In this work, we present a new concept for dynamic beam attenuation, the z-aligned sheet-based dynamic beam attenuator (z-sbDBA). Like the previously presented sbDBA, it allows dynamically adapting the x-ray intensity over the projection angle and also the fan beam angle by changing the tilt angle of the z-sbDBA. In contrast to the sbDBA, however, the absorbing sheets of a z-sbDBA are parallel to the detector rows. In addition, the height of the absorbing sheets varies over the fan angle – very low near the central beam, increasing toward larger fan beam angles. This facilitates designing the absorption profile of the z-sbDBA, which was not feasible with the sbDBA. Due to the changed orientation of the attenuation sheets, pronounced structures along the fan beam width are avoided, reducing the risk of ring artifacts in the reconstructed image.

A prototype of the z-sbDBA, mounted on a drive, has been built and investigated. Using a clinical computed tomography (CT) scanner, we experimentally demonstrate that variable and smooth intensity profiles can be realized by the controlled change of the angular position of the z-sbDBA. Reconstructed images do not reveal substantial artifacts, thus proving the necessary stability of the acquisition technique. We also show that the variance across a reconstructed image can be changed as a function of the tilt angle.

Our experimental results demonstrate that the new z-sbDBA concept maintains the main advantage of the sbDBA concept, which is the dynamic fluence field modulation (FFM) of the emitted x-ray beam. In addition, our findings show that due to the improved z-sbDBA structuring several drawbacks of the sbDBA can be overcome by a) avoiding pronounced structures along the fan beam angle, b) requiring only small tilt angles and c) allowing for a flexible design of the transmission profile propagated toward the patient.

We present further progress on the implementation of C-arm CT imaging with the extended line-ellipse-line (LEL) trajectory. This novel data acquisition geometry is designed to enhance image quality in interventional radiology. Previously, we showed that robust extended LEL data acquisition is feasible using a state-of-the-art multi-axis robotic C-arm (ARTIS pheno, Siemens Healthcare GmbH, Germany) and we also showed that accurate reconstruction from real data can be obtained using an iterative algorithm. The extensive computational effort required by such an algorithm is however not suitable for clinical translation. Reconstruction using a filtered- backprojection (FBP) formula would be practical. To use such a formula, there needs to be a technique to handle imperfections in the data acquisition geometry, which result from mechanical vibrations and gravity effects. We recently presented such a technique, but this development was only carried out for a single cycle of the LEL trajectory. In this work, we address the more challenging issue of reconstructing the volume covered by multiple cycles of the trajectory. Specifically, we propose an extension of our single cycle approach to multiple cycles. We successfully demonstrate that our procedure now allows seamless volume reconstruction from real data using a cone-beam performance phantom as well as an anthropomorphic head phantom. Our results bring the extended LEL trajectory closer to clinical deployment for improved image quality in interventional radiology. Further work will focus on increasing the number of views to avoid few view artifacts and on thoroughly demonstrating image quality benefits.

Orthopedic tomosynthesis is emerging as an attractive alternative to digital radiography (DR), with increased sensitivity for some clinical tasks, including fracture diagnosis and staging and follow-up of arthritis. Commercially available digital tomosynthesis (DTS) systems are complex, room-sized devices. A compact tomosynthesis system for extremity imaging (TomoE) was previously demonstrated using carbon nanotube (CNT) x-ray source array technology. The purpose of this study was to evaluate the prototype device in preparation for an Institutional Review Board (IRB)- approved patient imaging study and evaluate initial patient images.

A tabletop device was constructed using a short CNT x-ray source array, operated in three positions, and a flat panel digital detector. Twenty-one x-ray projection images were acquired at incident angles from -20 to +20 degrees in various clinical orientations, with entrance doses matched to commercial in-room DTS scanners. The projection images were reconstructed with an iterative reconstruction technique in 1mm slices. Cadaveric specimen and initial participant images were reviewed by radiologists for feature conspicuity and diagnostic accuracy.

TomoE image quality was found to be superior to DR, with reconstruction slices exhibiting visual conspicuity of trabecular bone, delineation of joint space, bone erosions, fractures, and clear depiction of normal anatomical features. The scan time was fifteen seconds with mechanical translation. Skin entrance dose was verified to be 0.2mGy. TomoE device image quality has been evaluated in cadaveric specimens and dose was calibrated for a patient imaging study. Initial patient images depict a high level of anatomical detail an increase in diagnostic value compared to DR.

Purpose: We investigate the feasibility of slot-scan dual-energy x-ray absorptiometry (DXA) on robotic x-ray platforms capable of synchronized source and detector translation. This novel approach will enhance the capabilities of such platforms to include quantitative assessment of bone quality using areal bone mineral density (aBMD), normally obtained only with a dedicated DXA scanner. Methods: We performed simulation studies of a robotized x-ray platform that enables fast linear translation of the x-ray source and flat-panel detector (FPD) to execute slot-scan dual-energy (DE) imaging of the entire spine. Two consecutive translations are performed to acquire the low-energy (LE, 80 kVp) and high-energy (HE, 120 kVp) data in <15 sec total time. The slot views are corrected with convolution-based scatter estimation and backprojected to yield tiled long-length LE and HE radiographs. Projection-based DE decomposition is applied to the tiled radiographs to yield (i) aBMD measurements in bone, and (ii) adipose content measurement in bone-free regions. The feasibility of achieving accurate aBMD estimates was assessed using a high-fidelity simulation framework with a digital body phantom and a realistic bone model covering a clinically relevant range of mineral densities. Experiments examined the effects of slot size (1 – 20 cm), scatter correction, and patient size/adipose content (waist circumference: 77 – 95 cm) on the accuracy and reproducibility of aBMD. Results: The proposed combination of backprojection-based tiling of the slot views and DE decomposition yielded bone density maps of the spine that were free of any apparent distortions. The x-ray scatter increased with slot width, leading to aBMD errors ranging from 0.2 g/cm² for a 5 cm slot to 0.7 g/cm² for a 20 cm slot when no scatter correction was applied. The convolution-based correction reduced the aBMD error to within 0.02 g/cm² for slot widths <10 cm. Reproducible aBMD measurements across a range of body sizes (aBMD variability <0.1 g/cm²) were achieved by applying a calibration based on DE adipose thickness estimates from peripheral body sites. Conclusion: The feasibility of accurate and reproducible aBMD measurements on an FPD-based x-ray platform was demonstrated using DE slot scan trajectories, backprojection-domain decomposition, scatter correction, and adipose precorrection.

This work aims to combine compressed sensing reconstruction with a deep learning framework to leverage their individual strengths and enable 123-view moderate-dose sparse-view reconstruction for diagnostic CT imaging systems. Specifically, linear FBP reconstruction was applied to reconstruct sparse data, followed by a trained U-Net to remove artifacts resulting from undersampling. This methodology has been exploited in many other deep learning applications for CT imaging. However, this approach may be subjected to the generalizability issue in which the reconstruction can either remove real lesions or add lesions that may not exist. This will be demonstrated in our results. In our proposed work, the output of the network is used as the prior image for a prior image constrained compressed sensing (PICCS) reconstruction. This step helps to ensure the reconstructed image is consistent with the measured data. Finally, the PICCS reconstructed image is further cleaned up by a trained light duty U-Net to improve noise texture and reduce noise to generate the final reconstructed image. Both simulation data and human subject data were used to validate the proposed image reconstruction framework. In simulation studies, it is demonstrated that the final output corrected the distorted structures in the deep learning-only reconstruction with respect to the shape, size and contrast of the structures. The final images also appeared streak-free with more natural noise texture when compared with the PICCS-only reconstruction. In human subject validation, the false positive lesion-like structures in the deep learning prior image were eliminated in the final output.

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

Helical CT has been widely used in clinical diagnosis. Sparsely spaced multidetector in z direction can increase the coverage of the detector provided limited detector rows. It can speed up volumetric CT scan, lower the radiation dose and reduce motion artifacts. However, it leads to insufficient data for reconstruction. That means reconstructions from general analytical methods will have severe artifacts. Iterative reconstruction methods might be able to deal with this situation but with the cost of huge computational load. In this work, we propose a cascaded dual-domain deep learning method that completes both data transformation in projection domain and error reduction in image domain. First, a convolutional neural network (CNN) in projection domain is constructed to estimate missing helical projection data and converting helical projection data to 2D fan-beam projection data. This step is to suppress helical artifacts and reduce the following computational cost. Then, an analytical linear operator is followed to transfer the data from projection domain to image domain. Finally, an image domain CNN is added to improve image quality further. These three steps work as an entirety and can be trained end to end. The overall network is trained using a simulated lung CT dataset with Poisson noise from 25 patients. We evaluate the trained network on another three patients and obtain very encouraging results with both visual examination and quantitative comparison. The resulting RRMSE is 6.56% and the SSIM is 99.60%. In addition, we test the trained network on the lung CT dataset with different noise level and a new dental CT dataset to demonstrate the generalization and robustness of our method.

A fast scan with a high helical pitch is desirable for many CT exams, such as pediatric, chest, and some of cardiovascular exams, to suppress patient motion artifacts. However, on a single-source scanner, the pitch typically cannot exceed ~1.5 without generating image distortion within the entire scanning field of view due to insufficient data acquired in a fast pitch mode. In this work, we developed a deep convolutional neural network-based approach to reducing artifacts on images reconstructed from insufficient data acquired in an ultra-fast-pitch mode (𝑝𝑝 ≥ 2.0). This custom-designed neural network, referred to as Ultra-fast-pitch image reconstruction neural network (UFP-net) consists of functional modules using both local and non-local operators, as well as the z-coordinate of each image, to effectively suppress the location- and structure-dependent artifacts induced by the fast-pitch mode. The UFP-net was trained using a customized loss function that involves image-gradient-correlation loss and feature reconstruction loss. Projection data at a regular pitch (𝑝𝑝 = 1.0) and a fast-pitch (𝑝𝑝 = 3.0) were simulated using 10 patient CT cases to generate training and validation datasets. Compared to filtered-back-projection (FBP), the UFP-net largely suppressed image artifacts and restored anatomical details. The structural similarity index (SSIM) was significantly improved (Mean SSIM: UFP-net 0.9, FBP 0.6), and the root-mean-square-error (RMSE) was largely reduced (Mean RMSE: UFP-net 57 HU, FBP 273 HU). The proposed method has the potential to enable ultra-fast-pitch data acquisition on single-source CT scanners to improve scanning speed while maintaining image quality.

Advancements in deep learning and GPU computing have exponentially driven the application of neural networks to classic medical imaging problems: denoising, segmentation, artifact removal, etc. Deep learning solutions are particularly attractive for processing multi-channel, volumetric image data, where processing and reconstruction methods are often computationally expensive. Convolutional neural networks (CNNs) are commonly applied to multi-channel image data by matching the number of network input channels to the number of data channels, learning explicit relationships between channels. This provides a high degree of specificity to a particular problem, but may fail to generalize to a broader class of closely related problems. We propose a solution to this generalization problem in the context of spectral x-ray CT, where the scanning kVps (energy bins) and contrast are often variable. Specifically, we propose a novel CNN architecture which handles variable numbers of input channels, variable noise levels between channels, and variable modes of spectral contrast. We demonstrate our architecture in the application of preclinical, photon-counting, micro-CT, effectively replacing 1-2 hours of iterative reconstruction, with <10 minutes of analytical reconstruction and CNN regularization. Experimental validation shows the effectiveness of our approach when applied to both in vivo photon-counting validation data (4 energy thresholds) and to simulated, dual-energy CT data virtually acquired with an energy integrating detector. In both cases, the results output by the CNN provide greater spectral accuracy than analytical reconstruction alone, but suffer from some degradation of spatial resolution. We conclude by proposing several extensions of our work to better preserve spatial resolution.

With the development of deep learning (DL), many deep learning (DL) based algorithms have been widely used in the low-dose CT imaging and achieved promising reconstruction performance. However, most DL-based algorithms need to pre-collect a large set of image pairs (low-dose/high-dose image pairs) and trains networks in a supervised end-to-end manner. Actually, it is not feasible in clinical to obtain such a large amount of paired training data, especially for high-dose ones. Therefore, in this work, we present a semi-supervised learned sinogram restoration network (SLSR-Net) for low-dose CT image reconstruction. The presented SLSR-Net consists of supervised sub-network and unsupervised sub-network. Specifically, different from the traditional supervised DL networks which only use low-dose/high-dose sinogram pairs, the presented SLSR-Net method is capable of feeding only a few supervised sinogram pairs and massive unsupervised low-dose sinograms into the network training procedure. The supervised pairs are used to capture critical features (i.e., noise distribution, and tissue characteristics) latent in a supervised way and the unsupervised sub-network efficiently learns these features using a conventional weighted least-squares model with a regularization term. Moreover, another contribution of the presented SLSR-Net method is to adaptively transfer learned feature distribution from supervised subnetwork with the paired sinograms to unsupervised sub-network with unlabeled low-dose sinograms to obtain high-fidelity sinogram with a Kullback-Leibler divergence. Finally, the filtered backprojection algorithm is used to reconstruct CT images from the obtained sinograms. Real patient datasets are used to evaluate the performance of the presented SLSR-Net method and the corresponding experimental results show that compared with the traditional supervised learning method, the presented SLSR-Net method achieves competitive performance in terms of noise reduction and structure preservation in low-dose CT imaging.

Recently, multi-energy inter-pixel coincidence counter (MEICC) has been proposed for charge sharing correction and compensation for photon counting detectors (PCDs), which uses energy-dependent coincidence counters between a pixel-of-interest (POI) and its neighboring pixels and keeps the book of charge sharing events. A Monte Carlo (MC) simulation study was performed to assess the performance of MEICC; however, the performance might have been overestimated in a previous study. The charge sharing increases the number of photons recorded at a PCD pixel at the expense of the spatial resolution and the effect on the spatial resolution was not taken into account. The aims of this study were, first, to outline a spatially-resolved irradiation scheme for a normalized spatial resolution, and second, to assess the performance of MEICC. The MC program was modified to set the x-ray spectrum and intensity for the POI and those for the neighboring pixels separately. Data were then generated by varying two basis materials, water and bone, by a small amount only for the POI, while those for neighboring pixels remained unchanged. Cramér–Rao lower bound (CRLB) of water and bone estimation was computed from the changes in the mean and covariance matrix estimated. This setting allowed for assessing the CRLB at the same spatial resolution regardless of with or without charge sharing. The CRLBs at the normalized spatial resolution were worse than those at not-normalized spatial resolutions. With eight energy windows, MEICC provided the CRLB comparable to the ideal PCD with no charge sharing. The proposed MEICC has a potential to address charge sharing in PCDs.

Recently we have reported indirect X-ray photon-counting imaging using CMOS photon detectors (CPD) and have shown its high spatial resolution with MTF of over 0.7 at 10LP/mm [1]. However, at that time its energy resolution potential and the quantum efficiency were totally unknown. Thus, there was a question about whether it can detect relatively low energy X-ray photons, for example around 20keV used for mammography, with sufficient quantum efficiency while eliminating counting errors.

In this study we exposed the CPD test devices to near single energy X-ray photons of 19.5keV adopting a clinical mammography equipment and additional Mo filters, and measured output intensity distributions. We also fitted our intensity distribution model to the results estimating signal yield per keV and parameters for signal variation.

We tested with 2 types of Hamamatsu Photonics CsI(Tl) scintillators. A CPD with J6675-01 scintillator plate has a signal distribution of 76% FWHM, still showing sufficient capability of photon counting with a little loss of actual signals. The signal yield is about 3.6 e-/keV. In the meanwhile, a CPD with J6675 has a superior distribution of 49% FWHM with signal yield of 7.1 e-/keV.

In both types, the dominant factor of the variation is not shot noise or sensor noise but something that is proportional to the signal intensity. This variation factor is possibly attributed to random noise of scintillation intensity, which is about 28% rms in J6675-01 and 17% rms in J6675.

One of the existing prototype detector systems for full-field photon-counting CT is a silicon detector developed by our group. Spatial resolution is clinically important to resolve small details and can enable more efficient phase-contrast imaging. However, improving the resolution is difficult as decreasing the pixel size is associated with technical challenges. By integrating CMOS electronics into the silicon sensor, it is possible to reduce the pixel size drastically while also introducing on-sensor data processing capabilities. In this work, we evaluate the feasibility of measuring the charge cloud shape of Compton interactions in a silicon strip detector to increase the spatial resolution. With an incident spectrum of 140 kVp, Compton interactions constitute 66.2% of the detected interactions. By combining a Monte Carlo photon simulation with a charge transport model, we study the charge cloud distributions and induced currents as functions of the interaction position. For a simulated silicon strip detector with a pixel size of 12×500 μm² , we present a method in which the interaction position can be determined. For an ideal case without electronic noise an average absolute error of 0.65 μm is obtained in the direction along the wafer and 13.08 μm in the trans-wafer direction. With simulated electronic noise and a lowest threshold of 0.88 keV the corresponding values are 1.38 μm and 122.83 μm. Our results show that the proposed method has the potential to very significantly increase the spatial resolution in a full-field photon-counting detector for CT.

The maturation of photon-counting detector (PCD) technology promises to enhance routine CT applications with high-fidelity spectral information, transforming X-ray CT into a functional and molecular imaging modality, while maintaining the high spatial resolution, fast scanning times, and relatively low cost. We present our novel dual source pre-clinical micro-CT prototype system that combines a photon counting detector (PCD) and an energy integrating detector (EID) in a single system. We compare performance for PCD-only with combined hybrid (i.e. PCD and EID) CT imaging with equal dose. All CT data sets (EID-only, PCD-only, and hybrid) were reconstructed using our iterative, multi-channel algorithm based on the split Bregman method and regularization with rank-sparse kernel regression. We used a post-reconstruction decomposition method with iodine (I), gold (Au), gadolinium (Gd), and calcium (Ca) basis functions. Performance was assessed using micro-CT phantoms. Our results show the spatial resolution of the PCD-only and the hybrid reconstructions were similar and slightly better than for EID-only reconstruction with ~3.5 lp/mm (10% MTF). The NPS of the hybrid reconstruction was similar to PCD-only reconstruction. By co-reconstructing EID and PCD data, we achieved better image quality in the material decomposition but marginal differences in terms of concentration accuracy. Hybrid spectral micro- CT can benefit both nanotechnology and cancer research by providing imaging that can help test and optimize various NPs for theranostics.

