantitative applications.

Metal Artifact Reduction (MAR) is one of the notorious problems in dental CT imaging. The presence of metallic implants often introduces severe metal artifacts in the reconstructed CT images, which obstruct the visualization of dental structures. This highly ill-posed problem has been addressed with the rise of artificial intelligence in respect to deep learning. However, majority of the approaches rely on the supervision of large paired dataset, which is often infeasible in the clinical practice. In this work, we present NeMAR, a Neural fields-based Metal Artifact Reduction method for dental CT. NeMAR leverages coordinate-based neural representation along with two key components: the masked loss and the regularization loss. These elements synergistically empower the neural fields to generate metal-artifact-reduced CT image with high fidelity. Notably, NeMAR requires only the original metal-artifact-corrupted image as an input, thus eliminating the need for extensive paired data. The validation using simulated dental CT datasets demonstrates the effectiveness of NeMAR in accurately recovering the shape of dental structures. In essence, NeMAR presents a promising data-free approach to enhance dental CT imaging.

We exploit the development of a direct conversion intraoral sensor (IOS) utilizing a novel material, Perovskite (PV), employing a CMOS (Complementary metal-oxide-semiconductor) readout array with a pixel pitch of 14.8um for dental x-ray imaging. A 100um-thick Perovskite film was fabricated onto a CMOS readout array using a powder-based LTVAC (low-temperature and vacuum-assisted crystallization) process developed in this study. Fundamental x-ray imaging characteristics were assessed in terms of MTF (modulation transfer function) and DQE (detective quantum efficiency) and compared with a CsI-based commercial IOS using sample images under dental x-ray spectrum. The developed PV-based IOS demonstrated >20% MTF at the 33.8 lp/mm Nyquist frequency of 14.8um pixel device and superior DQE performance above a spatial frequency of 9 lp/mm, offering enhanced x-ray imaging capabilities compared to CsI-based indirect conversion commercial IOS products.

Sparse CT reconstruction continues to be an area of interest in a number of novel imaging systems. Many different approaches have been tried including model-based methods, compressed sensing approaches, and most recently deep-learning-based processing. Diffusion models, in particular, have become extremely popular due to their ability to effectively encode rich information about images and to allow for posterior sampling to generate many possible outputs. One drawback of diffusion models is that their recurrent structure tends to be computationally expensive. In this work we apply a new Fourier diffusion approach that permits processing with many fewer time steps than the standard scalar diffusion model. We present an extension of the Fourier diffusion technique and evaluate it in a simulated breast cone-beam CT system with a sparse view acquisition.

Phantom-based quality control, the current standard of QC in medical imaging, calibrates image quality at a population level, but does not account for the influence of patient variation on quality. In this work, we present a method to evaluate task-based image quality directly in individual clinical CT exams. Noise power spectrum (NPS) is measured in selected local image regions satisfying linearity and noise stationarity constraints, and globally over the volumetric image. Together with a semi-empirical model of image resolution, NPS is used to calculate noise-equivalent quanta (NEQ), a fundamental metric of image fidelity and information content. The NEQ may be extended to task-based detectability (d’) via a specified task function and model observer. We show that this method can: 1) elucidate intra-patient variations in signal detectability, and 2) task performance variations across a patient population. The method may be implemented in a hospital-wide online system that monitors imaging performance in CT exams in real-time.

Sparse-view computed tomography (CT) has great potential in reducing radiation dose and accelerating the scan process. Although deep learning (DL) methods have exhibited promising results in mitigating streaking artifacts caused by very few projections, their generalization remains a challenge. In this work, we proposed a DL-driven alternative Bayesian reconstruction method that efficiently integrates data-driven priors and the data consistency constraints. This methodology involves two stages: universal embedding and consistency adaptation respectively. In the embedding stage, we optimize DL parameters to learn and eliminate the general sparse-view artifacts on a large-scale paired dataset. In the subsequent consistency adaptation stage, an alternative Bayesian reconstruction further optimizes the DL parameters according to individual projection data. Our proposed technique is validated within both image-domain and dual-domain DL frameworks leveraging simulated sparse-view (90 views) projections. The results underscore the superior generalization and context structure recovery of our approach compared to networks solely trained via supervised loss.

Diffusion models have been demonstrated as powerful deep learning tools for image generation in CT reconstruction and restoration. Recently, diffusion posterior sampling, where a score-based diffusion prior is combined with a likelihood model, has been used to produce high quality CT images given low-quality measurements. This technique is attractive since it permits a one-time, unsupervised training of a CT prior; which can then be incorporated with an arbitrary data model. However, current methods only rely on a linear model of x-ray CT physics to reconstruct or restore images. While it is common to linearize the transmission tomography reconstruction problem, this is an approximation to the true and inherently nonlinear forward model. We propose a new method that solves the inverse problem of nonlinear CT image reconstruction via diffusion posterior sampling. We implement a traditional unconditional diffusion model by training a prior score function estimator, and apply Bayes rule to combine this prior with a measurement likelihood score function derived from the nonlinear physical model to arrive at a posterior score function that can be used to sample the reverse-time diffusion process. This plug-and-play method allows incorporation of a diffusion-based prior with generalized nonlinear CT image reconstruction into multiple CT system designs with different forward models, without the need for any additional training. We demonstrate the technique in both fully sampled low dose data and sparse-view geometries using a single unsupervised training of the prior.

Automatic exposure control based on tube current modulation (TCM) can effectively reduce dose while maintaining image quality. Conventional TCM uses total exposure from the tube and noise in the center of CT slices as surrogates of dose and image quality, respectively. In this abstract, we present an automated method to optimize TCM at the organ level, offering increased flexibility and aligning with the concept of organ-specific radiation risk assessment. We applied our method to a retrospective CT dataset and incorporated automatic organ segmentation, Monte Carlo simulation for dose calculation, and an empirical model for noise estimation. This method was fully automated and readily scalable to massive clinical data, allowing the generation of ground-truth data for any data-driven approach to prospective planning, including methods utilizing scout images.

Multi-source array (MXA) Computed Tomography systems pose challenges related to sampling and x-ray scatter. We present a semi-stationary head CT system and image formation pipeline including adaptive scatter estimation and image reconstruction based on learned diffusion models. The CT was evaluated on a robotic bench system including a miniaturized carbon-nanotube x-ray source and a curved-panel detector. Scatter correction was achieved with an Adaptive Deep Scatter Estimation (ADSE) method combining geometry-invariant projection-based scatter estimation with geometry-adaptive registration and scaling. Image reconstruction followed a Diffusion Posterior Sampling method (DPS-Recon) combining an unconditional diffusion model with measured data consistency. Image quality was assessed using anthropomorphic phantoms for a semi-stationary protocol involving a 21-source MXA rotated to three positions. ADSE resulted in 118% mean increase in feature contrast accuracy, 1.75 to 13-fold improvement in CNR for variable contrast features (-337HU to 885HU), and 3.56-fold improvement in CNR for variable size features (2mm to 12mm, 110HU) compared to uncorrected reconstructions. Non-uniformity reduced 50% for the three slices. DPS-Recon reduced limited sampling artifacts and improved visualization of soft-tissue structures, particularly in less densely sampled and bony anatomy locations, and further reduced non-uniformity by 20% in the superior brain location. We present first experimental results from a semi-stationary, multi-source CT utilizing CNT x-ray sources and curved-panel detector coupled to an imaging chain that addressed the main challenges inherent to the architecture. Metrics of CT number accuracy, image uniformity, and soft-tissue visualization showed promising performance for visualization of stroke radiological markers with the proposed approach.

Dual-energy CBCT acquires projection images using two different x-ray spectra, enabling material quantification and synthesis of virtual monoenergetic images to reduce metal artifacts. Current methods for dual-energy CBCT imaging include using either a fast kVp switching x-ray source or a dual-layer detector. In this work, we demonstrate a novel approach to dual-energy CBCT by spectral filtration of an x-ray source that contains two focal spots. The method generates two distinct x-ray spectra and independently programmable dose and dose rate for the two energies, without a significant cost increase over a conventional CBCT. Two spectral filters were attached to the x-ray exit windows of a dual-focus carbon nanotube x-ray source operating at a constant tube voltage. As the source and detector rotate around the object, alternating high and low energy projections were acquired by alternatively activating electron emission from the two cathodes. A one-step material decomposition based on separable quadratic surrogate functions was implemented. A water equivalent plastic phantom with calcium and iodine contrast inserts was imaged to evaluate the accuracy of material decomposition. Results demonstrate a functional low-cost dual-energy CBCT system by spectral filtration, with the accuracy for material quantification comparable to those from the dual-layer detector and fast kVp switching based DE-CBCT.

We proposed a deep learning-based method for single-heartbeat 4D cardiac CT reconstruction, where a single cardiac cycle was split into multiple phases for reconstruction. First, we pre-reconstruct each phase using the projection data from itself and the neighboring phases. The pre-reconstructions are fed into a supervised registration network to generate the deformation fields between different phases. The deformation fields are trained so that it can match the ground truth images from the corresponding phases. The deformation fields are then used in the FBP-and-wrap method for motion-compensated reconstruction, where a subsequent network is used to remove residual artifacts. The proposed method was validated with simulation data from 40 4D cardiac CT scans and demonstrated improved RMSE and SSIM and less blurring compared to FBP and PICCS.

Deep Learning (DL) image synthesis has gained increasing popularity for the reconstruction of CT and cone-beam CT (CBCT) images, especially in combination with physically-principled reconstruction algorithms. However, DL synthesis is challenged by the generalizability of training data and noise in the trained model. Epistemic uncertainty has proven as an efficient way of quantifying erroneous synthesis in the presence of out-of-domain features, but its estimation with Monte Carlo (MC) dropout requires a large number of inference runs, variable as a function of the particular uncertain feature. We propose a single-pass method–the Moment Propagation Model–which approximates the MC dropout by analytically propagating the statistical moments through the network layers, removing the need for multiple inferences and removing errors in estimations from insufficient dropout realizations. The proposed approach jointly computes the change of the expectation and the variance of the input (first two statistical moments) through each network layer, where each moment undergoes a different numerical transformation. The expectation is initialized as the network input; the variance is solely introduced at dropout layers, modeled as a Bernoulli process. The method was evaluated using a 3D Bayesian conditional generative adversarial network (GAN) for synthesis of high-quality head MDCT from low-quality intraoperative CBCT reconstructions. 20 pairs of measured MDCT volumes (120kV, 400 to 550mAs) depicting normal head anatomy, and simulated CBCT volumes (100 to120kV, 32 to 200mAs) were used for training. Scatter, beam-hardening, detector lag and glare were added to the simulated CBCT and were corrected (assuming unknown) prior to reconstruction. Epistemic uncertainty was estimated for 30 heads (outside of the training set) containing simulated brain lesions using the proposed single-pass propagation model, and results were compared to the standard 200-pass dropout approach. Image quality and quantitative accuracy of the estimated uncertainty of lesions and other anatomical sites were further evaluated. The proposed propagation model captured >2HU increase in epistemic uncertainty caused by various hyper- and hypo-density lesions, with <0.31HU error over the brain compared to the reference MC dropout result at 200 inferences and <0.1HU difference to a converged MC dropout estimate at 100 inference passes. These findings indicate a potential 100-fold increase in computational efficiency of neural network uncertainty estimation. The proposed moment propagation model is able to achieve accurate quantification of epistemic uncertainty in a single network pass and is an efficient alternative to conventional MC dropout.

Additive manufacturing (i.e. 3D printing) offers transformative potential in the development of biomedical devices and medical imaging systems, but at the same time presents challenges that continue to limit widespread adoption. Within medical imaging, 3D printing has numerous applications including device design, radiographic collimation, anthropomorphic phantoms, and surgical visualization. Continuous technological development has resulted in improved plastic materials as well as high-throughput fabrication in medical-grade metal alloys. Nonetheless, additive manufacturing has not realized its full potential, due to a number of factors. The regulatory environment for medical devices is geared towards conventional manufacturing techniques, making it challenging to certify 3D-printed devices. Additive manufacturing may still not be competitive when scaled up for industrial production, and the need for post-processing may negate some of the benefits. In this talk, we will describe the current state of 3D printing in medical imaging and explore future potential, including links to 3D design and finite-element modeling.

Respiratory diseases impose an immense worldwide health burden. Traditional high resolution computed tomography is regarded as the gold standard for structural lung imaging but does not routinely provide functional information due to cardiac and respiratory motion artifact. We aim to develop and evaluate a low-dose dynamic CT technique with temporal resolution as fast as 66 ms for airway imaging using a dual-source photon-counting-detector (PCD)-CT. An inflatable pig lung phantom that can simulate respiratory cycle was used. On the back of the phantom small spheres with different radius were attached for quantification. Phantom was imaged with two scanning modes, a dual-source prospectively gated adaptive sequential mode with slow (30 bpm) electrocardiogram (ECG) signal and standard dynamic cine mode on a PCD-CT. Motion artifacts were reduced on images acquired from slow ECG-gated dynamic images compared to traditional cine. Improved temporal resolution led to more accurate quantitative measurements. Compared to standard dynamic CT the proposed slow ECG-gated dynamic imaging on PCD-CT can achieve a better temporal resolution, reduces motion artifacts and improves quantification.

Pulmonary emphysema is a progressive lung disease that requires accurate evaluation for optimal management. This task, possible using quantitative CT, is particularly challenging as scanner and patient attributes change over time, negatively impacting the CT-derived quantitative measures. Efforts to minimize such variations have been limited by the absence of ground truth in clinical data, thus necessitating reliance on clinical surrogates, which may not have one-to-one correspondence to CT-based findings. This study aimed to develop the first suite of human models with emphysema at multiple time points, enabling longitudinal assessment of disease progression with access to ground truth. A total of 14 virtual subjects were modeled across three time points. Each human model was virtually imaged using a validated imaging simulator (DukeSim), modeling an energy-integrating CT scanner. The models were scanned at two dose levels and reconstructed with two reconstruction kernels, slice thicknesses, and pixel sizes. The developed longitudinal models were further utilized to demonstrate utility in algorithm testing and development. Two previously developed image processing algorithms (CT-HARMONICA, EmphysemaSeg) were evaluated. The results demonstrated the efficacy of both algorithms in improving the accuracy and precision of longitudinal quantifications, from 6.1±6.3% to 1.1±1.1% and 1.6±2.2% across years 0 to 5. Further investigation in EmphysemaSeg identified that baseline emphysema severity, defined as >5% emphysema at year 0, contributed to its reduced performance. This finding highlights the value of virtual imaging trials in enhancing the explainability of algorithms. Overall, the developed longitudinal human models enabled ground-truth based assessment of image processing algorithms for lung quantifications.

Cardiac CT (CCT) is of vital importance in heart disease diagnosis but is conventionally limited by its complex workflow that requires dedicated phase and bolus monitoring devices [e.g., electrocardiogram (ECG) gating]. Our previous work has demonstrated the possibility of replacing ECG devices with deep learning (DL)-based monitoring of continuously acquired pulsed mode projections (PMPs, i.e., only a few sparsely sampled projections per gantry rotation). In this work, we report the development of a new projection domain DL-based cardiac phase estimation method that uses ensemble learning [i.e., training multiple convolution neural network (CNN) in parallel] to estimate and reduce DL uncertainty. The estimated DL uncertainty information was then used drive an analytical regularizer in a principled time-dependent manner (i.e., stronger regularization when DL uncertainty is higher). Combined with our previous work on PMP-based bolus curve estimation, the proposed method could potentially be used to achieve autonomous cardiac scanning in a robust (i.e., reduced uncertainty) manner without ECG and bolus timing devices.

CT diagnostic imaging is a major contributor to ionizing radiation exposure in the United States. Unfortunately, a reduction in radiation dose often results in degraded image quality. Automatic Exposure Control (AEC) is the most commonly used method to balance image quality and dose in x-ray CT, generally by modifying the scan’s tube current modulation (TCM) parameters. To allow current AEC techniques to be better personalized to the patient size, organ dose, and clinical task, our team previously proposed Scout-Dose and Scout-IQA to prospectively estimate dose and noise from frontal and lateral scouts, scan range, and TCM map. In this study, we evaluate for the first time the performance of our scout-based organ dose and noise predictions in an optimization framework to prospectively determine real-time, personalized TCM maps from a patient’s acquired scouts and scan ranges.

Spectral computed tomography (CT) is a powerful diagnostic tool offering quantitative material decomposition results that enhance clinical imaging by providing physiologic and functional insights. Iodine, a widely used contrast agent, improves visualization in various clinical contexts. However, accurately detecting low-concentration iodine presents challenges in spectral CT systems, particularly crucial for conditions like pancreatic cancer assessment. In this study, we present preliminary results from our hybrid spectral CT instrumentation which includes clinical-grade hardware (rapid kVp-switching x-ray tube, dual-layer detector). This combination expands spectral datasets from two to four channels, wherein we hypothesize improved quantification accuracy for low-dose and low-iodine concentration cases. We modulate the system duty cycle to evaluate its impact on quantification noise and bias. We evaluate iodine quantification performance by comparing two hybrid weighting strategies alongside rapid kVp-switching. This evaluation is performed with a polyamide phantom containing seven iodine inserts ranging from 0.5 to 20mg/mL. In comparison to alternative methodologies, the maximum separation configuration, incorporating data from both the 80kVp, low photon energy detector layer and the 140kVp, high photon energy detector layer produces spectral images containing low quantitative noise and bias. This study presents initial evaluations on a hybrid spectral CT system, leveraging clinical hardware to demonstrate the potential for enhanced precision and sensitivity in spectral imaging. This research holds promise for advancing spectral CT imaging performance across diverse clinical scenarios.