Practical photon counting detectors that have been adopted for commercial use are typically based on crystalline or polycrystalline materials. However, these types of materials are challenging to scale to large-area medical imaging applications because of yield and cost issues associated with the crystal growth and bonding technology required to interface the sensor with the readout IC. An alternate approach is to use a large-area-compatible, mature, direct conversion X-ray-detection sensor such as amorphous selenium (a-Se). The technical challenges for photon counting with a-Se lie in overcoming (1) the slow carrier-transport material property of a-Se, which leads to count-rate limitations due to pile-up, and (2) the lower X-ray-to-charge conversion gain, which degrades SNR and can be resolved by improved design of pixel readout circuits. In this paper, we address the a-Se material limitation by leveraging a unipolar charge sensing detector design. We demonstrate that the proposed unipolar charge sensing detector provides an effective method to detect charge of the polarity type having a higher mobility-lifetime product, obviating the need for detection of the opposite polarity slow transport charge. Transient signal measurements indicate that a quasi depth independent signal rise-time is achieved with the unipolar charge sensing detector. Moreover, two orders of magnitude improvement is observed compared to the conventional a-Se detector rise-time (0.15 μs vs. 25 μs).

Wide-angle digital breast tomosynthesis (DBT) gives better depth resolution and tissue separation while the microcalcification (MC) detectability is impacted by many factors, such as detector performance, focal spot motion (FSM), increased noise due to scatter radiation, angular dose distribution and image reconstruction methods. This study aims at developing an in-silico experimental pipeline to compare these factors’ influence on the MC detectability, so that new system design can be proposed to improve the performance. The VICTRE tool developed by Food and Drug Administration was used to generate a 49 mm-thick anthropomorphic digital breast phantom and the projection images using Monte Carlo simulation. We inserted 38 MC clusters with size of 120 μm at 40 mm from the bottom of the phantom. The Monte Carlo simulation used parameters of a clinical wide-angle DBT system, with 25 projections over 50-degree angular range, and 28 kVp W/Rh energy spectrum. Projection images were simulated under different scenarios: uniform/nonuniform angular dose distribution, with/without FSM of 2 mm, with/without scatter radiation, and TFT/CMOS detector types. A four-alternative forced choice (4AFC) methodology was employed to evaluate the MC detectability. The percentage correct and visibility score under all scenarios were compared. The results show that MC detectability decreases with the presence of either increased noise due to scattered radiation, or image blur due to FSM. Nonuniform angular dose distribution improves the detectability of MCs when using FBP reconstruction with narrow slice thickness filter. MC sharpness is improved with CMOS detector with smaller pixel size and lower electronic noise.

The limited angle and limited number of projections in digital breast tomosynthesis (DBT) produce under-sampled datasets that may compromise calcification detection. Small breast lesions, such as microcalcifications, may not be discernible without sufficient sampling in the reconstructed DBT images. We propose a virtual clinical trial (VCT) method to evaluate the calcification detection in DBT using computer simulations of breast phantoms, images, and virtual readers. We used multiple-reader multiple-case (MRMC) receiver operating characteristic (ROC) analyses to evaluate the performance of channelized Hotelling observers (CHOs) in calcification detection. The angular motion path of the x-ray source was varied to simulate different DBT acquisition geometries. We simulated continuous and step-and-shoot x-ray source motion and three angular motion paths: ±7.5°, ±15°, and ±25°. The detection of calcifications is affected by the angular motion path, particularly for the ±25° angular range, combined with continuous tube motion, larger detector element sizes (0.14 mm) and larger reconstructed voxel sizes (0.10 mm). When an angular range of ±25° is compared to ±7.5°, the difference in the area under the curve (AUC) is -0.030 (d’ ratio=0.633) and -0.067 (d’ ratio=0.584), for one and two- voxel calcifications (0.1 mm³ and 0.2 mm³), respectively. There is no significant difference in calcification detection using images acquired with ±7.5° and ±15°. The results provide insight on the impact of angular range for calcification detection, an ongoing limitation of tomosynthesis.

Aim: Investigate 3D structured DBT phantoms with lesion models for use in the evaluation of synthetic mammography (SM) imaging performance.

Methods: 4 phantoms were investigated: CDMAM, L1, CIRS BR3D and Modular DBT Phantom (two different inserts). The phantoms were imaged on recent DBT models: Fujifilm Amulet Innovality (ST mode), GE HC Senographe Pristina, Hologic 3Dimensions, IMS Giotto Class and Siemens Mammomat Revelation. Images were acquired at automatic exposure control (AEC) level, half AEC and twice AEC. SM was calculated. The CDMAM and L1 phantom were read by human readers via a 4-Alternative Forced Choice method and thresholds were established. CIRS BR3D and Modular DBT Phantom were analysed by counting visible lesions.

Results: The scores obtained from the phantoms had the same tendencies among systems. The phantoms highlight many specific characteristics of the SM algorithms such as tuning contrast enhancement to a range of sizes. The phantoms confirm, as in 2D and DBT, an impact of dose on detectability of microcalcification-like inserts but not on masses. None of the phantoms evaluate the SM for different glandular tissue or thickness distributions.

Conclusion: For all phantoms, SM found a number of lesion-like targets and an impact of dose as expected. Whether these phantom readings are representative for quality in SM in real practice is not yet proven. More elaborated sensitivity studies should be done prior to the use of the phantoms in routine QC. Ultimately, accurate assessment of SM may have to be done via virtual trials.

Purpose: In this work we present equivalent breast thickness and dose sensitivity of a next iteration 3D structured breast phantom with lesion models to demonstrate its potential use for quality assurance measurements in breast imaging. Methods: PMMA equivalent thickness was determined employing the automatic exposure control (AEC) of Siemens Mammomat Inspiration and Siemens Mammomat Revelation. A 2D projection image of the phantom was acquired and the corresponding AEC settings recorded as reference. Equivalent PMMA thickness was found by interpolating between three PMMA thicknesses with mAs values close to the reference settings selected by AEC. Dose sensitivity of the reconstructed digital breast tomosynthesis (DBT) images was assessed by two experienced readers using a four alternative forced choice (4-AFC) study. Three different dose levels for lesion models and microcalcifications were evaluated. Results: PMMA equivalent thickness of the phantom was 46.8 mm and 47.0 mm for measurements on Siemens Mammomat Inspiration and Siemens Mammomat Revelation which equals to a breast equivalent thickness of 55.5 mm and 55.8 mm, respectively, compared to a physical phantom thickness of 53.5 mm. For lesion models dose sensitiviy of the detectability was not obvious. For microcalcification the diameter threshold was found to increase for decreasing dose from high dose to AEC to low dose. Conclusions: We found the measured equivalent breast thickness of our phantom to be close to its physical thickness. It can be concluded that changes in dose can be detected by the presented phantom for the tested dose levels.

This study introduces a methodology for generating high resolution signal profiles of microcalcification (MC) grains for validating breast CT (bCT) systems. A physical MC phantom was constructed by suspending calcium carbonate grains in an agar solution emulating MCs in a fibroglandular tissue background. Additionally, small Teflon spheres (2.4 mm diameter) were embedded in the agar solution for the purpose of fiducial marking and assessment of segmentation accuracy. The MC phantom was imaged on a high resolution (34 μm) commercial small-bore μCT scanner at high dose, and the images were used as the gold-standard for assessing MC size and for generating high resolution signal profiles of each MC. High-dose bCT scans of the MC phantom suspended in-air were acquired using 1×1 binning mode (75 μm dexel pitch) by averaging three repeat scans to produce a single low-noise reconstruction of the MC phantom. The high resolution μCT volume data set was then registered with the corresponding bCT data set after correcting for the bCT system spatial resolution. Microcalcification signal profiles constructed using low-noise bCT images were found to be in good agreement with those generated using the μCT scanner with all differences <10% within the VOI surrounding each MC. The MC signal profiles were used as detection templates for a non-prewhitening-matched-filter model observer for scans acquired in a realistic breast phantom at 3, 6, and 9 mGy mean glandular dose. MC detectability using signal templates derived from bCT were shown to be in good agreement with those generated using μCT.

Based on chemical composition, two types of microcalcifications are found within the breast. Type I calcifications consist of calcium oxalate (CaOx), whereas Type II calcifications are composed of calcium phosphates, mainly hydroxyapatite (CaHa). Previous tissue analysis studies^{1, 2} have shown that Type I calcifications are typically found in benign lesions, and are not associated with carcinoma. Type II calcifications, on the other hand, can be found in both benign and malignant breast lesions. It was estimated¹ that lesions with solely Type I calcifications represented 12-17% of benign biopsies performed. Therefore, an imaging method that could differentiate calcified lesions with solely CaOx calcifications might help in reducing unnecessary breast biopsies. Previous studies^3-5 have shown that dual-energy mammography can be an effective tool in discriminating between these two classes of calcifications. In this work we investigate if a convolutional neural network (CNN) can be trained to distinguish between these two types of calcified clusters, using dual-energy x-ray mammograms. We implement a Monte Carlo mammography simulator package to model a dual-energy FFDM system and use a realistic anthropomorphic digital breast phantom model⁶ with embedded microcalcifications. A Resnet CNN architecture was used to classify the ROIs containing either CaOX, or CaOx clusters. We report classification accuracy results in terms ROC curve and the corresponding area under it.

Dense breast tissue has been shown to reduce the sensitivity and specificity of breast cancer detection. Estimation of breast density (BD) in digital breast tomosynthesis (DBT) may provide masking risk assessment in quasi three dimensions (3D). Dual energy (DE) mammography has been used for breast tissue decomposition, which may provide more accurate BD map for DBT projections. This study aims to investigate the feasibility of DE DBT for volumetric BD estimation and 3D masking risk assessment. A compressed digital breast phantom (with ground truth for BD) was simulated using a virtual clinical trial software (VICTRE). Low and high energy projection images of the phantom were simulated emulating the Siemens Mammomat Inspiration DBT system. DE decomposition was performed for each projection angle using a library of calibrated DE attenuation data for various combinations of breast thickness and glandularity. The resulted BD maps for all projection angles were used for simple-back-projection image reconstruction, and the dense regions for each image slice were segmented based on the reconstructed voxel values. This method was applied to clinical images from DE DBT acquisition after scatter correction. Our results show that DE decomposition provides consistent BD map for each projection angle and throughout the entire breast (including the periphery with thickness roll-off), and consistent measurements of the total breast volume, fibroglandular tissue volume and volumetric BD. The 3D segmented fibroglandular tissue map can be used to assess the masking risk for the DBT image volume.

Mammography and breast CT are important tools for breast cancer screening and diagnosis. Current implementations are limited by scattered radiation and/or spatial resolution. In this work, we propose and develop a slot scan-based system to be used in both mammography and CT mode that can limit scatter and collect sparse CT data for improved image quality at low radiation exposures. Monte Carlo simulations of an anthropomorphic breast phantom show a factor of 10 reduction in scattering amplitude with our slot scan-based system compared to that of a full-field detector mammography system (area mode). Similarly, slot-scan improved the MTF (particularly the low-frequency response) compared to an area detector. Investigation of sparse CT sampling with doubly sparse acquisition data return better quality reconstruction, for which our slot-scanning system is capable, over angle-only projection. Thus, a system with the combined ability for slot-scanning mammography and slot-scanning breast CT has the potential to deliver improved dose-efficient imaging performance and become viable breast cancer screening and diagnostic tools.

To reduce noise and enhance the contrast-to-noise ratio (CNR) of microcalcifications (MCs) in digital breast tomosynthesis (DBT), we conducted a study to investigate the feasibility of performing denoising on the projection views (PVs) using a deep convolutional neural network (DCNN) before reconstruction. We fine-tuned a modularized adaptive processing neural network (MAPNN) based on a pretrained model for CT image denoising. Four phantom DBTs containing over 700 simulated MCs of 3 nominal sizes were scanned with a DBT system that acquired 9 PVs within a 25° scan angle at two dose level settings to form the training and validation sets. Nine human subject DBTs were used as an independent test set. We trained the DCNN with low dose PVs as the input and the corresponding high dose PVs as the reference. We marked the MCs in the PVs and designed a loss function for DCNN training that balanced the effect of noise reduction and signal preservation. The loss function was a weighted sum of the perceptual loss, the adversarial loss and the CNR loss. A visual comparison of the DBT volumes reconstructed from the denoised PVs indicated that the proposed method could reduce noise and preserve the texture of the background without blurring subtle MC signals. A quantitative comparison showed a significant CNR improvement (p < 0.0001) in both the validation phantom and the human subject DBTs.

The recent development of compact, cold cathode X-ray emitters presents the opportunity to use multiple emitters arranged in an array to perform medical imaging from a stationary source. This study considers some of the potential advantages in their application to digital breast tomosynthesis and breast specimen imaging. It presents modelling results from simulated images based on a validated toolkit that models the X-ray source, breast tissue and X-ray image receptor performance. It also presents results from physical experiments using flat panel source geometries. Stationary sources eliminate focal spot motion blur which can improve the visualization of small features such as microcalcifications. A rectangular array of emitters can provide enhanced depth resolution compared to a linear sweep of emitter positions, as the projection images are obtained over a ‘sweep’ in two directions. Moreover, a rectangular array gives the opportunity to reduce the source to image distance (SID). Halving this distance reduces the required beam current by four times, which could significantly reduce the size, weight, and cost of an overall system, albeit with an increase in the angular width of the focal spot. Both the simulation and physical experiments show that depth resolution is considerably improved when using a rectangular array in comparison to a linear array. If a rectangular array is used at half the SID, and a quarter of the power, then this benefit is maintained, even though the focal spot has a larger angular width. This could provide enhanced imaging from lower-cost, smaller devices compared with conventional systems.

Developing PET reconstruction algorithms with improved low-count capabilities may provide a timely and cost- effective means of reducing radiation dose in promising clinical applications such as immuno-PET that require long-lived radiotracers. For many PET clinics, the reconstruction protocol consists of postsmoothed ordered-sets expectation-maximization (OSEM) reconstruction, but penalized likelihood methods based on total-variation (TV) regularization could substantially reduce dose. We performed a task-based comparison of postsmoothed OSEM and higher-order TV (HOTV) reconstructions using simulated images of a contrast-detail phantom. An anthropomorphic visual-search model observer read the images in a location-known receiver operating characteristic (ROC) format. Acquisition counts, target uptake, and target size were study variables, and the OSEM postfiltering was task-optimized based on count level. A psychometric analysis of observer performance for the selected task found that the HOTV algorithm allowed a two-fold reduction in dose compared to the optimized OSEM algorithm.

Most existing fully automatic or semi-automatic medical imaging segmentation methods start from reconstructed images. However, a framework for joint segmentation and image reconstruction can be beneficial because both tasks can be mutually dependent. Better segmentation can improve image reconstruction and vice-versa. We propose to perform joint PET image reconstruction and fully automatic PET image segmentation, using the CT image from a PET-CT scanner as a given input. Within a unified framework, our proposed method generates a PET image and a segmentation mask utilizing two connected trained networks: 1) a network dedicated to denoising the PET image with boundary information from the segmentation network. While reconstructing the PET image, the algorithm exploits the denoised image recovered from the trained network. 2) a segmentation network dedicated to estimating the lesion and background (e.g., liver) masks using PET/CT information. A boundary indicator image is generated based on the gradients of segmentation masks. We simulated extremely low-count PET, typical for Y-90 imaging, where traditional segmentation and reconstruction methods tend to perform poorly. For PET reconstruction, proposed method using true boundary improves CNR (RMSE) by 28.9 % (49.1%) and 16.8 % (13.2%) compared to EM and proposed method without using boundary. For multi-modal segmentation, our proposed method improved global Dice score in tumor by 70.6% compared to our proposed segmentation framework using only CT information.

Medical image reconstruction is often an ill-posed inverse problem. In order to address such ill-posed inverse problems, prior knowledge of the sought after object property is usually incorporated by means of regularization. For example, sparsity-promoting regularization in a suitable transform domain is widely used to reconstruct images with diagnostic quality from noisy and/or incomplete medical data. However, sparsity-promoting regularization may not be able to comprehensively describe the actual prior information of the objects being imaged. Deep generative models, such as generative adversarial networks (GANs) have shown great promise in learning the underlying distribution of images. Prior distributions for images estimated using GANs have been employed as a means of regularization with impressive results in several linear inverse problems in computer vision that are also relevant to medical imaging. However, in practice, it can be difficult for a GAN to comprehensively describe prior distributions, which can potentially lead to a lack of fidelity between the reconstructed image and the observed data. Recently, an image-adaptive GAN-based reconstruction method (IAGAN) was proposed to guarantee stronger data consistency by adapting the trained generative model parameters to the observed measurements. In this work, for the first time, we apply the IAGAN method to reconstruct images from undersampled magnetic resonance imaging (MRI) measurements. A state-of-the-art GAN model called Progressive Growing of GANs (ProGAN) was trained on a large number of ground truth images from the NYU fastMRI dataset, and the learned generator was subsequently employed in the IAGAN framework to reconstruct high fidelity images from retrospectively undersampled experimental k-space data in the validation dataset. It is demonstrated that by use of the GAN-based reconstruction method with noisy and/or incomplete measurements, we can potentially recover fine structures in the object that are relevant for medical diagnosis that may be difficult to achieve using traditional reconstruction methods relying on sparsity-promoting penalties.