The advent of clinical photon counting x-ray CT (PCCT) has yielded gains in spatial resolution, noise performance, contrast resolution, and radiation dose management in numerous diagnostic applications. Pediatric populations, especially, stand to benefit from these gains due to their smaller size relative to adult patients and to greater concerns over lifetime cancer risk associated with ionizing radiation exposure. No where is this more applicable than in the congenital heart population who are repeatedly imaged during their care. Despite these potential advantages, limited protocol optimizations for pediatric patients, the heterogeneity of metallic implants in pediatric cardiac patients, and trade-offs between image quality and dose made by lowering kV and mAs values contribute to variable image quality in pediatric PCCT. Here, we adapt and demonstrate two deep-learning strategies for denoising pediatric cardiac PCCT data which do not require paired training data and which can be trained using data sets of varying image quality. We train a CycleGAN to map CT images between high and low image quality, as gauged by signal-to-noise ratio measurements. The domain mappings are performed with vision transformers (ViTs) which adeptly preserve high spatial frequencies and enable 50% noise reduction across high and low intensity structures (enhanced vasculature, fat). We adapt prior work using masked autoencoders and 2D natural images to the domain of 3D CT data. Specifically, we propose a ViT network structure and cost function which yields robust interpolation of deleted 3D CT data, similarly reducing image noise by 50% in an assessment of validation data. Comparing CycleGAN and masked autoencoder results, the CycleGAN reduces noise by identifying and attenuating high frequency features associated with noise while the masked autoencoder adeptly removes noise textures but visibly smooths high-contrast structures.

Spectral CT systems enable material decomposition (MD) and generate material-specific images and virtual monoenergetic images (VMIs). Multi-energy CT (MECT) phantoms with accurate attenuation properties enable image quality evaluation and optimization of these spectral CT systems. Previous works have proposed radiopaque 3D printing methods that print potassium iodide (KI) solution on different types of papers using office inkjet printers to create realistic customized patient phantoms of different anatomies. However, these methods failed to reproduce the accurate energy dependent x-ray attenuation properties of different human tissues and can only be used in conventional CT systems. The KI solution used in this inkjet printing technique also results in clogged printer nozzles and leads to inconsistent printed structures and high cost of printer maintenance. In this work, we propose a 3D anthropomorphic MECT phantom generation process with office laser printers and customized iron oxide toner. These phantoms reproduce realistic patient anatomical structures and accurate energy dependent attenuation properties of different tissue types. We printed a 5 mm thick slice of a non-contrast anthropomorphic brain phantom to validate our method. This printed phantom was imaged using two different x-ray spectra, which were used to form VMIs. Our results show that our method can print customized anthropomorphic MECT phantoms for spectral imaging with accurate energy-dependent x-ray attenuation properties.

Virtual imaging trials of malignancies require realistic models of lesions. The purpose of this study was to create hybrid lesion models and associated tool incorporating morphological and textural realism. The developed tool creates a lesion morphology based on input parameters describing its shape and spiculation. Internal heterogeneity is added as 3D clustered lumpy background (CLB), allowing for various sub-classes of lesions including full solid, semi-solid, and ground-glass lesions. To insert a lesion into a full body human model (e.g., XCAT phantom), the edges of the lesion are blended into the surrounding background using a parameterizable Gaussian blurring technique. The developed lesion tool allows users to define lesion sizes either manually or automatically following population distribution of lesion sizes. Similarly, the tool allows users to insert lesions either manually or automatically while avoiding intersections with pulmonary structures. The utility of the developed lesion tool was demonstrated by modeling both homogeneous and heterogeneous lung lesions and inserting them into 5 human models (XCAT). The human models were imaged using a validated CT simulator (DukeSim). Images of heterogeneous lesions were visually comparable to clinical images. The first order and texture radiomics features (58 features) were extracted from all image series and compared using the Pearson correlation. The two lesion generation techniques for full solid lesions (homogeneous vs. heterogeneous) were observed to have a weak correlation (r<0.4) for 35 of 58 features using a soft kernel, and for 43 of 58 features using a sharp kernel—capturing the structural differences between the two models. The lesion tool proved capable of forming different lung lesion sub-classes (full-solid, semi-solid, and ground-glass) through its input parameters to emulate the lesion characteristics of interest for a virtual lesion study.

Deep learning CT reconstruction (DLR) has become increasingly popular as a method for improving image quality and reducing radiation exposure. Due to their nonlinear nature, these algorithms result in resolution and noise performance which are object-dependent. Therefore, traditional CT phantoms, which lack realistic tissue morphology, have become inadequate for assessing clinical imaging performance. We propose to utilize 3D-printed PixelPrint phantoms, which exhibit lifelike attenuation profiles, textures, and structures, as a better tool for evaluating DLR performance. In this study, we evaluate a DLR algorithm (Precise Image (PI), Philips Healthcare) using a custom PixelPrint lung phantom and perform head-to-head comparisons between DLR, iterative reconstruction, and filtered back projection (FBP) with scans acquired at a broad range of radiation exposures (CTDI_vol: 0.5, 1, 2, 4, 6, 9, 12, 15, 19, and 20mGy). We compared the performance of each resultant image using noise, peak signal to noise ratio (PSNR), structural similarity index (SSIM), feature-based similarity index (FSIM), information theoretic-based statistic similarity measure (ISSM) and universal image quality index (UIQ). Iterative reconstruction at 9mGy matches the image quality of FBP at 12mGy (diagnostic reference level) for all metrics, demonstrating a dose reduction capability of 25%. Meanwhile, DLR matches the image quality of diagnostic reference level FBP images at doses between 4 to 9mGy, demonstrating dose reduction capabilities between 25% and 67%. This study shows that DLR allows for reduced radiation dose compared to both FBP and iterative reconstruction without compromising image quality. Furthermore, PixelPrint phantoms offer more realistic testing conditions compared to traditional phantoms in the evaluation of novel CT technologies. This, in turn, promotes the translation of new technologies, such as DLR, into clinical practice.

Segmentation of diagnostic radiography images using deep learning is progressively expanding, which sets demands on the accessibility, availability, and accuracy of the software tools used. This study aimed at evaluating the performance of a segmentation model for digital breast tomosynthesis (DBT), with the use of computer-simulated breast anatomy. We have simulated breast anatomy and soft tissue breast lesions, by utilizing a model approach based on the Perlin noise algorithm. The obtained breast phantoms were projected and reconstructed into DBT slices using a publicly available open-source reconstruction method. Each lesion was then segmented using two approaches: 1. the Segment Anything Model (SAM), a publicly available AI-based method for image segmentation and 2. manually by three human observers. The lesion area in each slice was compared to the ground truth area, derived from the binary mask of the lesion model. We found similar performance between SAM and manual segmentation. Both SAM and the observers performed comparably in the central slice (mean absolute relative error compared to the ground truth and standard deviation SAM: 4±3%, observers: 3±3%). Similarly, both SAM and the observers overestimated the lesion area in the peripheral reconstructed slices (mean absolute relative error and standard deviation SAM: 277±190%, observers: 295±182%). We showed that 3D voxel phantoms can be used for evaluating different segmentation methods. In preliminary comparison, tumor segmentation in simulated DBT images using SAM open-source method showed a similar performance as manual tumor segmentation.

We develop the XCAT series of phantoms for medical imaging research. The phantoms model different individuals over various ages, heights, and weights, but a current drawback is they do not include small intestine variability. Each phantom has a small intestine derived from a common anatomical template due to the difficulty of fitting a regular tubular model to patient segmentations. Building upon previous work, we develop a software pipeline to add realistic variability in the phantoms by generating tubular small intestine surface models with random length and diameter fit within the constraints of a given intestine segmentation. The pipeline first alpha wraps a given segmentation into a 3D mesh. Using a random walk algorithm, a path is then constructed within the constraints of the surface mesh while avoiding both path and surface collisions given a set diameter, start point, and end point that bounds the random walk of the passage line. After generating the passage line, the program smooths the path by fitting a cubic spline curve to it. A cubic NURBS cylinder is lofted along the path to create an initial target model. The cylinder is then grown radially, avoiding self-intersection and bounded by the mesh surface, to achieve a user-defined volume. The pipeline was tested on 45 sets of patient CT data. From the results, we find we can generate variable mean passage lengths and mean diameters within realistic ranges found for the general population.

In this work, we aimed to investigate the sensitivity of spectral CT to changes in temperatures. Moreover, we assessed the accuracy and precision of spectral CT to provide temperature mapping within the treatment volume in CT-guided hypo- and hyper-thermal tumor ablations. Leveraging optical temperature sensors and a 3D printed phantom holder designed to precisely position metallic thermal ablation applicators and reduce the associated image artifacts, we collected spectral CT data spanning a wide range of temperatures (-100 to +90°C) and used a robust theoretical model to fit these distributions for both conventional and spectral CT data. From these expressions, look-up tables that map temperature to CT were generated within the temperature range of interest for cryoablation and hyperthermal ablations, respectively. Consistent with expectations from theoretical models for a water-based gel material, we demonstrated similar sensitivity for conventional and spectral results, with a 3-fold increase in sensitivity in the heating temperature range compared to frozen material. Next, we used repeated scans of a multi-energy phantom to determine the imaging protocol that yield the highest precision (standard deviation of average signal through repeated measurements) in CT signal at the radiation dose levels used in our practice for treatment monitoring during thermal ablation. Our data revealed the complex interplay among radiation dose, image noise, and material decomposition accuracy for spectral data. With a 5mGy CTDIvol limit, we obtained a standard deviation of the average value within a 3³ mm³ VOI value of 0.85HU-equivalent for electron density images, corresponding to a precision of 7°C and 2°C for frozen and heated gel, respectively. Conventional image data, including with the application of an in-house CNN denoising algorithm, was unable to reduce the standard deviation of ROI measurements below 1HU. This information was used to develop a clinical-ready CT thermometry protocol that was subsequently independently tested on hyper-thermal measurements, demonstrating an accuracy (root mean square error) in temperature estimates of <8°C for heated gel.

Continuous-sweep limited angle (CLA) fluoroscopy is a proposed technique for real-time 3D device guidance during catheter-based procedures, where the x-ray C-arm continuously rotates back-and-forth within a limited angle during fluoroscopic image acquisition. The 3D device reconstruction relies on accurate and robust tracking of the device centerline in the 2D projection images. The purpose of this study was to investigate the influence of radiation dose per frame on deep learning-based device segmentation and tracking accuracy. A convolutional neural network was trained based on simulated and manually annotated images from retrospectively analyzed animal studies. After binarization, images were postprocessed using topology preserving thinning and a graph-based path search was used to extract the device centerline. To evaluate the accuracy of the approach, a porcine study was conducted, where a microcatheter and guidewire were navigated through the hepatic arteries. Image sequences with three different detector dose levels were acquired (80nGy/frame, 170nGy/frame, and 380nGy/frame at the detector). The deep learning-based device tracking results were compared to manually annotated reference device centerlines. The average root mean squared distance (RMSD) between reference and automatic segmentation was 0.45±0.07mm and 0.50±0.07mm for the two higher dose levels and 2.75±7.71mm for the lowest level. A significant difference was only detected for the lowest dose level (p<0.00001). However, when only the first 1.5cm from the tip of the device was evaluated, no significant differences were found for all dose levels with average RMSDs between 0.56 and 0.62mm. This suggests that accurate device-tip tracking is possible even at the lowest dose level.

Histotripsy is an emerging focal tumor therapy that utilizes focused ultrasound (US) to mechanically destroy tissue. Cone beam CT (CBCT) guidance has been developed to overcome limitations of diagnostic US for visualizing and targeting tumors. Existing workflow for CBCT-guided histotripsy utilizes targeting based on CBCT images acquired with the patient in position for treatment; therefore, treatment planning is currently performed intraprocedurally right before delivering therapy. To provide a framework for planning liver tumor treatments in advance, a biomechanical, nonrigid model is proposed to predict volumetric liver deformations between a pre-procedural diagnostic scan and the intraprocedural CBCT. In the proposed registration approach, the liver, gallbladder (GB), and hepatic blood vessels were segmented from CBCT images acquired at two different points of motion. Segmented structural surfaces were registered using a rigid iterative closest point algorithm followed by deformable registration (demons algorithm). These internal and external hepatic structural surface registrations were used as boundary conditions for a finite element model (FEM) to determine internal liver deformations. Four FEM models were constructed: the liver alone (L-FEM), liver and GB (LG-FEM), liver and vessels (LV-FEM), and liver, GB, and vessels (LGV-FEM). Registration accuracy was measured as the Euclidean distance between manually annotated vessel bifurcations. Bifurcation error was 3.5±3.5mm for LGV-FEM, a 52% improvement from L-FEM (7.4±6.1mm). Including vessels and GB in the model both individually reduced bifurcation error compared to corresponding models excluding those structures. FEM-based liver registration guided by internal and external hepatic structures is a feasible method to facilitate CBCT-guided histotripsy treatment planning.

Deep learning-based metal artifact reduction methods often struggle to apply the trained network to real data due to the difference between simulated training and real application datasets. To solve this problem, we propose a patient-specific data augmentation method that can effectively train the convolutional neural network of metal artifact reduction with a single patient dental CT volume. The idea of the proposed method is to generate the dataset using metal-unaffected slices from the patient data that require metal artifact reduction. The dataset generated by leveraging the adjacency of both metal-affected and metal-unaffected slices closely resembles real metal artifact images in terms of teeth shape and CT system geometry. To overcome the problem of small data size due to using only a single patient's data, we augment the data by generating various patient-specific metal masks. We segment the bone and label each tooth to get the size and position of the tooth. We determine the size and shape of the metal objects in the metal mask based on the information of the labeled teeth. Metal artifacts are simulated from metal masks and patient data using a polychromatic sinogram simulation method and iterative estimation of metal attenuation coefficients. For experiments, we train the same U-net structure network with different train datasets and tested with real metal artifact. The results show that the patientspecific dataset of the proposed method is more suitable for reducing the real metal artifact than the dataset generated using large amounts of data.

Deep learning based denoising techniques have effectively reduced noise in low-dose computed tomography (LDCT) images. However, the need for pairs of LDCT and normal-dose CT (NDCT) training images is the main reason that makes the use of deep learning networks difficult. Therefore, we propose a method of generating synthesized LDCT images by adding synthesized quantum noise to the NDCT images. For this, we estimate the quantum noise power spectrum (NPS) of patient CT images by modeling anatomical NPS and quantum NPS, respectively. We use the quantum NPS as a noise filter and generate synthesized quantum noise by filtering random Gaussian noise. Using the proposed method, LDCT denoiser can be trained without using pairs of LDCT and NDCT images, and the trained denoiser effectively reduces noise in LDCT images. The proposed method facilitates data acquisition for training deep learning networks and reduces the risk of radiation exposure to patients.

Existing micro-CT systems can visualize the trabecular bone microstructure; however, its application is limited because of the small scan field of view and the long scan time for clinical use. We developed a high-resolution CT for human extremities (HRCTe) with clinical feasibility, which can achieve a spatial resolution of ~100μm and a short acquisition time of 6.5 seconds. A complementary metal-oxide-semiconductor x-ray detector with a 99μm pixel pitch and a compact x-ray tube with a focal spot size of 0.8mm were installed in the HRCTe. In order to maximize the performance of the x-ray tube and the detector, we introduced a low-magnification geometry. This resulted in an effective focal spot size of 0.14mm, ensuring the resolution performance of the pixel pitch at the iso-center of 81μm. The scan z-directional coverage was 50mm. For the technical assessment, modulation transfer function (MTF) and noise power spectrum (NPS) were measured using a wire phantom and a water phantom, respectively. A squared system performance function (SPF²) was calculated as MTF²/NPS. These results were compared with those measured using a high-resolution mode of a whole-body CT with a 0.25mm detector (WCT) and a dental CT (DCT). The 5%MTF of the HRCT, WCT, and DCT were approximately 3.8, 1.9, and 1.5mm^-1. The SPF² at 2.0mm^-1 of the developed CT was 260% higher than that of WCT. The trabecular bones in an anthropomorphic foot phantom were visualized much clearer with HRCTe than with both WCT and DCT.

Dynamic stress myocardial CT perfusion (CTP), a functional imaging modality that is proven highly valuable for evaluating myocardial ischemia, is one of the most challenging CT examinations being carried out in practice. The implementation varies among current CT systems on the image acquisition mode, the dose management, as well as the strategy to combine myocardial CTP with coronary CT angiography (CCTA), which may result in a large difference in the efficiency of the clinical workflow. Following the so-called “one-stop-shop” concept, a new genuine protocol has been implemented by using a 320-row CT imaging system, with rapid volumetric image acquisition in axial mode, adaptive temporal sampling rate, CCTA extracted from the boosted arterial phases within the CTP, and the deep learning algorithms ensuring the reconstructed image quality. This paper describes the imaging protocol and the associated technical parameters in detail and compares this protocol to those reported previously in the literature.

To develop a Machine Learning (ML) model that accurately estimates the patient fat-free mass (FFM) in order to personalize intravenous (i.v.) contrast volume injection and ensure reproducible target liver enhancement. A dataset of previously collected abdominal CT data from 689 adult patients referred for liver lesion characterization or cancer follow-up was utilized. The data was obtained from eleven different radiology centers, utilizing various CT vendors, spanning the period from 2018 to 2022. The dataset encompasses diverse patient characteristics and measurements, including age, gender, weight, height, Body Mass Index (BMI), Size Specific Dose Estimate (SSDE), and FFM measured with a Bioelectrical Impedance meter. A multivariate linear regression model was developed to estimate FFM. The relationships between the investigated variables and the measured FFM were examined, and the most highly correlated variables were selected for inclusion in the final model. The dataset was divided into training and test sets according to the 80/20 rule and validated using the K-fold technique. We built several multivariate linear regression models and evaluated the performance of the trained model using the testing dataset against metrics such as the GT using Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and R-squared. The cross-validation results for the best model reveal the model's robustness in typical patient profiles within our clinical settings. Specifically, the model exhibits low MAPE (0.033+/-0.003), high R2 (0.91+/-0.02), and relatively low standard deviation of residuals RMSE (2.13+/-0.23). External validation is being performed and preliminary results confirm the validity of the implementation of the theoretical FFM estimation in our personalized algorithm for contrast injection. Our model provides reliable and efficient estimation of patient FFM, facilitating the personalization of i.v. contrast volume in adult abdominal CT examinations while eliminating the need for expensive equipment.