Structural MRI is the standard-of-care imaging modality for screening and diagnosis of most neurological conditions. However, the ability to reliably evaluate brain MRIs for disease characterization (whether by machines or experts) is often hampered by the presence of artifacts such as magnetic field inhomogeneity, aliasing, or patient motion; some of which may not be visually apparent. Reliable quality assessment of brain MRI scans would allow for excluding noisy acquisitions and reducing errors in subsequent downstream analyses. Since visual inspection is impractical for large volumes of data and subject to inter-observer variability, there is a need for accurate, automated Quality Assessment (QA) of MR images. Previous studies have investigated image quality metrics (IQMs) in order to quantify the effect of specific artifacts and quality degradation in an MR image. There has also been some recent work in developing machine learning models which use IQMs to enable robust QA across multiple sites. We build on this approach by leveraging a total of 64 IQMs (quantifying noise, artifacts, information, and general brain measurements) within a Dense Neural Network (DNN) model for QA of brain MRI scans using the publicly available multi-site ABIDE-I cohort (17 sites, 1102 subjects). In attempting to predict the "mean opinion score" of MR image quality (as assessed by expert radiologists), the DNN model yielded an accuracy of 87.3±2.1% in training (15 sites, 885 subjects) which generalized to a 79.5±1.9% accuracy in validation (2 sites, 246 subjects). This initial DNN model suggests the promise of deep learning to improve automated QA of MRI scans in large multi-site, multi-scanner cohorts.

Direct active matrix flat panel imagers (AMFPIs) using amorphous selenium (a-Se) offer high intrinsic spatial resolution but have limited x-ray quantum efficiency at general radiographic energies due to selenium’s low atomic number. Conversely, indirect AMFPIs using inorganic scintillators typically have superior x-ray quantum efficiency at these energies, but inferior spatial resolution and increased noise due to optical effects in the scintillator. These inherent limitations motivate alternative AMFPI designs to further improve detector xray sensitivity and signal-to-noise performance. Towards this goal, this work constructs and experimentally investigates the x-ray imaging performance of a novel direct-indirect prototype imager referred to as Hybrid AMFPI. The imager comprises a direct conversion a-Se layer that may be coupled to an interchangeable scintillator screen through a transparent blocking layer and bias electrode. In this direct-indirect “hybrid” configuration, a-Se serves as both an x-ray and optical sensor. Readout is performed by a thin-film transistor array with 85 μm pixel pitch. The prototype imager’s x-ray sensitivity, modulation transfer function (MTF), noise power spectrum (NPS) and detective quantum efficiency (DQE) are measured in a direct AMFPI configuration (i.e. a-Se alone) and in a Hybrid configuration under identical x-ray exposure conditions and the results are compared. Contrast-detail and spatial resolution phantoms are also imaged using direct, Hybrid and indirect AMFPI configurations under identical exposure conditions to evaluate differences in their imaging performance.

A new imaging x-ray detector with 194μm pixel conventional FPD imaging mode for large area visualization with detector field-of-views (FOV) ranging from 12 inch to 6 inch and a 76μm pixel high resolution high-definition (Hi-Def) zoom mode for smaller region-of-interest (ROI) visualization with FOV’s from 3 to 1.5 inches in one detector panel was built to improve visualization of endovascular devices during neurovascular interventions. In this work we investigate the effect of simultaneous use of both imaging modes on entrance kerma area product (KAP) and scatter at the detector. The entrance KAP and the scatter reaching the detector during simultaneous acquisition is the same as the large FOV FPD mode acquisitions, both higher than just Hi-Def mode acquisition. When a 0.7mm thick differential ROI Cu attenuator with a circular 10mm hole was aligned to the Hi-Def detector area, the entrance air kerma is reduced by 70%. To study the effect of scatter on image quality, Hi-Def images of the devices such as guide catheter and stents placed on a head equivalent phantom were acquired with the x-ray field opened to 8inch FOV, with and without the grid. Compared to the 3 inch Hi-Def FOV, with the grid the contrast of the guide catheter tip was reduced by 11% when the FOV was opened to 8 inch; however, without the grid the reduction in contrast was 36%. With the ROI attenuator and no grid the contrast was reduced by 16% when the FOV was opened to 8inch.

For a light burst generated by an x ray-scintillator interaction, the quasi-isotropic spreading of optical quanta determines that both the light collection efficiency and the spread function of the light burst depend on the depth (z) of interaction: the closer the initial position of light burst to the light sensor, the less light loss and spatial resolution degradation a detector system encounters. For scintillator-based indirect-conversion detectors, the variation of the x-ray conversion gain and the MTF along the z direction contributes to a significant fraction of DQE loss and is known as the Lubberts effect. Understanding the physics behinds the effect and developing solutions to reduce the effect have led to tremendous improvements in the performance of indirect-conversion detectors. In contrast, for direct-conversion photon counting detectors (PCDs), since no optical quanta generation or transport is involved, there is no classical Lubberts effect. However, the transport properties of the generated charge carriers in semiconductors do depend on their travel distance. As a result, the performance of direct conversion detectors may depend on z. This dependence can be considered as a counterpart of the classical Lubberts effect. The purpose of this work is to analyze the contributions of a variety of physical mechanisms that may contribute to a Lubberts-like effect in direct-conversion PCDs. The results show that the Lubberts-like effect is primarily caused by the depth-dependent charge trapping and recombination likelihood. Based upon the physics analysis, PCD electrode design can be optimized to minimize the Lubberts-like effect and the associate DQE loss.

For active matrix flat-panel imagers (AMFPIs), relatively high levels of additive noise compared to the average imaging signal per interacting x-ray limits DQE under conditions of low dose per frame – degrading imaging performance in applications such as digital breast tomosynthesis. The development of particle-in-binder, polycrystalline mercuric iodide (PIB HgI2) as an alternative converter is of interest since the material can provide significantly larger imaging signal per x-ray than converters based on CsI:Tl or a-Se. In order to address the high degree of charge trapping exhibited by HgI2 (leading to unacceptably large image lag), incorporation of a Frisch grid structure into the bulk of the converter is under investigation. Earlier theoretical studies demonstrated that such a grid can greatly reduce the contribution of hole charge to imaging signal – which should reduce the effects of charge trapping. In those studies, the grid was assumed to be positioned between two polycrystalline HgI2 layers – a configuration referred to as a floating grid. In this paper, early results from a theoretical examination of the performance of converters containing a grid supported on insulating pillars over an underlying AMFPI array and a single layer of PIB HgI2 material, are reported. The simulation results indicate that, while charge accumulating on the pillars changes the shape of electric fields that allow collection of electron signal, high electron collection efficiency accompanied by substantial suppression of hole signal can nevertheless be maintained through judicious selection of grid design and operational conditions. This work is supported by NIH grant R01- EB022028.

Cassette type digital radiography detectors consisting of indirect conversion flat panel detectors (FPDs) find extensive applications in general radiography. Recently, these detectors are being employed not only for conventional hospital and clinical use but also in the field of home medical care. Therefore, image quality improvement, X-ray dose reduction, and weight reduction are essential for such detectors. To achieve these goals, we have developed novel FPDs with high image quality and light weight by combining the irradiation side sampling (ISS) system technology and flexible photodiode and thin film transistor (PD-TFT) technology. The ISS system enables effective detection of light emitted from a scintillator near the photodiode, resulting in higher sensitivity and resolution than those obtained with common penetration side sampling (PSS) systems. In addition, by employing a 40-μm flexible polyimide film as the PD-TFT array substrate instead of a conventional 0.5-mm rigid glass, the X-ray attenuation and scattering at the substrate are expected to be reduced, resulting in further improvements in sensitivity and resolution. We have developed two types of ISS system flexible FPDs with Gd₂O₂S:Tb (GOS) and CsI as scintillators. These FPDs realized higher sensitivities, modulation transfer functions, and detective quantum efficiencies than the conventional ISS system rigid FPDs for both the scintillators. In addition, we achieved approximately 35% weight reduction with the flexible FPDs compared to similarly sized conventional rigid FPDs.

Microcolumnar CsI:Tl scintillator screens have been the gold standard in X-ray imaging for many years due to their high density, high atomic number, and scintillation efficiency. The structured screens provide an improvement in performance by channeling the light to the detector, improving detection efficiency and spatial resolution. We have taken this concept a step further by laser-machining the CsI:Tl scintillator to provide pixels that match the detector pixels. This allows for still thicker CsI:Tl layers up to 700 μm pixelated with pitch of 100 μm to match CMOS flat panel pixels, thus improving X-ray absorption and resolution. We are investigating the applications of CMOS detectors with pixelated scintillators for imaging of bone microarchitecture on diagnostic Cone Beam CT (CBCT) systems to provide improved quantitative metrics for diagnosis of osteoporosis and osteoarthritis.

The scintillator design includes reflective coatings applied to the laser-cut grooves to improve optical isolation between pixels. Such coatings are created by atomic layer deposition (ALD), a unique approach, which permits formation of reflectors over inter-pixel grooves with aspect ratios as high as 140:1. Here we present initial results quantifying performance gains in CMOS detector resolution and their impact on the quality of bone microstructure segmentation. We demonstrate 77% gain in spatial resolution at 2 lp/mm and extension of the limiting resolution from 3 lp/mm to 4.5 lp/mm for the CMOS detector with a pixelated screen compared to a commercial sensor. In a bench-top CBCT study emulating diagnostic systems for orthopedic applications (extremity CBCT), we achieved >0.75 correlations in metrics of trabecular microarchitecture between pixelated CsI:Tl based CBCT and gold-standard micro-CT. The pixelated scintillator is expected to have significant impact for many other applications including mammography and digital radiography, where resolution and dose efficiency (DQE) of the detector are of critical importance.

This Conference Presentation, "Impacts of pulse pileup, charge sharing, and anti-charge sharing on the noise power spectra of photon counting detectors: theoretical analysis and experimental demonstrations," was recorded at Medical Imaging 2020 held in Houston, Texas, United States.

Iodinated contrast agent is frequently used in computed tomography (CT) imaging to enhance organ contrast enhancement and improve diagnostic sensitivity. Despite this importance, there currently is a lack of standardization in contrast administration protocol across institutions, leading to many safety and clinical diagnostic risks. To solve this, we built three liver contrast enhancement/perfusion models: two using simple linear regression and another by combining a pre-existing pharmacokinetics mathematical model with clinical data with the eventual goal of individualizing contrast administration protocol to optimize contrast-enhanced CT imaging for each patient. These models primarily use patient attributes, such as height, weight, sex, age and contrast administration information, and bolus tracking information to make such predictions. 418 Chest/Abdomen/Pelvis CT scans were used in this study. 75% of cases were used to train these models and the rest were used to test the prediction accuracy. Pearson’s correlation coefficient test was used to find the correlations between the patient attributes and contrast enhancement in liver parenchyma. Weight, height, BMI, and lean body mass were found to be statistically significant predictors for contrast enhancement (P<0.05), with weight as the strongest predictor. Of the predictive models, we found that including bolus tracking information increases predictive accuracy (r2=0.75 v. 0.42) and that in the absence of bolus tracking information, combining clinical data with pre-existing pharmacokinetics model may provide the needed enhancement curve.

In recent years, there has been an increasing interest in replacing digital subtraction angiography (DSA) as method of choice for the diagnostic imaging of patients suffering from lower extremity peripheral arterial disease (PAD). Due to small vessel diameters and suboptimal resolution, examinations of below-the-knee arteries however remain extremely challenging. The advent of wide beam CT scanners allows to perform multiple CT acquisitions over a wide patient volume. A sequence of these CT acquisitions at timed intervals could provide additional hemodynamic information, and as such allows to track a contrast bolus that propagates through the arterial conduit. The aim of this study was to evaluate the accuracy and precision of ow velocity measurements using time-resolved computed tomography angiography (CTA). To this end, we constructed a mechanical ow phantom (single lumen, 6 mm inner-diameter). Six consecutive time-resolved CTA acquisitions were performed at a constant ow rate to achieve six reference velocities (21.2 mm/s, 38.9 mm/s, 60.1 mm/s, 81.4 mm/s, 99.0 mm/s and 120.3 mm/s). The mean centerline ow velocity was obtained from the contrast propagation over three different segmental lengths (160 mm, 80 mm and 40 mm) and then compared to the reference ow velocity. The results of this study suggest that mean ow velocities within the range of typical blood ow velocities in the below-the-knee arteries (40 mm/s - 70 mm/s), can be accurately measured with high precision in a 6 mm ow phantom using time-resolved CTA when considering a minimal path length of 80 mm.

Catheter-based vascular interventions performed in the interventional radiology suite rely heavily on digital subtraction angiography (DSA) for intraprocedural evaluation of vessel morphology. Recently, DSA methods have been reported which quantify blood flow based on the tracking of temporal oscillations in contrast agent density through vessels. This signal, termed contrast pulsatility, arises naturally from the mixing of a fixed rate contrast injection with periodic arterial blood flow. Although these tools promise quantitative intraprocedural evaluation of flow, quantification is limited by a decline in contrast pulsatility strength in distal arterial vasculature, and, in venous structures, minimally observable or no pulsatility. To improve pulsatility strength in DSA images, a custom dual-barrel power injector was constructed that dynamically modulates the injected contrast flow (or concentration). Each barrel can follow sinusoidal and square wave injection profiles of specified frequency, amplitude, phase, and duty cycle. Improvement in contrast pulsatility was evaluated in a vessel phantom with simulated 60 bpm arterial flow and low-frequency sinusoidal contrast injections of 0.25, 0.50, and 1.0Hz. The ability to establish contrast pulsatility in a venous structure was evaluated in a porcine study. In the arterial phantom, sinusoidal contrast injections at 0.25Hz, 0.5Hz, and 1.0Hz displayed increased distal contrast pulsatility relative to a DC contrast injection at the same mean rate, with energy spectral density (ESD) ratios of 215, 59.9, and 5.3 respectively. In the animal study, a custom 0.67Hz square wave injection into the inferior vena cava showed a large improvement in pulsatility relative to DC injection (ESD ratio of 7208).

Quantitative perfusion maps of cerebral blood flow, cerebral blood volume, and transit time generated using dynamic imaging data enable physicians to evaluate and prescribe the optimal plan of care for stroke patients. Validation is needed to increase the accuracy and reproducibility of this data, which can vary depending on the scanning technique and post-processing algorithm. In this work, we expand the XCAT brain phantom to incorporate a realistic model of the contrast agent dynamics in the cerebral vasculature and establish the ground truth to which the perfusion maps can be compared. For a specific stroke case, each tissue region's flow demand is calculated and used to determine the feeding flow in the upstream arteries and draining flow in the downstream veins. Using the flow values and a contrast agent injection curve, the model calculates the input and output concentration curves for each structure in the brain. The concentration curve within each structure is then calculated as the difference between the total amount of contrast agent that has entered and exited the structure up until that timepoint. A dynamic simulation framework utilizes the curves to define the contrast agent concentration within the phantom at each time point and generates simulated CT perfusion imaging data sets compatible with commercially available post-processing software. This development provides a realistic set of ground truth test data that enables quantitative validation and optimization of perfusion imaging and post-processing methods for stroke assessment.

Quantitative estimation of contrast agent concentration is made possible by spectral CT and material decomposition. There are several approaches to modulate the sensitivity of the imaging system to obtain the different spectral channels required for decomposition. Spectral CT technologies that enable this varied sensitivity include source kV-switching, dual-layer detectors, and source-side filtering (e.g., tiled spatial-spectral filters). In this work, we use an advanced physical model to simulate these three spectral CT strategies as well as hybrid acquisitions using all combinations of two or three strategies. We apply model-based material decomposition to a water-iodine phantom with iodine concentrations from 0.1 to 5.0 mg/mL. We present bias-noise plots for the different combinations of spectral techniques and show that combined approaches permit diversity in spectral sensitivity and improve low concentration imaging performance relative to the those strategies applied individually. Better ability to estimate low concentrations of contrast agent has the potential to reduce risks associated with contrast administration (by lowering dosage) or to extend imaging applications into targets with much lower uptake.

By integrating three absorption gratings with a medical X-ray tube and a medical flat panel detector, a preclinical X-ray multi-contrast lung imaging prototype with a large field-of-view up to 40 × 50 cm² was developed in this work. The prototype system provides phase-contrast and dark-field images simultaneously in addition to the conventional X-ray absorption imaging. Unlike the traditional grating-based phase-contrast imaging (GPCI) system, the scanning object is placed behind the pattern-generating grating G1, reducing the radiation dose to probably half of the original system design. For a fast scan process, three absorption gratings inducing a large field-of-view were utilized. Besides, the X-ray tube, the three gratings and the detector are all fixed on the same mechanical arm, scanning the object along the vertical direction. A numerical simulation platform was developed to help optimize the imaging parameters. Finally, the preliminary experimental results of a full-body enthanized rabbit demonstrate the capability of the multi-contrast prototype system, and its dark-field images are capable of providing additional diagnostic value for the diagnosis of lung diseases. Direct evaluation of the potential clinical unity of the multi-contrast lung imaging can be further performed on this prototype system.

Three-material decomposition is crucial for material quantification when more than two elemental materials, including a K-edge material, are presented in an image object. In principle, three-material decomposition requires a triple energy scan which cannot be directly accomplished using a conventional dual energy CT system. In this work, a new scheme to enable three-material decomposition by employing phase contrast CT was presented. When a grating interferometer is added, a conventional absorption dual energy CT system can be upgraded to a phase contrast dual energy CT system which provides an additional phase signal related to the real part of the refractive index of an image object, along with the absorption signal under two different x-ray spectra. In this work, a three-material decomposition method was proposed for the aforementioned dual energy phase contrast CT system. Physical experimental studies were performed on a benchtop x-ray Talbot-Lau interferometer system to validate the proposed method. A physical phantom, containing calcium and iodine inserts of known concentrations, was used as the image object. A rotation-rotation dual energy phase contrast CT scan was performed under 40 and 80 kVp tube potentials. For each view angle, a phase stepping procedure with five phase steps was performed. After the phase retrieval procedure and image reconstruction using the standard filtered-back projection, the solutions were decomposed into the calcium, iodine and water bases based on the proposed decomposition method. For all the solutions, the relative quantification errors of the concentrations were within 10%.