In this study, we proposed a wavelet domain-based deep residual learning strategy for reducing metal artifacts in computed tomography (CT) images. A fully-connected neural network (FCN) was constructed for learning the end-to-end non-linear mapping between the images including metal artifacts and the residual images. Training CT images were transformed into subband images using the 2D wavelet transformation for providing the high-frequency features during network training. The residual learning was implemented by using the subband images. The performance of the proposed technique was compared to that of the O-MAR algorithm. The results showed that metal artifacts were sufficiently suppressed by the proposed technique, and the proposed technique reduced the NRMSE by 12.34% and improved the SSIM by 0.84% compared to the O-MAR algorithm. In conclusion, the proposed model is able to efficiently reduce metal artifacts in CT images and has the superior performance compared with the commercial algorithm.

The static CT by Nanovision, as a new CT scanning formula, assembles a multi-source array and a ring detector array on two parallel planes with a fixed offset. The advantage of this configuration is that each source only needs to be rotated over a smaller angle range to complete a full scan than with conventional CT systems. However, the large cone angle from the source to the detector and the distribution of multiple sources lead to severe incomplete projections during the scanning process. To address this issue, this paper proposes a deep iterative network based on directional TV regularization. The network employs a tensorization module suitable for the static CT geometry in the forward and back-projection steps, and the regularization term adopts a directional TV deep learning model, which enables end-to-end reconstruction of incomplete data in the static CT. Experimental results demonstrate that the proposed method can effectively eliminate sparse artifacts, uneven artifacts and noise, and can obtain high quality images.

Computer tomography (CT) imaging is an essential diagnostic tool in clinical practice. Because of its radioactive nature. It is required to minimize the dose delivered to the patient. Currently, in clinical settings, the adjustment of radiation dose for CT imaging primarily relies on parameters such as tube voltage and tube current, which are often adjusted based on experience, leading to potential unreliability and instability. In this work, we propose a reinforcement learning (RL) based approach for tuning the tube current and voltage according to the principle of As Low As Reasonably Achievable (ALARA). Our method involves the development of an automatic parameter adjustment network (APAN) to determine the optimal policy for parameter adjustment. In this primary study, APAN is trained in a simulation environment and images are reconstructed by the Feldkamp-Davis-Kress (FDK) method, the experiments demonstrate its ability to optimize the parameters to obtain a better dose distribution than a uniform or energy absorption based distribution.

Cardiac computed tomography (CT) imaging faces challenges from cardiac and respiratory motion, which can result in motion artifacts. We have previously developed an innovative Triple-Source CT (TSCT) architecture which enables parallel acquisition of three projections to improve temporal resolution. However, forward and cross-scattering induced by multi-source exposures can severely degrade image quality. In this work, we evaluate various scatter correction approaches including a data-driven deep learning approach to mitigate scatter in a physical TSCT system. Phantom studies were performed under various configurations to investigate scatter effects and evaluate image quality pre- and post-correction. Evaluation metrics including HU profiles, HU uniformity and contrast-to-noise ratio (CNR) were analyzed. Among all evaluated scatter mitigation methods, the collimator-based hardware method achieved the best performance. Among all evaluated software-based methods, our deep-learning method performed slightly better than other deep learning methods.

Metal artifacts in CT images severely degrade image quality and decrease the diagnostic value of CT examinations. In this study, a dual-domain deep learning neural network framework that fuses information from both the sinogram and image domains is proposed. Our approach leverages the powers from the imaging-physics-driven model and the data-driven approach. It involves a sinogram improvement module, an image enhancement module, and an image fusion module, and promotes mutual learning from training data in both the sinogram domain and image domain. The experimental results show that our method can successfully reduce metal artifacts while preserving the tissue structures in the regions near the metal implants.

Robotic CT is a novel imaging platform built on two manipulators with great flexibility and convenience. But it suffers from limited mechanical motion accuracy, which brings artifacts into images. Acquiring true geometry parameters is critical for accurate image reconstruction. While it’s impractical to monitor all geometry positions in practice. Down-sampling the projection number and using sparsity reconstruction offers a feasible way of solving this problem. Score-based generative model (SGM) is a powerful generative model able to produce directional samples guided by prior information. Through combining prior data and generated data, images quality can be significantly improved. In this work, we trained a score-based generative model using images from two real CT scan datasets. In sampling of score-based net, prior sparsity projection was added through cone-beam projection and image reconstruction. Full projection under non-standard geometry was simulated by adding deviation into standard circular geometry. We compared performance of several algorithms on sparsity and full data, under true and ideal geometry. Image was evaluated by typical indexes and visible details. Images of SART under ideal geometry showed severe artifacts, with strip artifacts in soft tissue. Compared with images under ideal geometry, SGM-based sparsity reconstruction showed visual fuzzier image but with higher index, which improved by 59.0% and 41.1% for PSNR and SSIM. Compared with sparsity reconstruction under true geometry using SART, SGM-based method showed clearer image and higher indexes, with 11.4% and 24.7% improvement of PSNR and SSIM. SGM-based sparsity reconstruction showed great potential in sparsity reconstruction under non-standard geometry.

The increasing application of deep learning in medical imaging has prompted the need for developing analysis tools to understand, evaluate, and enhance neural networks and their outcomes. However, image quality assessment of learning-based methods remains challenging due to their nonlinear and data-dependent nature. In this study, we introduce an analysis method for learning-based CT denoising algorithms based on the Jacobian matrix, which characterizes the response of the network to perturbations around its input. We trained a classic denoising network architecture using the normal dose images as label but using progressively lower dose images as input. For each network, the Jacobian matrices for an example input image were evaluated over 50 noise realizations in terms of their width and anisotropy. We further performed null space analysis of the Jacobians to investigate the preserved and null components of input perturbations. The variability in the network outputs and Jacobian matrices was found to increase with a reduction in dose levels. Such behavior, however, is only observed around edges in the anatomy. The width of the Jacobian matrix also increased as the dose decreased, suggesting an increase in the strength of the smoothing effect. The anisotropy of the Jacobian was more pronounced about the edges, indicating edge enhancement in the denoising process. The null space analysis showed a reduction in the preserved features of an input perturbation as the dose decreased, indicating greater misrepresentation of anatomical structures. The analysis proposed in this work aids in understanding and interpreting the denoising behavior of neural networks and can potentially serve as a means to regulate and optimize network performance.

Our study investigated the impact of irregular respiration on correlation between surrogate respiratory signal and internal organ motion in 4DCT. Based on the proposed quantitative index of respiratory irregularity, the study cases were divided into two: the case with regular breathing and the case with irregular breathing. By groups, internal organ movements were measured from 4CT images, and correlation between external surrogate signal and internal organ motion was analyzed. The results showed that the correlation with internal organ motion was obviously deteriorated in the case with irregular breathing, rather than the case with regular breathing. As observed that the irregular respiration may lead to such errors on gating interval up to 20% either phase or amplitude level, it would be necessary to employ the respiratory guidance and feedback system that can minimize the respiratory irregularity.

In recent years, the importance of spectral CT scanners in clinical settings has significantly increased, necessitating the development of phantoms with spectral capabilities. This study introduces a dual-filament 3D printing technique for the fabrication of multi-material phantoms suitable for spectral CT, focusing particularly on creating realistic phantoms with orthopedic implants to mimic metal artifacts. Previously, we developed PixelPrint for creating patient-specific lung phantoms that accurately replicate lung properties through precise attenuation profiles and textures. This research extends PixelPrint's utility by incorporating a dual-filament printing approach, which merges materials such as calcium-doped Polylactic Acid (PLA) and metal-doped PLA, to emulate both soft tissue and bone, as well as orthopedic implants. The PixelPrint dual-filament technique utilizes an interleaved approach for material usage, whereby alternating lines of calcium-doped and metal-doped PLA are laid down. The development of specialized filament extruders and deposition mechanisms in this study allows for controlled layering of materials. The effectiveness of this technique was evaluated using various phantom types, including one with a dual filament orthopedic implant and another based on a human knee CT scan with a medical implant. Spectral CT scanner results demonstrated a high degree of similarity between the phantoms and the original patient scans in terms of texture, density, and the creation of realistic metal artifacts. The PixelPrint technology's ability to produce multi-material, lifelike phantoms present new opportunities for evaluating and developing metal artifact reduction (MAR) algorithms and strategies.

As medical technologies advance with increasing speed, virtual imaging trials (VITs) are emerging as a crucial tool in the evaluation and optimization of new imaging techniques. Widely used in many VITs is the four-dimensional extended cardiac-torso (XCAT) phantom, a comprehensive computational model that accurately represents human anatomy and physiology. While the XCAT phantom offers a powerful tool for imaging research, it offers only a limited model of blood flow to compartmentalized organs, potentially limiting the realism and clinical applicability of contrast-enhanced scan simulations. This study bridges that gap by combining realistic CT simulation with an accurate model of blood flow dynamics to enable more realistic simulations of contrast-enhanced imaging. To achieve this, a validated one-dimensional blood flow simulator, HARVEY1D, was used to model flow throughout the vessels of the XCAT phantom. DukeSim, a validated CT simulation platform, was then modified to incorporate the resulting flow into its simulations, thus enabling the genera􀆟on of simulated CT scans reflective of real-world blood-based contrast-enhanced imaging scenarios. To demonstrate the utility of this pipeline in an initial application to cardiac imaging, three heart models were studied: a non-diseased model, a 50% stenosis model, and an 80% stenosis model. Three seconds of contrast propagation were tracked in each heart model, and CT scans corresponding to two timepoints were simulated. Results demonstrated that the presence of stenosis significantly impacted blood flow, with greater resistance to blood flow leading to altered flow patterns visible in the simulated CT images. This work showcases a pipeline that leverages both computational fluid dynamics and medical imaging simulations to enhance the realism of virtual imaging trials and facilitate the evaluation, optimization, and development of diagnostic tools for contrast-enhanced imaging.

Dual-energy computed tomography (CT) can improve image quality relative to single-energy CT through decomposition into two or three basis materials and synthesis into virtual monoenergetic images (VMIs). Decomposition into even more materials is possible using a multi-triplet material decomposition (MMD) algorithm, in which a different set of materials (i.e. “material triplet”) is chosen for each voxel. MMD could be particularly useful for certain tasks, such as atherosclerotic plaque risk assessment. However, in its current form, MMD requires manual tuning to optimize its performance in different imaging scenarios. This work aimed to quantitatively explore the sensitivity of the MMD algorithm to CT image noise and initial VMI basis materials. We simulated 80-kVp and 140-kVp CT images of a water cylinder with four inserts (soft tissue, fat, calcium, and iodine) at twenty dose levels (0.2 to 4.0mGy). The needed input VMIs were generated using initial two-material decomposition into either soft tissue/bone or water/aluminum material pairs. VMIs were then used for MMD into basis images of the four insert materials. We found that the choice of VMI basis materials affects MMD image quality at lower doses; tissue/bone was best for all materials except fat using a dose below 1mGy. Additionally, the image quality benefit of increasing dose plateaus at a certain point, possibly due to the reduction in voxels “jumping” to different material triplets. This sensitivity analysis offers insight into the nuanced effects of MMD input variations and may be useful for jointly optimizing the quality of many basis material images with minimal patient dose.

We investigated the impact of a CNN-based deep-learning (DL) image de-blurring algorithm on coronary artery calcium (CAC) detection performance in conventional CT imaging. Our approach comprises first de-noising the image with a state-of-the-art CNN-based image de-noising algorithm. With improved SNR, it is then possible to sharpen the image with a CNN-based image de-blurring algorithm. We train such networks using natural images, i.e., a large set of diverse photographs. The de-noising strength in the final image can be adjusted by blending back the estimated noise from the first step to the desired degree. To assess the impact of the de-blurring algorithm, we scanned an anthropomorphic phantom containing 100 small calcifications on a CT system using a CAC scoring protocol. Data were acquired at clinical and high dose, and subsequently reconstructed with and without the DL de-blurring algorithm, using 25% of the maximum de-noising strength. For each small CAC, detectability was defined as the ability to calculate an Agatston score (at least 3 adjacent voxels exceeding 130 HU). For the high dose scans, CAC detectability increased from 39% for the standard reconstruction to 49% with de-blurring. The same 39% CAC detectability at high dose without de-blurring was obtained with routine dose with de-blurring. In this work, we also show some visual impressions of applying our DL de-blurring method to clinical cardiac data.

In CT radiomics, numerical parameters extracted from CT images are analyzed to find biomarkers. Since these numerical parameters can vary with imaging parameters, there is a need to optimize acquisition protocols for radiomics. In this work, we investigate the effect of deep-learning-based image reconstruction on the accuracy of radiomic parameters of tumors. We image a 3D printed lung phantom containing four tumors (ellipsoidal, lobulated, spherical, and spiculated), using the CAD model as ground truth. The phantom was 3D printed using fused deposition modeling with a PLA filament and 80% fill rate with a gyroidal pattern to mimic soft tissue. CT images of the 3D printed phantom and tumors were acquired with a GE revolution scanner with 120kVp and 213mAs. We reconstructed images using FBP and a vendor-supplied deep learning image reconstruction (DLIR) method (TrueFidelity, GE HealthCare). We also applied 24 custom convolutional neural network denoisers with a U-net architecture, trained on the AAPM-Mayo Clinic Low Dose CT dataset. After segmentation, 14 radiomic features were extracted using SlicerRadiomics. Results show that the vendor-supplied DLIR gave a smaller relative error than FBP for 87% of radiomic features. 8 out of 24 custom denoisers yielded a smaller error than FBP in 50% or more of the radiomic measurements. One denoiser, (VGG16+L1 loss, 32 features, batch size 16), outperformed FBP in 84% of measurements and outperformed the vendor-supplied DLIR in 63% of the measurements. In conclusion, our results demonstrate that deep-learning-based denoising has the potential to improve the accuracy of CT radiomics.

As Deep Learning Reconstruction (DLR) begins to dominate computed tomography (CT) reconstruction, performance evaluation via conventional phantoms with uniform backgrounds and specific sizes may benefit from augmentation with simulated controlled test objects inserted into anatomical backgrounds. The purpose of this study is to validate a simulation tool with physics-based image quality metrics both in phantom and in patient data. An analytic forward projection tool, based on detector and source geometry with beam spectra was designed to match the specifications of Canon Medical’s Aquilion ONE Prism. The Catphan^TM 500 and two water phantoms, 24cm and 32cm in diameter, were scanned with Aquilion ONE Prism at various mA levels and reconstructed with FBP. Corresponding simulated images were generated. The CT number, noise power spectrum (NPS) and modulation transfer function (MTF) were evaluated and compared between the simulated images and actual images. Simulated projection data of a Catphan^TM sensitometry cylinder was also combined with a patient sinogram and reconstructed with a variety of kernels. The MTF of three different contrast rods were measured and compared with the MTF measured from Catphan^TM. The CT numbers were equivalent between the simulated data and the image acquired from the actual CT system. The MTF measured from the simulated data of both phantom and patient image matched with the MTF from Catphan^TM. The noise properties of the simulated data also aligned with the NPS of the 24cm and 32cm water phantom image. The simulation tool was able to generate images with image quality equivalent to the images scanned and reconstructed from the actual CT system. With this validation study, the simulation will be utilized to further evaluate the performance of deep learning reconstructions (DLRs).

The technique of Localizer Radiography (LR) can realize patient oriented automatic exposure control based on attenuation information. Rotating Projection based Localizer Radiography (RPLR), as a dynamic tube positioned scanning, aims to improve the whole clinical workflow. However, topogram (topo) reconstruction in RPLR is affected by sparse sampling. This paper proposed a deep learning model which contains transformers (power in modeling long-term relationship) and CNNs (high texture modeling capacity) to implement projection context restoration for topo reconstruction. With a coarse topo prior generated by the transformers based on sparse sampling data, high-fidelity topo texture can be rendered with CNNs, which reveals great potential for topo reconstruction in RPLR.

Virtual Imaging Trials, known as VITs, provide a computational substitute for clinical trials. These traditional trials tend to be sluggish, costly, and frequently deficient in definitive evidence, all the while subjecting participants to ionizing radiation. Our VIT platform meticulously mimics essential components of the imaging process, encompassing everything from virtual patients and scanners to simulated readers. Within the scope of this intended research, we aim to authenticate our virtual imaging trial platform by duplicating the results of the National Lung Screening Trial (NLST) for lung cancer screening through the emulation of low-dose computed tomography (CT) and chest radiography (CXR) procedures. The methodology involves creating 66 unique computational phantoms, each with inserted simulated lung nodules. Replicating NLST CT imaging via Duke Legacy W20 scanner matched essential properties. Virtual imaging was done through DukeSim. A LUNA16-trained virtual reader, combining a 3D RetinaNet model (front-end) with a ResNet-10 false positive reduction model (back-end), evaluated the virtually imaged data, ensuring rigorous assessment. The back-end model achieved a sensitivity of over 95% at fewer than 3 false positives per scan for both the clinical and virtual imaged CTs. Notably, nodule diameter-based analysis showcases even higher sensitivity for nodules measuring 10 mm or more. In conclusion, the integration of diverse computational and imaging techniques, culminating in a virtual reader, demonstrates promising sensitivity. To capture both arms of the trial, future research will compare virtual reader performance on CT with CXR. This affirms the transformative potential of virtual imaging trials in advancing evidence-based medicine, offering an efficient and ethically conscious approach to medical research and development.

Computed tomography (CT) imaging is widely used for medical diagnosis and image guidance for treatment. Metal artifacts are observed on the reconstructed CT images if metal implants are carried by patients due to the beam hardening effects. In this condition, the acquired projection data cannot be used for analytical reconstruction as they do not meet Tuy's data sufficiency condition. Numerous deep learning-based methods have been developed for metal artifact reduction (MAR), providing superior performance. Nevertheless, all the reported models are data-driven and require large-size referenced images for the manifold approximation. In this work, we propose a physics-driven sinogram manifold learning method, which fully exploits the projection data correlation in CT scanning for MAR, and the proposed method is ready to be extended to other data-incomplete CT reconstruction problems.