X-ray phase contrast imaging has been proposed as a way to improve contrast in soft tissue. Phase retrieval methods have been developed which only require a single image acquisition, allowing phase contrast imaging to be used in combination with computed tomography without a dangerously large dose applied to the subject. Many of the existing methods use approximations regarding the knowledge of object properties as well as make assumptions such as a constant ratio of absorption and phase properties throughout the object.¹ Our group has previously developed a phase retrieval algorithm which only requires a single image when using energy-resolving photon counting detectors and uses no additional approximations regarding object properties. This allows us to decompose phase properties even if the object is composed of a large number of materials with very close material properties. Our recent results on simple phantoms show that we can accurately retrieve multiple material properties corresponding to absorption and phase effects for a range of different materials. The strength of our spectral decomposition method is compounded by the use of photon counting detectors with zero dark noise. In this paper we demonstrate the first propagation-based phase contrast tomography study of a biological sample using our method. Our phase retrieval method yields excellent results with very low counts - 1 second or less per projection acquisitions using a weak microfocal X-Ray tube. We show this via known phantoms as well as satisfactory preliminary results from biological samples. Additional studies will explore dose aspects using mouse models in our benchtop micro computed tomography platform and compare against attenuation-based computed tomography.

Studies have shown that the conventionally estimated visibility and differential phase signals in grating-based Talbot-Lau imaging system are intrinsically biased signals. Since such bias are mainly caused by applying the conventional signal estimation approach on noisy data, therefore, it remains an open question whether there has a better signal estimation method to reduce such bias. To answer this question, we proposed an end-to-end supervised deep computed signal estimation network (XP-NET) to extract the three unknown signals, i.e., the absorption, the dark-field, and the phase contrast. Numerical phase stepping data generated from natural images are utilized to train the network. Afterwards, both numerical and experimental studies are performed to validate the performance of the proposed XP-NET method. Results show that for high radiation dose levels, signals retrieved from the XP-NET method are identical as obtained from the conventional analytical method. However, the XP-NET method has the capability of reducing phase signal bias by as much as 15% when the radiation dose levels gets lower. As the phase signal becomes more unbiased, the phase images get more accurate.

This work investigates a metal artifact reduction technique that combines weighted masking of projection data corrupted by metal, improved photon-counting detector modeling, and the constrained ‘one-step’ spectral CT image reconstruction (cOSSCIR) algorithm. The cOSSCIR algorithm directly estimates the basis material maps from the photon counts data using an optimization algorithm that places constraints on the basis maps. The improved photon-counting detector spectral modeling improves the accuracy of the polyenergetic forward model, which is expected to reduce beam hardening artifacts due to metal. This study also explores weighting schemes to reduce the contribution of counts measurements corrupted by metal during reconstruction, including selective masking across energy window. Unlike two-step decomposition approaches, cOSSCIR does not require energy windows to be registered, thus enabling energy-selective masking of data corrupted by metal. Preliminary feasibility of the proposed methods was investigated through experimental photon-counting CT acquisition of a tissue specimen with metal inserts. The cOSSCIR algorithm estimated acrylic and aluminum basis maps which were combined to form a 50 keV effective monoenergetic image. The effective monoenergetic image reconstructed by cOSSCIR from all counts data demonstrated reduced streak and view aliasing artifacts compared to the reference filtered backprojection image. Weighting of the data corrupted by metal further reduced the remaining beam hardening artifacts, with weighted masking further reducing the streak artifacts.

Accurate quantitation of bone density variation would allow monitoring changes due to aging, disease progression, therapy outcomes or drug response. We present results from our material decomposition method using spectral computed tomography (CT) with a photon counting detector (PCD) for estimating quantitative bone density variations in healthy and arthritic mice. Limited samples of healthy mice and collagen antibody-induced arthritis (CAIA) mice were imaged on our home-built benchtop micro CT system with a wide area photon counting detector with a high resolution 55 micrometer pixel size. A material decomposition algorithm was applied in the image domain to separate and quantify bone and soft tissue variations. The resulting bone densities were quantified globally and locally for both mice. Higher bone densities were seen in the control mice over their arthritis counterparts by about 6-12%. Based on our preliminary results, material decomposition of multi-energy CT images collected with a PCD appears promising for bone density quantification. Our results show that these methods could be successfully applied to arthritis diagnosis and treatment monitoring in a mouse model. We will further investigate the utility of our methods to simultaneously quantify bone density as well as tagged contrast agents. These investigations are expected to help with longitudinal imaging studies to understand disease progression in animal models.

The introduction of photon-counting detectors in x-ray computed tomography raises the question of how reconstruction algorithms should be adapted to photon-counting measurement data. The transition from energyintegrating to photon-counting detectors introduces new effects into the data model, such as pure Poisson statistics and increased cross talk between detector pixels, (e.g. due to charge sharing), but it is still not known in detail how these effects can be treated accurately by the reconstruction algorithm. In this work, we propose a new reconstruction method based on penalized-likelihood reconstruction that incorporates these effects. By starting from a simple, easily-solved reconstruction problem and adding correction terms for the additional physical effects, we obtain a series expansion for the solution to the image reconstruction problem. This approach serves the twofold purpose of (1) yielding a new, potentially faster method of incorporating complex detector models in the reconstruction process and (2) providing insight into the impact of the non-ideal physical effects on the reconstructed image. We investigate the potential for reconstructing images from simulated photon-counting energy-resolving CT data with the new algorithm by including correction terms representing pure Poisson statistics and interpixel cross talk; and we investigate the impact of these physical effects on the reconstructed images. Results indicate that using two correction terms gives good agreement with the converged solution, suggesting that the new method is feasible in practice. This new approach to image reconstruction can help in developing improved reconstruction algorithms for photon-counting CT.

There is potential to improve CT imaging by adding spectral capabilities as given by photon counting detectors (PCD). Here we describe and assess performance of a new spectral micro-CT prototype system using a CdTe-based PCD with 100-μm pixel size (model XC-Thor, made by Direct Conversion) benefitting from anti-coincidence corrections. To assess the PCD in terms of spectral response, a Ba-133 nuclear source was scanned using full spectrum scanning by sweeping thresholds with 2 keV increments. In a different experiment, we used small vials containing water, iodine (I), gadolinium (Gd), and gold (Au) placed on the surface of the detector and acquired X-ray data in full spectrum mode to verify that the PCD threshold positions yielded the expected changes in contrast around the K edges of these elements. Detector performance was assessed using micro-CT phantoms and during cancer experiments in mice injected with nanoparticle (NP)-based contrast agents. Both phantom and mouse micro-CT data were reconstructed using our iterative, multi-channel algorithm based on the split Bregman method and regularization with rank-sparse kernel regression. A post-reconstruction decomposition method was used. The system is capable of high resolution (11.9 lp/mm, 10% MTF) tomographic imaging. Despite the anti-coincidence corrections, the spectral performance of the PCD is, however, not perfect, and it seems to be affected mostly at lower keVs, making accurate iodine decomposition challenging. Our cancer imaging results illustrate that our spectral micro-CT can benefit both nanotechnology and cancer research by providing an imaging method that can help test/optimize various nanoparticle for theranostics.

Virtual non-contrast (VNC) CT images derived from multi-energy CT data have demonstrated many valuable clinical applications. However, VNC CT has not yet established itself as a technology that can reliably replace the true non-contrast CT. One commonly observed phenomenon is an erroneous removal or reduction of calcium signal in VNC images. The purpose of this work is to develop a photon counting CT-based method to decouple the iodine signal from the calcium signal to achieve VNC CT imaging. Thanks to the energy resolving capability of the photon counting detector (PCD), a photon counting CT enables all single kV CT data to carry additional spectral information needed for VNC CT reconstruction. However, the energy discriminating capability of a real PCD system is far from ideal, which can severely degrade the fidelity of the encoded spectral information and the efficiency of the material decomposition. In this work, a physics-based model of the PCD energy response function was developed and experimentally validated. By leveraging this model, a method was developed to correct the distorted spectral information in the measured PCD energy bin data, allowing the true post image object spectrum to be estimated to accomplish accurate three-material decomposition and VNC CT reconstruction. Both numerical simulation and experimental results demonstrated that the proposed spectral distortion correction method can effectively improve the CT number accuracy of both iodine-containing vessels and calcium-containing bony structures in VNC CT images.

Temporal resolution in time-resolved cone-beam CT (TR-CBCT) imaging is often limited by the time needed to acquire a complete dataset for image reconstruction. With the recent developments of performing nearly limited-view artifact-free reconstruction from data in a limited-view angle range and a prior image, temporal resolution of TR-CBCT imaging can be improved from the initial 0.2 frame per second (fps) to 1 fps which indicated a factor of five improvement in temporal resolution. In this paper, a new technique was introduced to further improve temporal resolution from 1.0 fps to about 7.5 fps which indicates a factor of more than 30 times improvement in temporal resolution for slow C-arm CBCT acquistions. This new technique was developed for multi-sweep data acquistion protocol and it is referred to as enhanced SMART-RECON (eSMART-RECON) in this paper. Numerical phantoms with ground truth and in vivo human subject data were used to validate the proposed eSMART-RECON method.

Interventional guidance aims at providing the radiologist with detailed information about the location and orientation of interventional tools such as guide wires and stents. Most commonly, this is done by acquiring fluoroscopic images using an interventional C-arm system. Due to its projective nature, fluoroscopy is restricted to provide information from two spatial dimensions, preventing an exact 3D localization of the interventional tools. Analogous to computed tomography for diagnostic imaging, four-dimensional (three spatial dimensions plus the temporal dimension) interventional guidance has the potential to drastically improve both the speed and accuracy of such interventions, but is currently impractical due to the excessively high dose that would be necessary for continuous cone-beam CT (CBCT) scanning at high frame rates.

In this work we develop a novel deep learning-based approach to reconstruct interventional tools from only four x-ray projections. We train and test this deep tool reconstruction (DTR) network on simulated data. Only small deviations from the ground truth (GT) reconstruction of the tools were observed, both quantitatively and qualitatively, showing that deep learning-based four-dimensional interventional guidance has the potential to overcome the drawbacks of conventional interventional guidance in the future.

Purpose: Cone-beam CT is increasingly used for 3D guidance in interventional radiology (IR), but long image acquisition time results in degradation from complex deformable motion of soft-tissue structures. Deformable motion compensation with multi-region autofocus optimization was shown to improve image quality. However, the high dimensionality and non-convexity of the optimization problem challenge its convergence. This work presents preliminary development and early results obtained from an automatic learning-based decision framework to obtain local estimates of basic properties of the deformable motion field, coupled to a preconditioning strategy to simplify the optimization. Methods: Deformable motion properties are estimated with a deep convolutional neural network (CNN) consisting of a concatenation of custom-designed residual blocks. The preliminary design provided an estimate of the local motion amplitude on an 8x8 grid covering an axial slice of a motion-contaminated CBCT volume. The decision framework is coupled to a preconditioning strategy that effectively favors more likely solutions through motion amplitude-driven spatially-varying regularization of the motion trajectory and spatially varying selection of the search range for the optimization problem. The network was trained on simulated data generated from publicly available CT datasets, including simple motion fields. Results: Predictions of local motion amplitude showed good agreement with the true values, with root mean squared error (RMSE) < 10 mm for the complete range of motion distributions explored (sufficient for the intended purpose of initialization). Combination of amplitude prediction with spatially varying regularization and search range setting resulted in improved motion compensation after 1000 iterations of the preconditioned multi-motion autofocus in an example case with complex deformable motion. Extensive validation in a large dataset of complex, multi-motion patterns is underway. Conclusion: The proposed approach shows promising initial results and the potential for automatic local motion estimation with learning-based methods. Pending ongoing development to extend this initial development, the method could simplify and accelerate complex deformable motion compensation with spatially varying preconditioning of the motion estimation.

Model-based iterative CT reconstruction suffers from its high computational cost. To increase the reconstruction speed, algorithms using a momentum acceleration can be combined with the ordered subset approach. As shown in previous works, the incorporation of ordered subsets strongly improves the reconstruction time but makes the outcome of the reconstruction difficult to predict due to its empirical nature. Specifically, when the number of subsets is too high, divergence or convergence to an unsatisfactory result can easily appear. In this work, we propose a new combination of ordered subsets with momentum to achieve fast reconstruction in a more robust manner. Our approach, referred to as EOS-MFISTA, is an Efficient Ordered Subset strategy based on the steps of MFISTA, a monotonic version of FISTA. In short, EOS-MFISTA uses objective function values to decide on the next update step and does this with little increase in computational effort. The performance of EOS-MFISTA is evaluated on real CT data of a head phantom and a human chest. Starting from a zero image, 200 iterations of EOS-MFISTA are calculated and compared with other algorithms. Quantitative analyses based on the RMSE show that EOS-MFISTA is much more robust than OS-FISTA, especially when the number of subsets is increased to accelerate reconstruction. Visual inspection of results obtained with 50 iterations further supports these findings.

We report on the modeling, characterization, benchmarking, and optimization of an interventional cone beam CT system based on a dual layer X-ray detector by means of physics based simulations.

By Monte Carlo methods, we log the interaction and dose deposition (i.e. signal generation) of X-ray photons in the dual layer geometry, including scattering processes and fluorescence photon emission. From the spatial dose distribution inside the detection volume, we derive typical detector properties like X-ray spectral responses, detective quantum efficiencies 𝐷𝑄𝐸(0), and noise characteristics for particular detector layouts.

We apply these results in subsequent full system simulations to generate 3D imaging scans of dual layer spectral projections, for custom virtual phantoms containing inserts of e.g. blood sediment or iodine with different concentrations. These simulated images are used to calculate key performance indicators of the imaging system, like e.g. receiver operating characteristic based analysis of material separation capabilities.

We present a new concept of full-scale cone-beam computed tomography (CT) spectral imaging with simultaneous scatter correction capability using a spectral modulator and flying focal spot technology. The design of the spectral modulator and the utilization of flying focal spot are investigated, revealing the feasibility of using the flying focal spot to improve the spectral data sampling for a stationary modulator. Based on our analytical analysis, with the flying focal spot, the spectral projection ray misalignment can be reduced by a factor of more than 8. For modulator design, common materials such as Cu, ZnSe, Mo, Ag and Sn could be good candidates for better energy separation at a given attenuation of photon flux. A proof-of-concept study of a Catphan600 phantom and an anthropomorphic chest phantom was conducted on a tabletop cone-beam CT system using a copper modulator and by moving the X-ray tube position by 1.24 mm in the X and Z directions to mimic the flying focal spot. We found the low and high energy rays were well aligned, while the scatter distribution remained constant between the flying focal spot positions.

Dual Energy CT is a modern imaging technique that is utilized in clinical practice to acquire spectral information for various diagnostic purposes including the identification, classification, and characterization of different liver lesions. It provides additional information that, when compared to the information available from conventional CT datasets, has the potential to benefit existing computer vision techniques by improving their accuracy and reliability. In order to evaluate the additional value of spectral versus conventional datasets when being used as input for machine learning algorithms, we implemented a weakly-supervised Convolutional Neural Network (CNN) that learns liver lesion localization and classification without pixel-level ground truth annotations. We evaluated the lesion classification (healthy, cyst, hypodense metastasis) and localization performance of the network for various conventional and spectral input datasets obtained from the same CT scan. The best results for lesion localization were found for the spectral datasets with distances of 8.22 ± 10.72 mm, 8.78 ± 15.21 mm and 8.29 ± 12.97 mm for iodine maps, 40 keV and 70 keV virtual mono-energetic images, respectively, while lesion localization distances of 10.58 ± 17.65 mm were measured for the conventional dataset. In addition, the 40 keV virtual mono-energetic datasets achieved the highest overall lesion classification accuracy of 0.899 compared to 0.854 measured for the conventional datasets. The enhanced localization and classification results that we observed for spectral CT data demonstrates that combining machine-learning technology with spectral CT information may improve the clinical workflow as well as the diagnostic accuracy.

Energy-integrating-detector (EID)-based triple-beam multi-energy CT (TB-MECT) was recently implemented on a dual-source (DS) CT platform by mounting a z-axis split-filter (0.05 mm Au, 0.6 mm Sn) on one of the two tubes. The purpose of this work is to perform a feasibility animal study on this new MECT platform for a small bowel imaging task using two contrast agents, iodine and bismuth. Optimal triple-beam configurations, 70/Au140/Sn140 kV were determined in a phantom study for this task and applied in the animal study for best material decomposition imaging performance. The results demonstrated that the TB-MECT can successfully separate and quantify the two contrast agents from one single scan for the task of small bowel imaging.

Proton radiotherapy has the potential to provide clinically effective treatment and highly conformal dose delivery when the rapid dose falloff at the end of its proton-beam range is correctly aligned to the distal margin of the clinical target volume. However, in current clinical practice an additional 2-3.5% safety margin must be added to the proton range to account for uncertainties in the estimation of proton-beam range when using stopping-power ratios (SPRs) derived from single-energy CT scans. Several approaches have been proposed to estimate stopping power by using dual-energy CT (DECT) and have been shown through theoretical analysis to outperform single-energy CT (SECT) under the presence of tissue composition and density variations. Our lab previously proposed a joint statistical image reconstruction (JSIR) method built on a basis-vector model (BVM) tissue parameterization for SPR estimation, which was shown to perform comparatively better than other DECT image- and sinogram-domain decomposition approaches on simulated as well as experimental data. This comparison, however, assumed theoretical SPR values calculated from the samples’ known compositions and densities as ground truth and used the mean excitation energy and effective electron density from ICRU reports along with a simplified version of the Bethe-Bloch equation to determine SPR reference values. Furthermore, CT scans were acquired with an assumed ideal point source at a narrow beam collimation; thus, the signal formation assumed by our JSIR process neglected scatter and off-focal radiation. In this paper, we verify the accuracy of our method by comparing the SPR values derived from JSIR-BVM to direct measurements of relative SPR, as well as present a preliminary study on the impact of fan-beam scatter radiation on JSIR-BVM SPR prediction accuracy.