Recently, we have released the first open-source version of our Multi-Channel CT Reconstruction (MCR) Toolkit (https://gitlab.oit.duke.edu/dpc18/mcr-toolkit-public). The initial release of the Toolkit represents 10 years of development and provides a complete set of GPU-accelerated tools for solving multi-channel (multi-energy, time-resolved) X-ray CT reconstruction problems with support for both analytical and iterative reconstruction in common preclinical and clinical geometries. This initial version of the Toolkit (v1.0) relies on MATLAB and its MEX interface for orchestrating CT reconstruction pipelines; however, heavy reliance on MATLAB comes with licensing restrictions and limited support for deep learning augmentation of reconstruction pipelines. In this work, we detail the features of v2.0 of the MCR Toolkit which ports all the Toolkit’s v1.0 features from MATLAB to the Python programming language, including the ability to perform regularized, iterative reconstruction of multi-energy photon-counting cardiac CT data. We demonstrate these new features through benchmarks which show comparable performance between our MATLAB (v1.0) and Python (v2.0) implementations of the BiCGSTAB(l) solver, following improved memory management in our Python implementation. We also demonstrate a high-level interface between v2.0 of the Toolkit and PyTorch, allowing the incorporation of a previously trained multi-energy CT denoising model, known as UnetU, directly in our multi-channel reconstruction framework. These preliminary reconstruction results show a reduction in intensity bias from 13HU, after a single pass of the UnetU denoising model, to 7HU after the same model is incorporated into our iterative reconstruction framework; however, some high-contrast edge features are exaggerated in the UnetU reconstruction, and the noise standard deviation increases from 21HU to 34HU.

Motion detection and correction are key to good image quality in x-ray CT. Various approaches have been explored in the past. Methods to combat patient motion include the use of gating devices to optimize the data acquisition, reducing the patient scan time via faster gantry rotation and large detector coverage, and the development of advanced reconstruction and post-processing algorithms to minimize motion artifacts. In clinical practice, these approaches often rely on CT operators to detect local motion artifacts and initiate appropriate corrective actions. Because physiologically induced motion is complex and often varies spatially and temporally, the effectiveness of the existing approaches is often compromised. In this paper, we propose a data consistency-metric to automatically detect and characterize motion. The metric takes advantage of the fact that the size and shape of small rigid objects, such as blood vessels or lung nodules, do not change significantly over the data acquisition window. The characteristics of the tomographic reconstruction ensures that the magnitude of the circular integral surrounding these objects is small. Therefore, the differential signal of the back-projected intensity between a stationary object and its background integral is linear with respect to the projection view. The “goodness of fit” of the differential signal, therefore, represents the level of motion present at each local region and provides a four-dimensional (spatial plus temporal) motion map of the reconstructed volume. We first present a mathematical model to establish a theoretical foundation. Extensive computer simulations are then utilized to confirm the validity of our theory. Our theory is further tested using a swine cardiac scan to demonstrate its efficacy and robustness. Future expansion and potential clinical benefits of the proposed approach are also discussed.

The Channelized Hotelling observer (CHO) is well correlated with human observer performance in many CT detection/classification tasks but has not been widely adopted in routine CT quality control and performance evaluation, mainly because of the lack of an easily available, efficient, and validated software tool. We developed a highly automated solution–CT image quality evaluation and Protocol Optimization (CTPro), a web-based software platform that includes CHO and other traditional image quality assessment tools such as modulation transfer function and noise power spectrum. This tool can allow easy access to the CHO for both the research and clinical community and enable efficient, accurate image quality evaluation. An example of its application to evaluating the low-contrast performance of a photon-counting-detector CT with varying scan modes, image types and reconstruction methods was demonstrated using this web platform.

Spectral capabilities of photon counting detectors (PCDs) can allow material decomposition. We recently showed multi-material decomposition using high resolution photon counting detectors (Medipix3). High-resolution photon counting spectral detectors have unique advantages and challenges. Some of the challenges arise from noise properties as well as spectral distortions. We show benefits of an empirical correction method to obtain accurate attenuation values in a spectral CT even with high resolution detectors and a combination of spectral distortions. Aided with accurate spectral correction, we show that a Gaussian mixture model assisted iterative decomposition can separate multiple materials at once. Our group has been investigating the role of image texture features in signal detection performance in tomographic images. Here we will explore utilizing variations in image texture features in spectral CT material decomposition. Along with attenuation variations for each energy bin, second order statistical texture feature variations associated with spectral data will be used to reduce the number of energy bins and imaging dose to perform multi-material decomposition. With promising preliminary results, we will show a more thorough investigation of image texture variations in spectral data to assist efficient and low dose material decomposition in spectral CT.

Contrast-enhanced mammography (CEM) has shown increased sensitivity for detecting breast cancer when compared to traditional full-field digital mammography with performance comparable to MRI. While all current CEM systems use a dual-energy approach, photon-counting detectors can similarly be used by acquiring two or more energy bins to subtract anatomical noise and highlight iodine uptake. Photon-counting detectors have several advantages over dual-energy such as the simultaneous acquisition of multiple energy bins and the potential for electronic noise rejection. However, photon-counting detectors suffer from several physical phenomena such as charge sharing and k-shell fluorescence that degrade its spatial and spectral response. Solving for true counts given measured photon counts constitutes an inverse problem. While, analytically difficult to solve, machine learning techniques can be an alternative method. In this simulated study, we investigated the use a light-weight convolutional neural network to correct for spatial and spectral degradations in CEM acquisitions using a photon-counting detector.

Energy-resolving photon counting detectors (ERPCDs) used for medical X-ray diagnosis have the potential to extract quantitative information concerning various organs. Previously, we proposed an analysis procedure to determine the effective atomic number (Z_eff), but parameter optimization is required. The purpose of this study is to develop an imaging simulator and to derive optimal imaging conditions. First, we created an imaging simulator taking into consideration the following three elements: (1) X-ray attenuation in objects, (2) detector response of cadmium zinc telluride (CZT), and (3) statistical fluctuation. Second, virtual phantoms consisting of soft tissue (Z_eff=6.5) and bone (Z_eff=13) with mass thicknesses (ρt) of 0.1 to 80g/cm² were made, and the Z_eff images were generated under various imaging conditions related to tube voltages (60 to 120kV) and energy bin settings (10keV intervals). Then, the optimal imaging condition was derived from the accuracy analysis of Z_eff images. Furthermore, a demonstration of clinical quantitative images was performed by using digital phantom which was created from voxel data of the CT image of human body phantom. The optimized imaging conditions were a tube voltage of 120kV with energy bin settings of 20–30–50–120keV. The applicable thickness ranges (achieving differences of +/- 1.0 from theoretical Z_effvalues) for Z_eff=13 and 6.5 were ρt=0.8 to 10g/cm² and 2 to 20g/cm², respectively. Using the above imaging conditions, we succeeded in generating accurate quantitative images of the pelvic region.

Dual-source photon-counting CT combines the high temporal resolution and high pitch of dual-source CT with the material quantification capabilities of photon-counting CT. It, however, results in cross-scatter that increases in severity with increased patient size and collimation. This cross-scatter must be corrected to ensure the removal of scatter artifacts and improve quantitative accuracy. To evaluate residual cross-scatter of a first-generation dual-source photon-counting CT and the effect of phantom size, collimation, and radiation dose, a phantom was scanned in single- and dual-source modes with and without its extension ring at three collimations and three radiation doses. Virtual monoenergetic images (VMI) at 50 keV, VMI 150 keV, and iodine density maps were reconstructed to determine variation between acquisition parameters in single- and dual-source modes. Additionally, differences relative to single-source acquisitions and to singlesource and small collimation acquisitions were calculated to reflect residual cross-scatter with and without matched collimation. At VMI 50 keV, inserts exhibited accuracy and similar variation between single- and dual-source modes, averaging 5.4 ± 2.6 and 6.2 ± 2.5 HU, respectively, across phantom size, collimation, and radiation dose. Differences relative to single-source measured 5.1 ± 8.5 and 0.4 ± 4.2 HU while differences relative to single-source and small collimation acquisitions were 6.4 ± 10.8 HU and -0.5 ± 3.9 HU for VMI 50 and 150 keV, respectively. This minimal residual cross-scatter increases confidence in the quantitative accuracy of spectral results necessary for clinical applications of dual-source photon-counting CT with motion, such as cardiac imaging.

Spectral Photon-Counting CT (SPCCT) has become the focus of researchers, because of its energy resolution ability. SPCCT has broad application prospects, especially in the field of medical imaging. In medicine, iodized oil is often used as an antitumor drug, and its diffusion and retention time in tumors greatly affect the effect of tumor treatment. It's becoming an increasing concern for researchers, therefore, it is necessary to quantify the distribution and content of iodized oil in the tumor. In this study, the distribution and content of iodized oil in tumors of living mice were quantitatively analyzed with SPCCT, and an effective and convenient method based on SPCCT was established. First, using iodized oil and water as the base materials, the method of material decomposition was carried out to obtain the decomposed image of iodized oil. Then the content of iodized oil was calculated from the decomposed images, and the change of iodized oil content with time was quantitatively analyzed. The experimental results show that the quantification method of iodized oil based on SPCCT established in this study could effectively quantify the change in the distribution and content of iodized oil over a long period, indicating the potential of SPCCT in the application of drug quantitative analysis.