As medical imaging technologies continue to accelerate in complexity, application, and multiplicity of design choices and use features, they should ideally be evaluated and optimized through human clinical trials. However, such trials are often impossible and impractical. Virtual imaging trials (VITs) offer a solution allowing scientists to conduct “human” trials of emerging imaging concepts and technologies via quick and cost effective simulations. VIT gives researchers a new experimental paradigm to answer fundamental questions that can only be practically answered using the precise controls and known ground truth afforded in the virtual domain. VIT is possible through realistic and accurate modeling of a) the patient, b) the imaging system, and c) the image interpretation process. This paper discusses these topics and offers demonstration examples of select applications of VIT in CT imaging.

The purpose of this study was to develop, validate, and evaluate a method for measuring noise texture directly from patient CT images (i.e., “in vivo”). The method identifies target regions within patient scans which are least likely to have major contribution of patient anatomy, detrends these regions locally, and measures noise power spectrum (NPS) there using previously phantomvalidated techniques. Method development and validation used scanner-specific CT simulations of computational, anthropomorphic phantom (XCAT phantom at three phases of contrast enhancement) with known ground truth of the NPS. Simulations were based on a clinical scanner (Definition Flash, Siemens) and clinically relevant settings (tube voltage of 120kV at 3 dose levels). Images were reconstructed with filtered backprojection (kernel: B31, B41, B50) and Sinogram Affirmed Iterative Reconstruction (kernel: I31, I41, I50) algorithms using a manufacturer-specific reconstruction software (ReconCT, Siemens). All NPS measurements were made in the liver. Ground-truth NPS were taken as the sum of 1) a measurement in parenchymal regions of anatomy-subtracted (i.e. “noise only”) scans, and 2) a measurement in the same region of “noise-free” (pre-noise-insertion) images. To assess the performance of the in vivo NPS, the accuracy and bias of NPS average frequency (f_avg), and integrated noise magnitude were reported across the simulated scan population representing 2 reconstruction algorithms, 3 kernels, 3 dose levels, and 3 liver vasculature-to-parenchyma contrast levels. Pearson and Spearman correlation coefficient pairs were 0.97 and 0.93, and 1.0 and 0.99 for f_avg and noise magnitude, respectively. Finally, the NPS estimation method was further deployed on clinical cases to assess the feasibility of clinical analysis.

X-ray fluorescence tomography (XFCT) using nanoparticles (NPs) as contrast agents shows promise for high-spatial- resolution 3D molecular imaging. The technique has been demonstrated experimentally on phantoms as well as in small animals, i.e., mice. Parallel to experimental development, simulations play a key role in investigating the performance of existing and proposed XFCT arrangements. Up until recently, however, realistic simulations of small-animal XFCT have been unavailable due to the lack of appropriate, accurate and accelerated Monte Carlo (MC) based tools.

We have developed the tools necessary to simulate highly realistic small-animal XFCT. Our MC tool, XRFGPU is based on the accelerated open-source MC-GPU and allows simulations with speeds >1000x faster than other existing MC codes capable of similar simulations. Using available high-resolution digital mouse phantoms, we can predict the performance of any XFCT imaging arrangement and produce the realistic data needed for future development of small-animal XFCT. Here we validate the simulations against our laboratory arrangement, capable of 200 μm spatial-resolution XFCT of mice. Comparisons using a phantom confirms that the simulations are accurate. Lastly, we compare simulated images with images acquired in vivo using our laboratory arrangement showing how simulations can be used to enhance the interpretation of experimental data.

Training of deep learning-based segmentation algorithms requires large datasets of annotated images. Obtaining large clinical datasets, particularly for interventional imaging, is difficult and annotating vessels manually in the images is very time-consuming. Simulated images can be used to supplement clinical datasets. We developed a technique for the simulation of realistic, anatomically- and physiologically-motivated hepatic artery trees and subsequent synthesis of fluoroscopic images. The proposed approach creates a network of main feeding arteries for each of the eight liver segments as defined by the Couinaud classification system. A constrained constructive optimization based approach was then used to connect a set of randomly generated endpoints within each segment to the corresponding feeding artery. Vessel curvature was created using cubic splines and the generated vasculature was inserted into the digital XCAT phantom. The simulated 2D fluoroscopic images were generated using ray tracing and included focal spot blur, detector blur and Poisson noise. The length ratio (1.1 ± 1.7) and two parameters from Murray’s law, branching angles (7.9 ± 7.7^𝑜 mean absolute difference from Murray’s law) and radius ratio (1.0 ± 0.1) of the generated vasculature were in accordance with values reported in literature (1.3, Murray’s law applied to branching angles, and 1.0 respectively). Simulated vasculature included main branches for each of the eight Couinaud segments, where 87% of all connected endpoints terminated in the same segment. Simulated 2D projection images were analyzed using a vessel phantom study with contrast-enhanced tubes (0.305-3.353 mm diameter). The normalized root mean squared difference between the measured and simulated vessel profiles averaged 3.5%. In conclusion, the proposed method provides realistic simulated fluoroscopic images of the liver vasculature and could prove useful for the training of machine learning based algorithms for vessel segmentation.

Coronary artery disease (CAD) is one of the leading causes of death world-wide. Because of its great clinical importance, technological and diagnostic advances occur to combat it at a similarly high rate. With the number of clinical trials that would be necessary, this becomes infeasible and virtual clinical trials (VCT) are necessary. These require virtual patients and virtual pathologies. A generative adversarial network (GAN) was used to create a library of coronary plaques which were physiologically validated with finite element analysis. The resulting plaque library consists of a large number of realistic, variable virtual pathologies for use in VCTs.

In photon-counting CT, detector energy thresholds directly affect image quality attributes such as contrast and noise. The purpose of this study was to identify optimum energy thresholds using a comprehensive virtual clinical trial platform. The virtual trial was done using a computational, anthropomorphic phantom and a photon-counting CT simulator. The phantom (adult male, 50^th percentile body mass index) was chosen from the library of extended cardiac-torso (XCAT) phantoms. A vessel growth algorithm was used to model detailed hepatic arteries within the liver. Five computational lesions were inserted to the liver. The phantom was “imaged” with iodinated contrast at three contrast phases, i.e. arterial, portal venous, and delayed phases. All virtual sinograms were simulated using a validated CT simulator (DukeSim) modeling a prototype photon-counting scanner (CounT, Siemens Healthcare). The simulator included an energy- and threshold-dependent detector model accounting for X-ray and electronic effects and noise. At each contrast phase, the phantom was imaged at 35 detector energy threshold combinations with five lower-energy thresholds (20-40 keV) and seven upper-energy thresholds (50-80 keV). For each scan, two “threshold” images (photons detected beyond a threshold) and one “bin” image (photons detected between the two thresholds) were acquired. Noise magnitude, image contrast, and contrast-to-noise ratio (CNR) were measured in the aorta and the liver lesions. Optimum energy thresholds were identified as those yielding higher CNRs. From the simulations, noise magnitude was found to increase and CNR decrease with the increase in energy thresholds. Keeping the low threshold constant, the noise decreased and CNR increased with the increase in the upper energy threshold. Therefore, our results suggest that a combination of a lower threshold at 20-30 keV and upper threshold at 80 keV maximize the image quality. This study demonstrates an application of a virtual trial platform to quantify and optimize the quality of photon-counting CT.

Model-based material decomposition (MBMD) directly estimates the material densities from the spectral CT data and has found opportunities for dose reduction via physical and statistical modeling and advanced regularization. However, image properties of material basis volumes can be complex. For example, spatial resolution, noise, and cross-talk can depend on acquisition parameters, regularization, patient size, and anatomical target. In this work, we propose a set of prospective prediction tools for the generalized local impulse response (LIR) that characterizes both in-basis spatial resolution and cross-basis response, as well as noise correlation. The accuracy of noise predictor was validated in a simulation study, comparing predicted and measured in- and cross-basis noise correlations. Employing these predictors, we composed a specialized regularization for cross-talk reduction and showed that such prediction tools are promising for task-based optimization in spectral CT applications.

Cone-beam CT (CBCT) is widely used in diagnostic imaging and image-guided procedures, leading to an increasing need for advanced CBCT techniques, such as dual energy (DE) imaging. Previous studies have shown that DECBCT can perform quantitative material decomposition, including quantification of contrast agents, electron density, and virtual monoenergetic images. Currently, most CBCT systems perform DE imaging using a kVp switching technique. However, the disadvantages of this method are spatial and temporal misregistration as well as total scan time increase, leading to errors in the material decomposition. DE-CBCT with a dual layer flat panel detector potentially overcomes these limitations by acquiring the dual energy images simultaneously. In this work, we investigate the DE imaging performance of a prototype dual layer detector by evaluating its material decomposition capability and comparing its performance to that of the kVp switching method. Two sets of x-ray spectra were used for kVp switching: 80/120 kVp and 80/120 kVp + 1 mm Cu filtration. Our results show the dual layer detector outperforms kVp switching at 80/120 kVp with matched dose. The performance of kVp switching was better by adding 1 mm copper filtration to the high energy images (80/120 kVp + 1 mm Cu), though the dual layer detector still provided comparable performance for material decomposition tasks. Overall, both the dual layer detector and kVp switching methods provided quantitative material decomposition images in DE-CBCT, with the dual layer detector having additional potential advantages.

Photon-counting spectral computed tomography (PCCT) reconstructs multiple energy-channel images to describe the same object, where there exists a strong correlation among all channels. In addition, reconstruction of each energychannel image suffers photon count starving problem. To make full use of the correlation among different channels to suppress the data noise and enhance the tissue texture in reconstructing each energy-channel image, this paper proposed a tensor convolutional neural network (TCNN) architecture to learn a tissue-specific texture prior for PCCT reconstruction. Specifically, we first model the spatial texture prior information in each individual channel using a convolution neural network, and then extract the correlation information among different energy channels by merging the multi-channel networks. Finally, we integrate the TCNN as a prior into Bayesian reconstruction framework. To evaluate the tissue texture preserving performance of the proposed method for each channel, a vivid clinical phantom which can simulate the real tissue textures was employed. The improvement associated with TCNN is remarkable relative to simultaneous algebraic reconstruction technique (SART) and tensor dictionary learning (TDL) based reconstruction. The proposed method produced promising results in terms of not only preserving texture feature but also suppressing image noise in each channel. The proposed method outperforms the competing methods in both visual inspection and quantitative indexes of root mean square error (RMSE), peak signal to noise ratio (PSNR), structural similarity (SSIM) and feature similarity (FSIM).

CT imaging is one of the primary diagnostic tools utilized in modern radiology departments. Current stateof- the-art spectral CT imaging systems have been implemented using advanced x-ray source and/or detector technologies that have enabled image objects to be rapidly scanned using two distinct x-ray spectra (i.e., different effective beam energies). In this paper, we study the possibility to extract the encoded spectral information from the measured data when a single polychromatic x-ray spectrum is used to acquire data using an energy integration detector. Based upon our physical analysis, a physics-based deep neural network architecture, termed the Deep Spectral Imaging Network, was trained to demonstrate the feasibility of achieving spectral CT imaging using an energy integration detector and a single-kV acquisition.

Dual-energy CT (DECT) has become increasingly popular in practice due to its unique capability of material differentiation. One typical implementation of DECT is to use fast kV switching acquisition technique, which rapidly alternates the X-ray tube voltage between two predetermined kVs in a frequent manner. However, usage of such technique may be limited in practice, as it typically requires sophisticated hardware of high cost and lacks of dose efficiency due to difficulty in tube current modulation. One possible solution is to reduce the frequency of voltage switching during acquisition. However, this alternative approach may potentially compromise the image quality, as it results in sparse measurements for both kVs. In this paper, we proposed a cascaded deep-learning reconstruction framework for sparse-view kV-switching DECT, where two deep convolutional neural networks were employed in the reconstruction, one completing the missing views in the sinogram space and the other improving image quality in the image space. We demonstrated the feasibility of proposed method using sparse-view kV-switching data simulated from rotate-rotate DECT scans with phantom and clinical data. Experimental results show that the proposed method on sparse-view kV-switching data achieve comparable image quality and quantitative accuracy as compared to traditional method on fully-sampled rotate-rotate data

Purpose: Chest tomosynthesis is an attractive alternative to computed tomography (CT) for lung nodule screening, but reductions in image quality caused by radiation scatter remains an important limitation. Conventional anti-scatter grids result in higher patient dose, and alternative approaches are needed. The purpose of this study was to validate a lowdose patient-specific approach to scatter correction for an upcoming human imaging study.

Method: A primary sampling device (PSD) was designed and scatter correction algorithm incorporated into an experimental stationary digital chest tomosynthesis (s-DCT) system for this study to directly compute scatter from the primary beam information. Phantom and an in-vivo porcine subject were imaged. Total scan time was measured and image quality evaluated.

Results: Comparison of reconstruction slice images from uncorrected and scatter-corrected projection images reveals improved image quality, with increased feature conspicuity. Each scan in the current setup required twelve seconds, in addition to one second for PSD retraction, for a total scan time of 25 seconds.

Conclusions: We have evaluated the prototype low-dose, patient-specific scatter correction methodology using phantom studies in preparation for a clinical trial. Incorporating only 5% of additional patient dose, the reconstruction slices exhibit increased visual conspicuity of anatomical features, with the primary drawback of increased total scan time. Though used for tomosynthesis, the technique can be easily translated to digital radiography in lieu of an anti-scattering grid.

Metal artifact remains a challenge in cone-beam CT images. Many two-pass metal artifact reduction methods have been proposed, which work fairly well, but are limited when the metal is outside the scan field-of-view (FOV) or when the metal is moving during the scan. In the former, even reconstructing with a larger FOV does not guarantee a good estimate of metal location in the projections; and in the latter, the metal location in each projection is difficult to identify due to motion. Furthermore, two-pass methods increase the total reconstruction time. In this study, a projection-based metal detection and correction method with a dual layer detector is investigated. The dual layer detector provides dual energy images with perfect temporal and spatial registration in each projection, which aid in the identification of metal. A simple phantom with metal wires (copper) and a needle (steel) is used to evaluate the projection-based metal artifact reduction method from a dual layer scan and compared with that of a single layer scan. Preliminary results showed enhanced ability to identify metal regions, leading to substantially reduced metal artifact in reconstructed images. In summary, an effective single-pass, projection-domain method using a dual layer detector has been demonstrated, and it is expected to be robust against truncation and motion.

Purpose: Metal artifacts remain a challenge for CBCT systems in diagnostic imaging and image-guided surgery, obscuring visualization of metal instruments and surrounding anatomy. We present a method to predict C-arm CBCT orbits that will avoid metal artifacts by acquiring projection data that is least affected by polyenergetic bias. Methods: The metal artifact avoidance (MAA) method operates with a minimum of prior information, is compatible with simple mobile C-arms that are increasingly prevalent in routine use, and is consistent with either 3D filtered backprojection (FBP), more advanced (polyenergetic) model-based image reconstruction (MBIR), and/or metal artifact reduction (MAR) post-processing methods. MAA consists of the following steps: (i) coarse localization of metal objects in the field of view (FOV) via two or more low-dose scout views, coarse backprojection, and segmentation (e.g., with a U-Net); (ii) a simple model-based prediction of metal-induced x-ray spectral shift for all source-detector vertices (gantry rotation and tilt angles) accessible by the imaging system; and (iii) definition of a source-detector orbit that minimizes the view-to-view inconsistency in spectral shift. The method was evaluated in anthropomorphic phantom study emulating pedicle screw placement in spine surgery. Results: Phantom studies confirmed that the MAA method could accurately predict tilt angles that minimize metal artifacts. The proposed U-Net segmentation method was able to localize complex distributions of metal instrumentation (over 70% Dice coefficient) with 6 low-dose scout projections acquired during routine pre-scan collision check. CBCT images acquired at MAA-prescribed tilt angles demonstrated ~50% reduction in “blooming” artifacts (measured as FWHM of the screw shaft). Geometric calibration for tilted orbits at prescribed angular increments with interpolation for intermediate values demonstrated accuracy comparable to non-tilted circular trajectories in terms of the modulation transfer function. Conclusion: The preliminary results demonstrate the ability to predict C-arm orbits that provide projection data with minimal spectral bias from metal instrumentation. Such orbits exhibit strongly reduced metal artifacts, and the projection data are compatible with additional post-processing (metal artifact reduction, MAR) methods to further reduce artifacts and/or reduce noise. Ongoing studies aim to improve the robustness of metal object localization from scout views and investigate additional benefits of non-circular C-arm trajectories.

Metal artifacts are a well-known problem in computed tomography - particularly in interventional imaging where surgical tools and hardware are often found in the field-of-view. An increasing number of interventional imaging systems are capable of non-circular orbits providing one potential avenue to avoid metal artifacts entirely by careful design of the orbital trajectory. In this work, we propose a general design methodology to find complete data solution by applying Tuy’s condition for data completeness. That is, because metal implants effectively cause missing data in projections, we propose to find orbital designs that will not have missing data based on arbitrary placement of metal within the imaging field-of-view. We present the design process for these missing-data-free orbits and evaluate the orbital designs in simulation experiments. The resulting orbits are highly robust to metal objects and show greatly improved visualization of features that are ordinarily obscured.

Two-dimensional anti-scatter grids (2D-ASGs) have been developed to selectively capture scattered photons while preserving image signal in flat-panel cone-beam CT systems (CBCT). However, 2D-ASGs affect the response of detector elements underneath grid-septa, producing grid-line artifacts (GLA), which render traditional gain-and-offset corrections ineffective. GLA in the projection images lead to ring-artifacts in CBCT reconstructions, which undermine the improvements in image quality associated with 2D-ASGs. We propose a novel implementation of an exposuredependent gain-correction and notch-Fourier filtering of the projection data to minimize GLA-related ring-artifacts in CBCT. A pixel-by-pixel gain-factor was calculated by dividing the intensities of a flat-field image (with no-ASG) by the intensities of a flat-field image with added ASG, at different exposure levels. Exposure levels were modified using copper filtration of the x-ray beam at six-different thicknesses (0 to 2.5 mm, 0.5 mm increments). Exposure-dependent gain-factors were stored in a multidimensional array and pixel-by-pixel exposure response was characterized using nonlinear curve-fitting. The exposure-dependent gain-correction was applied to 215 projection images of a 14 cm water phantom using a cobalt-chrome 2D-ASG. Residual faint grid-lines were removed using a customized Fourier-notch filter prior to Parker-weighted FDK reconstruction. Traditional gain-and-offset correction produced severe ring-artifacts (i.e., σ = 833.64 HU) when compared to the exposure-dependent gain correction (i.e., σ = 76.16 HU). Additionally, Fouriernotch filtering improved CT number accuracy by 43 HU. Our results suggest that characterization of the exposuredependent response of GLA-affected pixels can minimize ring-artifacts and improve CT-number accuracy, thus eliminating some of the difficulties of 2D-ASG implementation in CBCT systems.

Compared to conventional X-ray transmission imaging techniques, imaging systems based on field-emissive cathode based conformal transmission imaging schemes have smaller volumes. This imaging system combines multiple point sources to complete very low dose imaging of the region of interest using a small cone beam, and finally obtain high quality images by image restoration methods. However, more precise system geometry is required by the system architecture of the new imaging scheme and the special image-forming method. If the geometric parameters that are inconsistent with the actual use are used in image restoration, it will cause the geometric distortion in the final restored image. In this work, a proposed geometric calibration applied to the new imaging scheme. First, we use a partial point source on the periphery of the array ray source to project the side ball phantom of the imaged object. Then, a cost function that associates the geometrical parameters with the degree to which the back-projections of the ball phantom in projections from different point sources is constructed. Finally, the particle swarm optimization (PSO) is used for minimizing the cost function, and the resultant value of the cost function is the actual geometric parameter. With this calibration method, transmission imaging and geometric calibration will be done simultaneously. Besides, this calibration method does not depend on a complex calibration phantom and is simple and effective. In addition, the PSO can accelerate the implementation of the algorithm. The effectiveness and accuracy of this calibration method are verified by two sets of simulation experiments.

To suppress artifacts in X-ray digital tomosynthesis, a method which combines 3D and 2D region growing was proposed. However, it could not extract small metals such as pins, due to the discontinuity of the metal in the 3D projection data. A novel method is proposed to separate the metal from the projection data by combining pattern matching between neighboring projection angles and interpolating image pixels of enlarged projection data for each projection angle. The proposed method was compared with the conventional method based on 3D region growing. In a phantom experiment, the proposed method reduced the artifacts close to the metals. At a distance within 1 mm from the metal object, the mean absolute error evaluated without metal artifact correction (No MAC), with the conventional and proposed methods were 461.2, 461.2, and 100.2, respectively, and an improvement of more than 78% was demonstrated. When the distance from the object was shorter, the artifact became more significant in the No MAC and conventional methods, so the effectiveness of the proposed method was higher. Applying a suitable artifact correction on the basis of metal extraction makes it possible to effectively reduce artifacts in DT images.

Digital tomosynthesis (DTS) has been used in diagnosis and radiation therapy due to its performance and benefits. DTS is able to provide 3D images with good depth resolution comparing to conventional radiography and reduce radiation dose comparing to computed tomography (CT). However, DTS scans with limited scan angles and a few projections lead to the insufficiency in data acquisition. Such a drawback causes the alteration of image characteristics and the distortion of image quality in DTS imaging. These issues can be magnified by the geometric variations of the DTS systems and the imaging strategies. In this study, the effect of geometric variations on image characteristics was evaluated by the DTS systems simulated with various X-ray source scan trajectories and angles. The DTS images were analyzed in terms of noise property, contrast-to-noise ratio (CNR), and spatial resolution. The results showed that the quality of DTS images was highly dependent on X-ray source scan trajectories and angles, and the characteristics of DTS images varied according to their acquisition geometries. Therefore, the geometries and strategies for DTS imaging should be appropriately determined for optimizing their systems and applications.

Digital chest tomosynthesis that provides a reconstructed 3D chest image is a superior technique to detect chest diseases. As it is difficult to detect diseases like lung cancer with conventional 2D digital chest X-ray technology (CXT), digital chest tomosynthesis improves upon the many of the limitations inherent in the 2D digital CXT. In this study, we report a digital chest tomosynthesis system (D-CTS) that can generate multi X-ray information for the reconstruction of a 3D Xray chest image. The D-CTS reported herein employs an array of carbon nanotube (CNT) emitter-based cold cathode electron-guns that are triggered in sequence to provide a gantry-less system (Figure 1). The CNTs are achieved by direct growth on a metal substrate and have a spaghetti-like structure (Figure 2) with fast response to electrical bias under vacuum conditions. Unlike conventional rotating type systems with gantries, our CTS has the advantage of less motion blur in image acquisition, given its stationary position. Additionally, the switching from one electron-gun (e-gun) to the next is much faster than the speed of conventional gantries, allowing faster acquisition time t required for digital operation. This system shows outstanding field emission property for taking X-ray images. The design, fabrication process and imaging processing of the multi-beam CNT X-ray system will be discussed during the presentation.

Fluorescence molecular imaging is a promising tool for molecular tracking, and thus can visualize the vascular structure of small animal. However, with the strong scattering in biological tissues, traditional fluorescence molecular imaging in the visible spectrum or the first near-infrared spectrum (NIR-I, 700-900 nm) has a limitation for high-resolution vascular imaging. Recently, the novel in vivo fluorescence molecular imaging in the longer second near-infrared window (NIR-IIb, 1500-1700 nm) is successfully developed for small animal imaging. The NIR-IIb window affords high imaging resolution and deep tissue penetration because of the diminished photon scattering effect, which is suitable for vascular imaging. However, the clinical applications of NIR-IIb fluorescence molecular imaging have been severely limited for lack of the clinical fluorophores. Here, we show that the clinically available dye, indocyanine green (ICG), can also emit the NIR-IIb signal, which provides high-resolution vascular imaging of small animal. We construct a novel imaging system for NIR-I and NIR-IIb imaging simultaneously and perform two vascular imaging experiments. The results demonstrate that the NIR-IIb imaging using ICG shows great superiority for high-resolution vascular imaging of small animal compared with NIR-I imaging. It is believed that this study will facilitate the preclinical and clinical applications of NIR-IIb molecular imaging using ICG in the future.

Stress fracture is a common condition in athletes and military personnel yet objective methods for detection and tracking of stress fractures (e.g., MR and scintigraphy) remain complex and expensive. Methods involving ultrasound are in common use; however, they are thus far subjective (relying on patient report of pain). The goal of this work is to develop quantitative and sensitive ultrasonic evaluation methods that could be deployed broadly in smaller clinical setting for use by a wide range of medical professionals. Such methods would complement beam-formed images of the bone surface by adding information about even sub-wavelength features. We have demonstrated that small imperfections on the bone surface significantly alter the characteristics of the reflectivity as made evident in the relative magnitude of the specular vs. diffuse scatter. In this paper we will present our work toward developing methods for acquiring, enhancing, visualizing, and evaluating such effects. We will present results from measurements with ex vivo bone samples and phantoms, and discuss the ultimate applicability of our methods to in vivo diagnosis.

Bioluminescence tomography (BLT) is a promising molecular imaging tool in monitoring non-invasively physiological and pathological processes in vivo at the cellular and molecular levels. And the radiative transfer equation (RTE) has successfully been used as a standard model for describing the propagation of visible and near infrared photons trough biological tissues. However in practical application, implementation of RTE is extremely complicated for complex biological tissue. And several approximations of the RTE were applied to model the light transport in a turbid medium, such as the diffusion equation (DE) and the simplified spherical harmonic approximation equation (SP_N). However, DE provides a high computational efficiency and is only valid in the high scattering region, while SP_N has a large demand for memory space, which makes it difficult for SP_N to be used on the fine mesh and limits its application in practice. In this paper, we provided a new finite element mesh regrouping-based hybrid light transport model in BLT. Based on the optical property of biological tissue, the finite element mesh were grouped into high-scattering and low-scattering regions. And based on the theory of light transport, hybrid third-order simplified spherical harmonic approximate–diffusion equation model (HSDM) was used to forward light transport model. In numerical simulation experiments, accuracy and efficiency of our proposed method were evaluated. Results showed that the hybrid light transport model achieved a better balance between accuracy and efficiency compared with the DE and the SP₃ models. And it was best suited as a light transport model for Bioluminescence Tomography.

Contrast sensitivity decreases in keratoconus subjects even before the decrease is presented at high contrast visual acuity. The aim of our study was to estimate the subject’s contrast sensitivity and visual acuity in different keratoconus stages. In our study, there were included 45 keratoconus subjects (77 eyes) with the first, second and third keratoconus stages. Visual acuity and contrast sensitivity were measured at 3 m with the best possible spectacle correction and without vision correction using FrACT program 3.9.3. In the study, contrast sensitivity was measured at following spatial frequencies – 1, 3, 5, 7, 9, 11, 13 and 15 cpd. Contrast sensitivity analyse showed that there was a statistically significant difference among the contrast sensitivity of the first, second and third keratoconus stages in all spatial frequencies. If the keratoconus stage increased, the contrast sensitivity decreased more. The median visual acuity of first stage keratoconus subjects with spectacle correction was 0.56 (0.36, 0.65) decimal units, second stage keratoconus – 0.38 (0.21; 0.47) and third stage – 0.13 (0.08, 0.22).

Retrospective 4D image construction from continuously acquired 2D slices is a necessary step to achieve high-quality 4D images. Self-gating methods, which extract breathing signals only from image information without any external gating technology, have much potential, such as in pediatric patients with thoracic insufficiency syndrome (TIS) who suffer from extreme malformations of the chest wall, diaphragm, and spine, leading to breathing that is very complex with lots of abnormal respiration cycles, including very deep or shallow cycles. Existing methods do not work well in this clinical scenario and most are not fully automatic, requiring some manual interactive operations. In this paper, we propose a fully automatic 4D dMRI construction method based on the concept of flux to address the 4D image construction from 2D slices of subjects with complex respiration. Firstly, we extract the breathing signal for each location based on the flux of the optical flow vector field of the body region from the image series. Then, we give a full analysis for all cycles and extract several normal ones and map them to one cosine respiration model for each location. After that, we re-sample one normal cycle from the respiration model for each location independently. All of these resampled normal cycles form the final constructed 4D image. Qualitative and quantitative evaluations on 25 subjects show that the proposed method can handle datasets from subjects with more complex respiration and achieves good self consistency results while maintaining time and space continuity.

The size and shape of an x-ray source’s focal spot is a critical factor in the imaging system’s overall spatial resolution. The conventional approach to imaging the focal spot uses a pinhole camera, but this requires careful, manual measurements. Instead, we propose a novel alternative, simply using the collimator available on many x-ray systems. After placing the edge of a collimator blade in the center of the beam, we can obtain an image of its edge spread function (ESF). Each ESF provides information about the focal spot distribution – specifically, the parallel projection of the focal spot in the direction parallel to the edge. If the edge is then rotated about the beam axis, each image provides a different parallel projection of the focal spot until a complete Radon transform of the focal spot distribution is obtained. The focal spot can then be reconstructed by the inverse Radon transform, or parallel-beam filtered backprojection. We conducted a study on a clinical C-arm system with 3 focal spot sizes (0.3, 0.6, 1.0 mm nominal size), comparing the focal spot obtained using the rotating edge method against the conventional pinhole approach. Our results demonstrate accurate characterization of the size and shape of the focal spot.

Computed tomography (CT) is a medical imaging technique which is widely used in various clinical applications. Since the applied radiation is potentially harmful to humans, the applied doses are carefully monitored during CT image acquisition. On the contrary, the CT image quality is not analyzed or monitored during normal clinical operation. A wide range of parameters such as patient anatomy, implants, cardiac or respiratory motion, pathological tissue variations or an inappropriately configured CT system can affect the image quality. Hence, an automated image quality assessment would be desirable to track the CT image quality in daily clinical routine. In this work, the image noise and spatial resolution were characterized and analyzed in order to determine the image quality. While the image noise was characterized by the noise power spectrum (NPS), the spatial resolution was described by the modulation transfer function (MTF). Both methods were applied to human torso phantom images or real thorax CT images. It was shown that both parameters are useful to distinguish between different image qualities. Lower CT image quality was correlated with higher noise amplitudes (measured by the NPS) and a lower spatial resolution (measured by the MTF). Challenges for future works remain in generating appropriately annotated clinical CT image databases as well as in automatically selecting suitable regions-of-interest within the image.

The quantification of radiomics features is impacted by a complex interplay of image quality and segmentation performance. To understand and optimize the process, it is highly beneficial to isolate the impact of each factor quantitatively. The purpose of this study was to develop a segmentation performance index (𝑠′) to describe the segmentation performance relative to an ideal estimation, and further apply the index across several segmentation algorithms and image quality conditions in the task of characterizing lesion morphology. First, an estimator was developed with prior knowledge of the lesion and the imaging resolution properties to represent the best that an ideal segmentation operator would be expected to yield given the information content available from the image. Using three anthropomorphic lesion models of a mean contrast of 600 HU and mean radius of 5 mm, the method was applied to three segmentation operators based on (1) active contour, (2) thresholding, and (3) a commercial vendor offering (Siemens Healthineers). Images of the lesions were simulated in a uniform background for a Siemens Flash filtered back-projection (FBP) B31f reconstruction with noise magnitude ranging from 10 HU to 230 HU. Images were simulated 50 times at each noise magnitude to assess test-retest variability. The segmentation masks were used to calculate a root mean square error (RMSE) across test-retest images for morphology and size radiomics features. The ideal estimation RMSE was calculated for each feature and was divided by the segmentation RMSE resulting in the segmentation performance index (𝑠′). The 𝑠′ values were evaluated as a function of increasing noise. Traditional segmentation quality metrics including dice coefficient, volume similarity, and Hausdorff distance were also calculated for each noise magnitude to compare against the 𝑠′ index. Traditional segmentation quality metrics are not reflective of radiomics performance. The 𝑠′ values, in contrast, were found to be both feature-specific and noisespecific, useful in characterizing the performance of segmentation algorithms in measuring radiomics features. This allows for the choice of segmentation algorithms to yield optimum radiomic performance for a given imaging condition and or protocol.

X-ray spectrum plays an essential role in CT applications. Since it is difficult to measure x-ray spectrum directly in practice, x-ray spectrum is always indirectly obtained by using transmission measurements through a calibration phantom of known thickness and materials. These methods are independent of CT image reconstruction and bring extra cost. In this study, we propose a parametric physical model based spectrum estimation algorithm for CT. First, a physical model contains six parameters is proposed to represent the x-ray spectrum, which is derived from the x-ray imaging physics. Second, a template image contains different material components can be obtained by segmenting CT reconstructed images with a simple method. And the estimated projection can be calculated by reprojecting the template image with the proposed spectrum model. Finally, the six parameters expressing the spectrum can be solved by minimizing the error between the estimated projection and real measurements. The effectiveness of the proposed method has been validated on the simulated data. Experimental results demonstrate the proposed method can estimate the accurate spectra at low and high energies and provide a good reconstruction of characteristic radiations.

Imaging phantoms are test patterns used to measure image quality in computer tomography (CT) systems. A new phantom platform (Mercury Phantom, Gammex) provides test patterns for estimating the task transfer function (TTF) or noise power spectrum (NPF) and simulates different patient sizes. Determining which image slices are suitable for analysis currently requires manual annotation of these patterns by an expert, as subtle defects may make an image unsuitable for measurement. We propose a method of automatically classifying these test patterns in a series of phantom images using deep learning techniques. By adapting a convolutional neural network based on the VGG19 architecture with weights trained on ImageNet, we use transfer learning to produce a classifier for this domain. The classifier is trained and evaluated with over 3,500 phantom images acquired at a university medical center. Input channels for color images are successfully adapted to convey contextual information for phantom images. A series of ablation studies are employed to verify design aspects of the classifier and evaluate its performance under varying training conditions. Our solution makes extensive use of image augmentation to produce a classifier that accurately classifies typical phantom images with 98% accuracy, while maintaining as much as 86% accuracy when the phantom is improperly imaged.

Purpose. Explore the potential value of displaying information collected by stationary intraoral tomosynthesis (sIOT) as multi-view synthetic radiographs, using vertical root fractures (VRFs) as a model system.

Methods. Filled and unfilled extracted tooth roots containing artificially-induced VRFs were imaged by sIOT and standard periapical radiography. sIOT collected 7 views across a 12° angle span, providing information for an image processing chain that included reconstruction, weighting, and forward projection to generate a set a synthetic two-dimensional (2D) images. Qualitative assessments of fracture conspicuity were used for comparison.

Results. The conspicuity of VRFs changed significantly with the angle of imaging, suggesting benefit to displaying a set of synthetic images across a span of viewing angles. Although high-density in-plane and out-of-plane artifacts, which could limit the conspicuity of VRFs, were prominent in the three-dimensional (3D) stack of reconstructed image slices, these artifacts were minimal in the synthetic radiographs. As such, some fractures were displayed more clearly in the synthetic 2D images compared to the reconstructed 3D image stack. Also, in some cases, the fractures were more conspicuous in the sIOT-generated synthetic images than the standard periapical radiographs.

Conclusion. Multi-view synthetic radiography can improve the display of VRFs in images generated by sIOT. As such, this approach to dental imaging may offer a useful clinical tool, with potential application to a host of imaging tasks.

Optical flow-based methods are commonly used to detect and correct patient motion in modalities such as cone-beam computed tomography. With such methods, the rotational motion deriving from the acquisition trajectory itself is obscuring the patient motion and therefore considered a perturbation. In this work, the question commonly posed in motion estimation is reversed. Instead of considering the rotational motion as obscuring the patient motion, it can be used to derive shape information about the patient. This is done by computing the optical flow from projection images in order to find point correspondences in different frames of the acquisition. Finally, projective geometry is used to localize a given pair of corresponding 2D image points in 3D object space.

The advantage of this method is that it allows for the localization of structures which are not contained within the scan field of view of truncated acquisitions but within an extended projection field of view. Therefore, it is shown that for certain high contrast structures, e.g. the skull, this localization method can be used to estimate the maximum extent of a patient from truncated projection data, which is an important information for extrapolation methods, or to localize highly absorbing structures outside of the scan field of view which contribute to the severity of the typically observed truncation artifacts.