Proton radiation therapy has the potential of achieving precise dose delivery to the tumor while sparing non-cancerous surrounding tissue, owing to the sharp Bragg peaks of protons. Aligning the high dose region with the tumor requires accurate estimates of the proton stopping power ratio (SPR) of patient tissues, commonly derived from computed tomography (CT) image data. Photon-counting detectors within CT have demonstrated advantages over their energy-integrating counterparts, such as improved quantitative imaging, higher spatial resolution and filtering of electronic noise. In this study, the potential of photon-counting computed tomography for improving SPR estimation was assessed by training a deep neural network on a domain transform from photon-counting CT images to SPR maps. XCAT phantoms of the head were generated and used to simulate photon-counting CT images with CatSim, as well as to compute corresponding ground truth SPR maps. The CT images and SPR maps were then used as input and labels to a neural network. Prediction of SPR with the network yielded mean root mean square errors (RMSE) of 0.26-0.41 %, which is an improvement on errors reported for methods based on dual energy CT (DECT). These early results show promise for using a combination of photon-counting CT and deep learning for predicting SPR, which in extension demonstrates potential for reducing the beam range uncertainty in proton therapy.

Coronary computed tomography angiography (cCTA) is a widely used non-invasive diagnostic exam for patients with coronary artery disease (CAD). However, most clinical CT scanners are limited in spatial resolution from use of energy-integrating detectors (EIDs). Radiological evaluation of CAD is challenging, as coronary arteries are small (3-4 mm diameter) and calcifications within them are highly attenuating, leading to blooming artifacts. As such, this is a task well suited for high spatial resolution. Recently, photon-counting-detector (PCD) CT became commercially available, allowing for ultrahigh resolution (UHR) data acquisition. However, PCD-CTs are costly, restricting widespread accessibility. To address this problem, we propose a super resolution convolutional neural network (CNN): ILUMENATE (Improved LUMEN visualization through Artificial super-resoluTion imagEs), creating a high resolution (HR) image simulating UHR PCD-CT. The network was trained and validated using patches extracted from 8 patients with a modified U-Net architecture. Training input and labels consisted of UHR PCD-CT images reconstructed with a smooth kernel degrading resolution (LR input) and sharp kernel (HR label). The network learned the resolution difference and was tested on 5 unseen LR patients. We evaluated network performance quantitatively and qualitatively through visual inspection, line profiles to assess spatial resolution improvements, ROIs for CT number stability and noise assessment, structural similarity index (SSIM), and percent diameter luminal stenosis. Overall, ILUMENATE improved images quantitatively and qualitatively, creating sharper edges more closely resembling reconstructed HR reference images, maintained stable CT numbers with less than 4% difference, reduced noise by 28%, maintained structural similarity (average SSIM = 0.70), and reduced percent diameter stenosis with respect to input images. ILUMENATE demonstrates potential impact for CAD patient management, improving the quality of LR CT images bringing them closer to UHR PCD-CT images.

Coronary CT angiography (cCTA) is a fast non-invasive imaging exam for coronary artery disease (CAD) but struggles with dense calcifications and stents due to blooming artifacts, potentially causing stenosis overestimation. Virtual monoenergetic images (VMIs) at higher keV (e.g., 100 keV) from photon counting detector (PCD) CT have shown promise in reducing blooming artifacts and improving lumen visibility through its simultaneous high-resolution and multi-energy imaging capability. However, most cCTA exams are performed with single-energy CT (SECT) using conventional energy-integrating detectors (EID). Generating VMIs through EID-CT requires advanced multi-energy CT (MECT) scanners and potentially sacrifices temporal resolution. Given these limitations, MECT cCTA exams are not commonly performed on EID-CT and VMIs are not routinely generated. To tackle this, we aim to enhance the multi-energy imaging capability of EIDCT through the utilization of a convolutional neural network to LEarn MONoenergetic imAging from VMIs at Different Energies (LEMONADE). The neural network was trained using ten patient cCTA exams acquired on a clinical PCD-CT (NAEOTOM Alpha, Siemens Healthineers), with 70 keV VMIs as input (which is nominally equivalent to the SECT from EID-CT scanned at 120 kV) and 100 keV VMIs as the target. Subsequently, we evaluated the performance of EID-CT equipped with LEMONADE on both phantom and patient cases (n=10) for stenosis assessment. Results indicated that LEMONADE accurately quantified stenosis in three phantoms, aligning closely wi th ground truth and demonstrating stenosis percentage area reductions of 13%, 8%, and 9%. In patient cases, it led to a 12.9% reduction in average diameter luminal stenosis when compared to the original SECT without LEMONADE. These outcomes highlight LEMONADE's capacity to enable multi-energy CT imaging, mitigate blooming artifacts, and improve stenosis assessment for the widely available EID-CT. This has a high potential impact as most cCTA exams are performed on EID-CT.

Computed tomography systems with very small detector pixels, such as they also occur in photon-counting computed tomography (PCCT), may have dead detector rows and columns whether due to the manufacturing process, the anti-scatter grid (ASG) that blocks the primary radiation behind the ASG lamellae, or other reasons. Such situations require a sophisticated inpainting algorithm to avoid possible artifacts in reconstructed images. To meet these requirements, we developed grid inpainting with deep learning (GRIDL), a neural network-based algorithm to allow the inpainting of arbitrary gaps in column and row direction. In our experimental setup, we corrupt detector images from spiral CT scans with a regular pattern of gaps in a comparable way as the ASG would be arranged in a PCCT system and perform an inpainting using GRIDL. Our approach yields reconstructed images with image quality comparable to the gapless ground truth. In comparison to a simple inpainting method utilizing linear interpolation or a more sophisticated diffusion-based inpainting, GRIDL demonstrates a reduction in aliasing artifacts and the root mean square error (RMSE) in reconstructed images.

One advantage of photon-counting CT compared to dual-energy CT is the possibility to perform K-edge imaging, where contrast agents such as iodine can be distinguished from other substances based on spectral characteristics. However, for iodine K-edge imaging in clinical CT, the three-basis decomposition problem is ill-conditioned due to the low K-edge energy of iodine, meaning that the decomposition is highly sensitive to both noise and miscalibrations. This makes robust three-basis decomposition difficult using standard techniques. In this simulation study we evaluate a novel method of performing K-edge imaging, which circumvents the challenging three-basis decomposition step by replacing it with multiple two-basis decompositions followed by a deep convolutional neural network to generate three basis images. Based on the XCAT phantom, we generated 1224 spectral phantom image slices of the neck, with iodine-filled blood vessels and calcifications, and simulated CT imaging in CatSim with a silicon-based detector model without quantum noise, i.e. in the high-dose limit. For each simulated slice, we used maximum likelihood to perform three two-basis decompositions, into PE-PVC, PE-iodine, and PVC-iodine, yielding six basis images in total. We then trained a U-Net to map these six input images to the ground-truth basis images, PE, PVC and iodine. Our results show that the proposed method can reproduce PE, PVC and iodine basis images with high accuracy, in the high-dose limit. This suggests that the proposed three-basis decomposition method may be a feasible way of performing K-edge CT imaging with iodine, with important potential implications for imaging of the carotid arteries.

The measurements of bone macro- and microstructures provide major insight into bone health and risk fracture. The accuracy of these measurements is limited when Conventional CT is used. However, CT technology has advanced with the introduction of photon-counting detectors that offer significant improvements in spatial resolution. This advancement offers potential improvements in resolving trabecular microstructures, and the quantification of bone through current or emerging biomarkers. Additionally, the spectral separation available in photon-counting CT (PCCT) may further aid in quantification. The purpose of this study was to objectively investigate PCCT capabilities in accurate quantification of bone macro- and micro-structures. To do so, 5 human bone specimens were scanned using a PCCT scanner (NAEOTOM Alpha, Siemens) and an energy-integrating CT (EICT) (FORCE, Siemens). Each specimen was imaged at a CTDIvol of 4 and 8mGy, and then reconstructed with 2 matrix sizes and at least 2 kernels. For PCCT, a 70keV virtual mono-energetic image series was acquired to evaluate the potential benefits of spectral maps. The same specimens were also scanned using a high-resolution peripheral quantitative CT to provide a ground truth for the bone metrics. Each image series was analyzed in terms of bone mineral density (BMD) and trabecular bone volume to total bone volume. PCCT demonstrated major improvements (5.5% compared to 17% error for EICT) in quantifying bone microstructures (BV/TV). However, the BMD measurements remained similar across imaging conditions and scanners, and did not significantly change by the PCCT spatial resolution enhancement. For BV/TV measurements, PCCT T3D was the most accurate when the sharpest kernel available and 1024-matrix size for (error: 5.53%±4.72%) were used. Similarly, EICT images were the most accurate for BV/TV measurements (error: 16.70%±10.55%) when a medium-sharpness kernel and 1024-matrix size were used. The overall results suggest that PCCT technology can further improve trabecular bone measurements and thus enhance the clinical decision making for patients with bone disease.