Purpose: The aim of this study was to implement, validate and apply a tool for the assessment of the MTF and NPS of dental CBCT scanners using up-to-date measurement methods. Methods: The MTF was calculated using the circular edge method, and the NPS as the 2D Fourier transform of homogeneous background ROIs. The validation framework has been set-up as follows: images with known MTF/NPS are simulated, and from these images the MTF/NPS was calculated and the results compared. The influence of the phantom dimensions and the CNR on the MTF/NPS accuracy has been studied for a 10x10 cm² system FOV and corresponding voxel size of 0.25 mm³. For the practical measurements, two specific phantoms have been designed, based on the simulated results. Results: For the MTF the minimum number of slices needed to achieve an error of at most 5 % has been studied as a function of the CNR of the phantom rod insert (26 mm diameter). For the NPS, the number of 128x128 ROIs needed to achieve an error of at most 5 % have been determined. Measurement results on two different dental CBCT scanners for a selection of imaging conditions are presented. Conclusion: The simulation results provide guidelines for proper use of the developed tool to obtain accurate MTF and NPS measurements in practice. From the actual measurements it can be concluded that sometimes more information about system design parameters from the manufacturer is needed in order to explain the results with absolute certainty.

Three dimensional CT images have enabled dentists to visualize the overall anatomic structure of teeth and jawbones. Dental cone-beam CT(CBCT) images include a lot more information as compared to panoramic radiographs, but it is hard to assess overall teeth and jaw structures at a single glance. Although panoramic images facilitate the evaluation process of overall anatomic structures, they may include geometric distortions, blurs, and superimposition of multiple structures due to the spine and ghost effects. Some image viewers have a cut-viewing function that enables orthogonal viewing along the user-set dental arch. However, the orthogonal viewing looks different from the panoramic image physically taken by a panoramic scanner. To make more convenient use of CBCT images, we have developed an approach to synthesize panoramic images from the CBCT images. In the synthesis of the panoramic images, we removed the ghost and spine superimposition to improve the visibility of the synthesized panoramic images.

The panoramic image synthesis has been done in three steps. At first, we extracted the panoramic dental arch from the CBCT images. At the next step, we removed the ghost-inducing bone parts from the CBCT images. Lastly, we synthesized panoramic images by stitching thousands of partial-view panoramic projection images.

With the advantage of the synthesized panoramic image, the additional panoramic scan is not further necessary, and undesirable features of conventional panoramic radiographs (e.g., ghost artifacts or cervical spine superimposition) can be entirely removed.

The properties of carbon nanotubes (CNTs) show a variety of application possibilities in the next generation industries. Among them, application to a field emission device is expected due to a high emission current at a low critical electric field. Compared with conventional thermionic emission X-ray sources, field emission X-ray sources have many advantages. Their electric field can be turned on and off quickly, minimizing radiation exposure times. Additionally, the size of the imaging system can be reduced because no cooling device is required. Using these advantages, we created a micro computed tomography (CT) system based on a CNT field emission X-ray source. The main advantages of the micro- CT lie in the high-spatial resolution, sensitivity to skeletons and lungs, and low cost. We optimized the field emission characteristics of the CNTs and the system geometry for high resolution.

While two-dimensional antiscatter grids (2D grid) reduce scatter intensity substantially in Cone Beam Computed Tomography (CBCT), a small fraction of scattered radiation is transmitted through the 2D grid to the detector. Residual scatter limits the accuracy of CT numbers and interferes with the correction of grid’s septal shadows, or footprint, in projections. If grid’s septal shadows are not adequately suppressed in projections, it will lead to ring artifacts in CBCT images. In this work, we present a new method to correct residual scatter transmitted through the grid by employing the 2D grid itself as a residual scatter measurement device.

Our method, referred as grid-based scatter sampling (GSS), exploits the spatial modulation of primary x-ray fluence by 2D grid’s septal shadows. The shape of the signal modulation pattern varies as a function of residual scatter intensity registered by detector pixels. Such a variation in signal pattern was employed to measure residual scatter intensity in each projection, and subsequently, residual scatter was subtracted from each projection.

To validate the GSS method, CBCT imaging experiments were conducted using a 2D antiscatter grid prototype in a linac mounted CBCT system. The effect of GSS method on the ring artifact reduction was quantified by measuring noise in CBCT images. In addition, the nonuniformity of Hounsfield Units (HU) and HU accuracy was measured in both head and pelvis-sized phantoms.

In qualitative evaluations, GSS method successfully reduced ring artifacts caused by 2D grid’s footprint. Image noise reduced by 23% due to reduction of ring artifacts. HU nonuniformity in water-equivalent sections was reduced from 20 HU to 10 HU, and streak artifacts between high density inserts were reduced. The phantom size dependent variations in HU was also reduced after application of GSS method. Without GSS method, HU of density inserts reduced by 9% on the average when phantom size was increased from head to pelvis. With GSS method, HU values reduced only by 5% under the same conditions.

In summary, GSS method complements the 2D grid’s scatter suppression performance, by correcting the scatter transmitted through the grid. This approach does not require additional scatter-measurement hardware, such as beam-stop arrays, since the grid itself is employed as the scatter measurement device. By suppressing residual scatter in projections, our proposed method successfully reduced artifacts caused by 2D grid’s footprint, and further improved CT number accuracy.

In CT system, the distance between the scanned object and the detector gives a significant impact on the total scatter intensity arriving at the detector. As a result, if an object is scanned off-center, its relative distance to the detector is changing when CT system is rotating, and the detector received scatter intensity will also be changing throughout the scan. In general only after the location information is obtained from the entire 360 degree raw data, the precise scatter radiation can be predicted. This could slow down reconstruction performance a lot. In this paper, we would like to propose a purely projection based scatter correction algorithm, which can be processed independently inside each projection. The promising results of the proposed algorithm suggests the variance of scatter received at different locations could be well predicted, meanwhile the reconstruction speed could be well maintained.

Physical phantoms with realistic anatomical texture and composition (including contrast media) are of high value and relevance in evaluating the performance of clinical computed tomography (CT) imaging systems. They can offer assessments of image quality in a manner that is more relevant than existing phantoms. The goal of this project was to fabricate anatomically structured 3-D liver texture with added contrast-enhancement. The study used an inkjet printer and customized ink. Patient-informed liver texture with contrast details was first defined computationally and the needed material for each tissue type was estimated for each voxel to produce attenuations from -200 to +300 HU, with subtle 5-20 HU lesions. Volumetric data were printed, one sheet of paper at a time, using “soft tissue” ink based on 200mg/cc sodium bromide (NaBr), and “enhanced tissue” ink based on Isovue-300. The phantom was scanned on a commercial CT system (Siemens SOMATOM Definition Flash) using a clinical protocol. The results show targeted heterogeneity and enhancement as designed with subtle artifacts associated with uneven air contamination. In ongoing work, we plan to use this method to create anatomically and morphologically accurate liver and lung lesions with texture for use in dual energy CT.

We aimed to investigate the feasibility of predicting pleural invasion or adhesion of lung cancers with dynamic chest radiography (DCR), using a four-dimensional (4D) extended cardiac-torso (XCAT) computational phantom. An XCAT phantom of an adult man (50^th percentile in height and weight) with forced breathing and normal heart rate was generated. To simulate lung cancers with and without pleural invasion, 30-mm diameter tumor spheres were inserted into the right lower lung lobe of the virtual patients. Subsequently, the virtual patient was imaged using an X-ray simulator in posteroanterior and oblique directions, and bone suppression (BS) images were then created. The measurement points (tumor, rib, and diaphragm) were automatically tracked on projection images by template matching. We calculated five quantitative parameters related to the movement distance and directions of the targeted tumor and evaluated the ability of the DCR parameters to distinguish between patients with and without pleural invasion. Precise tracking of the targeted tumor was achieved on the BS images without any interruption by the rib shadows. The movement distance was an effective parameter to evaluate tumor invasion; however, with regard to the other parameters, similar results were obtained between the lung cancers with and without pleural invasion due to the lack of three-dimensional information on the projection images. The oblique views were useful for evaluation of the space between the chest wall and the moving tumor. DCR could help distinguish between patients with and without pleural invasion based on the two-dimensional movement distance in both oblique and posteroanterior projection views.

Dynamic Computed Tomography is a promising tool to investigate affections, predominantly of vascular and perivascular origin. Advances in CT technology are opening up the possibility to also evaluate musculoskeletal diseases using dynamic CT. Wide beam CT scanners have the potential to acquire images from the same anatomic area over lengths up to 16 cm. As such, detailed analysis of joint morphology and dynamic phenomena can be investigated. Based on image processing techniques, kinematics can then be obtained from these images. Motion artifacts are however a major concern in dynamic CT for musculoskeletal applications. Such artifacts are influenced by the acquisition protocol and the speed of motion. In this study, we designed a phantom that could rotate in the gantry of the CT in an attempt to quantify motion artifacts. Dynamic scans were obtained while the phantom was rotating at 720/s in the wide beam CT scanner. Visual inspection revealed two main types of artifacts: high intensity artifacts and blurring or distortion of the holes in the phantom. Two quantitative metrics were defined to assess the degree of motion artifacts induced by the rotation of the phantom. These metrics were used to compare two different dynamic CT acquisition protocols; cine and cardiac. Results showed that the cardiac protocol significantly preserved the original shape of the holes by as much as 75%±8 while that of cine was 62% ± 4. However, the cardiac protocol showed 2 times more pixels with high intensity spots than the cine protocol and that difference was significant.

High-contrast Computed Tomography (CT) imaging provides a non-invasive, detailed look at internal anatomy (soft tissues, bone, and cartilage) and is used throughout the clinical, research, and academic anatomical fields. Cadaveric specimens are valuable teaching and research tools, but contrast enhancement (CE) of cadavers for clear CT imaging is challenging. The lack of blood flow and pumping mechanisms produces clotting and a barrier to injection of contrast agents. Recently, cadaver preparation techniques for introducing contrast agents have been developed; however, evaluation of cadaveric contrast agents has been qualitative in nature and does not give enough information to adequately assess its efficacy. In this study, we used both quantitative and qualitative techniques to evaluate a cadaveric contrast agent (BriteVu) and its ability to resolve kidney boundaries, structure, and surrounding tissues. We found that BriteVu was able to enhance the macrostructures of the kidneys, and some substructures could be differentiated but not adequately enough for intricate vasculature modeling. Our approach to evaluation of cadaveric contrast agents provides a structural, quantitative understanding that frames cadaveric contrast agents as a tool for structural learning and research.

Early diagnosis of skin cancer offers the patient more favorable treatment options. Color fidelity of skin images is a major concern for dermatologists as adoption of digital dermatoscopes is increasing rapidly. Accurate color depiction of the lesion and surrounding skin are vital in diagnostic evaluation of a lesion.

We previously introduced VCT-Derma, a pipeline for dermatological Virtual Clinical Trials (VCTs) including detailed and flexible models of human skin and lesions, which represent the patient in the entire dermatoscopy-based diagnostic process. However, those initial models of skin and lesions did not properly account for tissue colors.

Our new skin model accounts for tissue color appearance by incorporating chromophores (e.g., melanin, blood) into the tissue model, and simulating the optical properties of the various skin layers. The physical properties of the skin and lesion were selected from clinically plausible values. The model and simulated dermatoscope images were created in open modelling software, assuming a linear camera model. We have assumed ambient white lighting, with a 6mm distance to the camera.

Our model of color appearance was characterised by comparing the brightness of the lesion to its depth. The brightness of the lesion is compared through the variability of the mean gray values of a cropped region around the lesion. We compare two skin models, one without extensive chromophore content and one with. Our preliminary evaluation of increasing chromophore content shows promise based on the results presented here. Further refinement and validation of the model is ongoing.

The brain is an organ of interest in high-consequence infectious disease. Subtle blood-brain-barrier disruption and edema can be identified and quantified using T₂ relaxation time as a biomarker. Accuracy is required to correlate changes in imaging findings with biological processes and therapeutic effectiveness. Dual echo time Fluid-Attenuated Inversion Recovery images, which minimizes the partial volume effects from cerebrospinal fluid, can be used to quantify T₂ relaxation. This sequence also meets the acquisition time requirements of biosafety level-4 facilities, where cohorts of animals are imaged on a tight schedule until terminal stages, when the animals may not survive long scans. However, the echo time used to compute T₂ varies among slices due to k-space mapping in fast spin-echo sequence. Therefore, an effective echo time instead of the prescribed echo time should be considered. We hypothesize that accurate effective echo time can be estimated by optimization for each slice ordering. A longer and more accurate T₂-weighted spin-echo (multiecho) sequence to compute T₂ maps from mono-exponential fitting was validated using gel phantoms and used as ground truth. A brain of a non-human primate was imaged in vivo using both T₂ sequences. T₂ maps were generated using both methods and a correction factor was computed by linear fitting for each set of echo times and slice ordering. The coefficient of variation within phantoms and the Pearson’s correlation coefficient between in vivo ground truth T₂ and each map were used to assess its accuracy and quality.

In an acute stroke parts of the human brain undergo subtle physiological changes, which are often visible as hypo- or hyperdense regions in Computed Tomography (CT) images. In the case of ischemic stroke usually an edema develops due to undersupplied cells forming regions of a core (dead tissue) and a penumbra (salvable tissue). For stroke diagnosis and outcome control it is very important to know the location and size of these different kinds of damaged tissue.

We have modelled the changes in elemental composition of brain tissue in different phases of an ischemic stroke. Influence of a number of factors on the absolute Hounsfield units is investigated as possible causes of intra- and interpatient variation. The modeled pathological changes are included in different software brain models. Subsequently we have simulated X-ray images of these brain models acquired by dual energy Cone Beam Computed Tomography (CBCT). Our modelling is based on a combination of analytical and Monte-Carlo methods. As an example of spectral processing virtual monoenergetic images are reconstructed from the simulated projections.

Simulated images are intended to optimize acquisition parameters for clinical studies beforehand and to develop new image processing algorithms to enhance the diagnostic value. As example a water map is calculated to better visualize the formation of an edema after ischemic stroke.

Angiographic parametric imaging (API) is a quantitative imaging method which uses digital subtraction angiography (DSA) to calculate biomarkers related to hemodynamics. This method has been used for neurovascular disease diagnosis and treatment outcome predictions in clinical settings but results are regarded with caution since derived biomarkers are strongly correlated with contrast injection parameters. This study aimed to assess utilization of truncated singular value decomposition (TSVD) in correcting API maps across various injection rates. Digital angiography data was collected using two neurovascular phantoms embedded in a simulated flow loop. Contrast volumes of 5 and 10 mL along with injection rates of 5, 10, 15, and 20 mL/sec were utilized during testing. API maps were generated with baseline and stenosis models using gamma variate fitting along with TSVD of the arterial input of the phantom. Surrogate regional blood flow (sRBF) and regional blood volume (sRBV) maps indicate consistent values across varying injection rates along with decreases in flow and volumes following introduction of a stenosis (Baseline: sRBF=53.3±4.20 arbitrary volume units (AVU)/min, sRBV=2.66±0.14 AVU, Stenosis: sRBF=28.6±3.78 AVU/min, sRBV=1.75±0.45 AVU). Mean transit time (MTT) and time of maximal residue function (Tmax) maps indicate consistent and decreasing parameter values respectively as injection rates increase along with increases in each parameter in the presence of a stenosis (Baseline: MTT=0.72±0.14 sec, Tmax=1.36±0.14 sec, Stenosis: MTT=0.94±0.13 sec, Tmax=1.71±0.18 sec). This study indicates TSVD has the potential to normalize API parameter maps across various injection rates potentially allowing for the implementation of API in ischemic stroke diagnosis and treatment.

Cerebral CT perfusion (CTP) imaging is widely used to guide the selection of acute ischemic stroke patients who are most likely to benefit from endovascular reperfusion therapy, as parametric CTP maps can assist physicians to identify the existence of hypoperfused but salvageable tissue upon reperfusion, and to differentiate this tissue from irreversibly damaged tissue for which reperfusion is futile. CTP is usually performed together with a noncontrast head CT (NCCT) for ruling out intracranial hemorrhage. Even for post stroke treatment follow-up and other neurological diseases, a CTP is often prescribed together with a NCCT to provide more comprehensive and mutually complementary diagnostic information. Despite these clinical utilities, the current CTP imaging technology still faces a major challenge: the poor contrast-to-noise ratio (CNR) of parametric CTP maps that hinders reliable stroke diagnosis for individual patients. Previous works have demonstrated that CNRs of some CTP maps such as cerebral blood volume (CBV) are predominately determined by the baseline CT images, and thus reducing the baseline image noise is very effective in improving the CNRs of CBV maps. The purpose of this work is to present a method that utilizes the high quality NCCT image to reduce the noise of baseline images and improve the detectability of ischemic core in CBV maps. The proposed method was evaluated using an in vivo canine subject where by utilizing the NCCT image the CNR was improved by 55%.

Purpose: Today’s state-of-the-art CT systems rely on a rotating gantry to acquire projections spanning up to 360 degrees around the head and/or body. By replacing the rotating source and detector with a stationary array of x-ray sources and line detectors, a head CT scanner could be potentially constructed with a small footprint and fast scanning speed. The purpose of this project is to design and construct a stationary head CT (s-HCT) scanner capable of diagnosis of stroke and head trauma patients in limited resource areas such as forward operating bases. Here we present preliminary imaging results which demonstrate the feasibility of such a system using carbon nanotube (CNT) x-ray source arrays.

Methods: The feasibility study was performed using a benchtop setup consisting of an x-ray source array with 45 distributed focal spots, each operating at 120kVp, and an Electronic Control System (ECS) for high speed control of the x-ray output from individual focal spots. The projection data was collected by an array of detectors configured specifically for head imaging. The basic performance of the CNT x-ray source array was characterized. By rotating the object in discrete angular steps, a potential s-HCT configuration was emulated. The collected projection images were reconstructed using an iterative reconstruction algorithm developed specifically for this configuration. Evaluation of the image quality was completed by comparing this image of the ACR CT phantom obtained with the s-HCT to that obtained by a clinical CT scanner.

Results: The CNT x-ray source array was found to have a consistent focal spot size of 1.3×1.1 mm² for all beams (IEC 1.0). At 120 kVp the HVL was measured to be 5.8 mm Al. Axial images have been acquired with slice thickness 2.5 mm to evaluate the imaging performance of the s-HCT system. Contrast-noise-ratio was measured for the acrylic (120 HU) and water (0 HU) materials in the ACR CT 464 phantom Module 01. A value of 5.2 is reported for the benchtop setup with an entrance dose of 2.9 mGy, compared to the clinical measurement of 30.5 found at 74.5 mGy. These images demonstrate that the s-HCT system based on CNT x-ray source arrays is feasible.

Conclusion: Customized CNT x-ray sources were developed specifically for head CT imaging. The feasibility of using this source array to construct a s-HCT scanner has been demonstrated by emulating a potential CT configuration. It is shown that diagnostic quality CT images can be obtained using the proposed system geometry. These preliminary images provide confidence that a s-HCT system can be constructed for clinical evaluation.

Purpose: The invention of carbon nanotube (CNT) x-ray source array has allowed development of many novel imaging systems including stationary tomosynthesis devices for breast, chest and dental imaging. This technology enables stationary computed tomography with potentially a fast data acquisition rate and a mechanically robust structure by eliminating the rotating gantry. It reduces the image blur caused by the mechanical motion. The purpose of this work is to explore possible system configurations of stationary head CT (s-HCT) using fixed-position linear CNT x-ray source arrays and detector arrays. Methods: Sinogram coverage is used for qualitative evaluation on the CT projection data collection efficiency for a given configuration. Accordingly, the configuration is optimized based on the coverage in sinogram space. To evaluate the system feasibility on imaging low-contrast brain tissues, a modified low-contrast Shepp-Logan phantom is implemented for quality assessment using quantitative metrics. Different Iterative Reconstruction methods are compared for both qualitative and quantitative assessment as well. Results: The sinogram coverage of s-HCT configurations changes significantly with different number of CNT source arrays used, as well as the layout of the geometry. Preliminary results suggest that a s-HCT configuration with three planes gives a nearly completed sinogram coverage which provides enough information to reconstruct image with good quality. Different reconstruction techniques are used for such configuration with a low-contrast head phantom. High-quality images are reconstructed for the proposed configuration. Conclusion: An optimized s-HCT system configuration can be built with few linear CNT x-ray source arrays. Given such configuration, Iterative Reconstruction algorithms in conjunction with Total-Variation Regularization provides highquality images even for low-contrast objects.

Three-dimensional Polarized Light Imaging (3D-PLI) enables the investigation of nerve fibers in the brain by probing their birefringence with polarimetric measurements of unstained brain sections. Based on a biophysical model, it allows the estimation of the three-dimensional fiber orientation. In this work, a novel probabilistic approach capable of providing confidence measures for the inferred fiber orientation is introduced. The developed analysis utilizes Bayesian inference to derive a distribution for the model parameters. Simulations show that the obtained confidence measures are accurate. On human brain data, the novel algorithm enables to quantify levels of confidence in 3D-PLI for the first time. The method shows that fiber orientation estimation in 3D-PLI is robust for white matter regions.

It is important to determine the patient’s skin dose accurately for fluoroscopic interventional procedures in order to estimate the risk of deterministic injury. The purpose of this study is to investigate how the patient’s skin dose changes as a function of x-ray beam incident angle for flat and curved surfaces. The primary and scatter dose was calculated averaged over a 2.0 mm depth at the surfaces of both cubic and cylindrical water phantoms to simulate different patient curvature. The total skin dose was calculated using EGSnrc Monte-Carlo (MC) software with 10¹⁰ photons incident and the primary dose was calculated at the central axis using the mass energy absorption coefficients published by NIST and integrated over the beam-energy spectrum. Simulations were done for incident angles from 90 to 10 degrees, beam field sizes from 5 to 15 cm, cylinder diameters from 20 to 30 cm, and beam energies from 60 to 120 kVp. The results show the scatter-plus-primary to incident-primary dose ratio decreases with decreasing incident angle due to increased primary attenuation and decreases from cubic to cylindrical phantom and with decreasing cylinder diameter at all angles due to reduced backscatter. These results can be used to determine angular correction factors needed to accurately estimate patient skin dose when the beam is not normal to the entrance surface during fluoroscopic procedures.

The imaging parameters used in neurointerventional procedures were evaluated to better understand the exposure techniques used clinically and their impact on patient dose. All parameters are available on the imaging system’s network bus in real time for each exposure pulse during a procedure. The Canon Dose Tracking System (DTS), which we developed, records the parameters of each exposure event in a raw data log file of controller area network (CAN) packets. We have collected such log files for 120 neurointerventional cases. Parameters are extracted by converting the raw data log file to a reformatted text file using a MATLAB script. The text file is input into a Microsoft Visual Studio project which outputs a new text file, which, with a reference table, allows the parameters to be identified. A written Python script is used to extract the specific parameters that were to be evaluated and output a .csv file. These were then input into MATLAB for analysis. The parameters extracted were the kVp, beam filter type, mAs, and the cranial/caudal angle as well as the RAO/LAO angle for the frontal and lateral gantries for DA and pulsed fluoroscopy (PF) modes. The gantry angles ranged from 34⁰ CRA to 42⁰ CAU and from 114⁰ RAO to 91⁰ LAO for DA and PF, respectively. The median kVp was 84 and 73 and the average per frame mAs was about 11 and 1.8 for DA and PF, respectively. This analysis should allow a better understanding of clinical practice in order to relate technique to patient and staff dose.

We have developed a prototype scatter-display system (SDS) which includes a top-down view, virtual reality (VR) representation of an interventional room containing a color-coded scatter dose rate distribution in real-time. To represent various attenuating objects of interest in the room, such as the C-Arm gantry and ceiling mounted shield, the STL toolbox in Matlab was implemented to produce a 3D VR description of the objects. Attenuation by objects in the room will alter the dose distribution and may lead to shielding of individual staff members, and thus representation of those objects in the software is needed for precise dose rate estimations. Determination of the spatial regions of attenuation requires accurate specification of object position. To retain identification of a ceiling mounted shield, we implemented an open-source package which maintains object recognition using the depth camera feed of a Microsoft Kinect V2 and the features-from-accelerated-segment- test (FAST) algorithm in OpenCV for a dense sampling of salient features. The depth information from the identified object is transferred to an open-source robot operating system (ROS) wrapper for specification of the 3D position to be fed into the SDS. To compute the C-Arm gantry position, we take advantage of a controller area network (CAN) bus interfaced with the angiography system’s application programming interface (API). Methods for computing gantry and ceiling mounted shield shadow regions are discussed and demonstrated. FAST was applied to the ceiling-shield assembly’s flange with reliable recognition. Including object attenuation of room scatter in the SDS will facilitate accurate dose rate computation.

The functionality of a real-time, top-down view virtual reality (VR) display of scattered radiation during fluoroscopic interventional procedures is being expanded to incorporate automatic input of staff member locations. Microsoft Kinect V2 depth sensing camera input was integrated into an open-source Robot Operating System (ROS) wrapper to facilitate automatic extraction of relative landmark body feature coordinates. Coordinates for the torso are selected to represent the staff member location in the selected plane of scatter; these coordinates are stored in a text file to be input into the real-time scatter display system (SDS). Accuracy of the depth sensing camera was evaluated using a pinhole camera model. This model was also implemented in an ROS wrapper to calibrate the Microsoft Kinect V2. Calibrated values were then implemented within a coordinate transformation algorithm which converts the physical distance measurements in the frame of the Kinect to normalized coordinates used in Matlab for visualization of the top-down horizontal plane of the interventional suite. Impact on real-time performance was evaluated for both staff member position update on-screen as well as for the update of SDS image frames.

The lens of the eye can receive a substantial amount of radiation during neuro-interventional fluoroscopic procedures, increasing the risk of cataractogenesis for the patient. The purpose of this study is to investigate the variation of eye lens dose with a variation of the location of the beam isocenter in the head. The primary x-ray beam of a Toshiba (Canon) Infinix fluoroscopy machine was modeled using EGSnrc Monte Carlo code and the lens dose was calculated using 2 × 10¹⁰ photons incident on the anthropomorphic Zubal computational head phantom for each simulation. The Zubal phantom is derived from a CT scan of an average adult male and has internal organs, including the lenses, segmented for dose calculation. Computations were performed with the head shifted vertically +/- 4 cm and in the cranial-caudal and lateral directions incrementally up to 6 cm in either direction. At each position, the gantry was rotated to various LAO/RAO and CAU/CRA angles, both 5 cm × 5 cm and 10 cm x 10 cm entrance field sizes were used and the kVp was varied. The results show that substantial changes in lens dose occur when the head is shifted and can result in a dose difference between eyes of over 6 times at certain beam angles for the 5 cm × 5 cm field size. The results of this study should provide increased accuracy in lens dose estimation during neuro interventional procedures and, when incorporated into our real-time dose-tracking system, help interventionalists manage patient lens dose during the procedure to minimize risk.

Fluoroscopy is a radiological technique that provides real-time x-ray viewing in interventional and angiographic procedures. In fluoroscopic procedures, there are several issues have to be solved, such as a risk of radiation exposure to the patients and operators and low image qualities by motion blur. To lower the radiation dose and motion blur, most of fluoroscopic systems provide a pulse-mode operation. However, conventional systems adopt filament-based thermionic analog x-ray tubes that generate relatively longer x-ray pulses than a few milliseconds due to intrinsic difficulty in modulating electron emissions, thus still have many problems of motion blur for fast objects, unnecessary x-ray radiation, and mismatched frame rate to the moving objects. In this work, we tried to solve these problems by suggesting an adaptively variable frame-rate fluoroscopy with an ultra-fast digital x-ray tube (DXT) based on carbon nanotube (CNT) field electron emitters.

We first fabricated a vacuum-sealed CNT DXT and its monoblock with a power generator for the fluoroscopic system. Ultra-short and high-frequency x-ray pulses of up to 500 ns at 1 MHz was achieved by the direct control of electron emission through an active current-control unit. X-ray pulse frames from the CNT DXT with a tube voltage of 120 kV and current of 20 mA were adaptively modulated in the range of 1-30 Hz according to the motion of objects, greatly improving temporal resolution with a reduced radiation dose. The adaptively variable frame-rate fluoroscopy could pave the way for both reducing x-ray doses and improving temporal and spatial resolution.

In recent years, it has been shown that it is possible to achieve time-resolved cone-beam CT imaging using a Carm cone-beam CT system equipped with a back-and-forth multi-sweep data acquisition mode. A reconstruction algorithm which has been referred to as SMART-RECON has demonstrated to achieve better than 1.0 frame per second (fps) temporal resolution. Recently, another innovation was introduced into SMART-RECON to achieve 7.5 fps, this innovation opens up a whole new opportunity to extract other physiological information besides the well-known cone-beam CT perfusion. In this work, we study the feasibility to obtain quantitative blood flow information from high temporal resolution time-resolved cone beam CT angiography. This new possibility would enable physicians to more accurately pinpoint occlusion sites and provide the needed image guidance to plan for endovascular therapy in acute ischemic stroke patients as an example. Numerical simulations with ground truth and preliminary clinical case studies have been conducted to demonstrate the feasibility of blood flow quantification from time-resolved CBCT angiography.

Proton radiography is a potentially valuable tool for the image guidance of proton therapy cancer treatment. While proton therapy is desirable due to its high dose deposition accuracy, proton radiography, which could, in theory, be applied simultaneously to treatment, has an intrinsically low soft tissue contrast, making it difficult to visualize tumors. To enhance this tumor contrast, high-Z nanoparticles could be targeted to a tumor before imaging. To assess the efficacy of gold as a contrast agent, phantoms consisting of gold leaf mounted on acrylic backing were developed to simulate a tumor tagged with gold nanoparticles (AuNPs). Calculations are presented to correlate a given thickness of gold with scenarios for tagging cancers with AuNPs, in terms of the size of the functionalized nanoparticles, the diameter of the tumor, as well as the efficiency by which the nanoparticles are taken up by the malignant cells. These calculations determined phantoms that best describe particular tagging conditions, and are also applicable to ex vivo specimens made by injecting AuNPs into a mouse model. Using a ×3 proton magnifying lens with the 800-MeV LANSCE proton beam, a 1-μm-thick Au foil was radiographically discernible within 1 cm of acrylic, representing sensitivity to a material percent density change of 0.2%. This indicates that AuNP-enhanced proton radiography could be used to characterize small tumors, allowing for early detection and treatment of malignant tissues, as well as for in situ imaging for enhanced treatment localization.

The system presented herein consists of a custom 200 kV electron tube with deflection magnets and stationary water-cooled targets for radiotherapy (RT) and imaging. The electron beam is deflected and dwelled along 41 discrete anode locations equally spaced by 1 cm in a line, at equivalent speeds of 1000 cm/s, to create a focused radiotherapy source. The treatment beam is collimated into a triangular polyhedron shape, producing a 4 cm focal spot at the isocenter and corresponding planning-treatment-volume (PTV). This beam shape would allow entry dose to be distributed over large areas for skin sparing. The source is mounted on a ring gantry that rotates at speeds up to 1.5 rpm. Preliminary dose rate measurements were collected in air at 140 kV beam energy, up to 80 mA beam current. Radiographic film was used to collect an image of the treatment beam at isocenter. Results are presented and can be extrapolated to a dose rate of 2 Gy/min for a 140 kV, 200 mA beam. The electron beam can be deflected within 1 ms from therapy path to an additional array of 19 imaging targets, which provides hardware capabilities for real-time tomosynthesis and image-guided RT. Onboard cone beam CT for patient positioning is also available. The utilization of 200 kV beam treatment energies compared to MV greatly reduces the required shielding (4-6 mm lead vs. 1-2 m concrete) and the cost of radiotherapy system installations. Systems can be mounted onto standard mobile trailers for use at remote locations.

The use of C-Arm-based cone-beam computed tomography (CBCT) plays an increasing role in interventions, especially for guidance and therapy control. The slow speed and the high dose limit the use of CBCT to research and prevent its widespread application in clinical routine. Acquiring less data using greater angular step or limited angular range is an obvious way to overcome these issues. However, images reconstructed from such datasets using standard reconstruction algorithms are deteriorated with severe artifacts. In this work, we investigate the use of a new nullspace-constrained modification scheme for sparse-view and limited-angle intra-operative CT image reconstruction. This scheme allows to perform fast unconstrained ART reconstruction, and, based on prior knowledge regarding the object to be reconstructed, some modifications restricted to the nullspace of the system can be easily applied as a post-processing step. Within this scheme, we enforce sparsity by integrating geometric prior information regarding the interventional tool itself, besides a high-quality pre-operative CT image. The presented method was compared to the compressed sensing-based algorithms NIHT and PrIDICT. Performance was evaluated qualitatively and quantitatively. This new scheme is shown to be promising for low-dose intra-operative image reconstruction. Compared to PrIDICT and NIHT, it shows higher reconstruction accuracy and demonstrates the ability to precisely visualize the instrument’s position even when only 15 projection views are acquired over a full angular range. It demonstrates an accurate reconstruction with a high degree of robustness against data incompleteness and sparsity level over-estimation.

Inverse geometry CT(IGCT) employs source array and small detector, which raises requirement for new reconstruction method. Gridding-based method and direct FBP method have been developed for reconstruction, with degraded spatial resolution or computational efficiency. We recently propose a new FBP reconstruction for IGCT that merges the projections from all the sources to the final images using designed weightings. Although this method achieves excellent spatial resolution and is simple in practical use, the necessary zero-padding step that extends the detector to cover the whole scanned object causes increased time consumption in projection filtering and backprojection. In this paper, we propose an accelerated reconstruction for IGCT via derivative back-projection filtration(DBF). Compared with the proposed FBP reconstruction, the DBF reconstruction employ the same weightings, with only changes in the filter kernel, i.e., substitute the local derivative filter for the global ramp filter. Therefore, the DBF reconstruction formula could be obtained via the same derivation as in FBP reconstruction, and the computational efficiency is expected to improve since local filter poses no requirement for zero-padding. However, initial DBF reconstruction present images with strong streaking artifacts, so we further simplify the reconstruction to a stable implementation using the data redundancy. Simulation studies and phantom studies reveal that proposed method present images with comparable numerical accuracy and spatial resolution to the FBP reconstruction. In comparison with FBP, proposed method achieves acceleration with a ratio of approximate 7 for reconstruction of IGCT system with 9 sources.

4D-CBCT reconstruction technique could provide a sequence of phase-resolved images to alleviate motion blurring artifacts as a result of respiratory movement during CT scanning. However, 4D-CBCT images are degraded by streaking artifacts due to the under-sampled projection used for the reconstruction of each phase. Based on the high correlation of these 4D-CBCT images, estimating the deformation vector fields (DVF) among them via a deformable registration algorithm is one of the possible solutions to improve the image quality. Often, the intensity-based similarity metric is utilized in the optimization problem by minimizing the squared sum of intensity differences (SSD) of the reference image and the target image. However, this metric is not suitable for the 4D-CBCT registration case, because the quality of both the reference image and the target image are not always guaranteed. As a result, the registration accuracy of the conventional SSD metric still has room to improve. In our method, by considering the characteristic of the phase-depended images, we design a novel similarity metric: 1) A prior image reconstructed by the whole projection set is regarded as the reference image; 2) Instead of an intensity-based similarity metric alone, we proposed a free-form based optimization function associating the gradient information in spatial domain with the projection-based constraint. To validate the performance of the proposed method, we carried out a phantom data and a patient data to compare with the classical Demons algorithm. To be specific, the quality of the registered image was improved to a great extent, especially in regions of interest of moving tissues. Quantitative evaluations were shown in terms of the rooted mean square error (RMSE) by our method when compared with existing Demons method.

Substantial researches have shown that the wildly used statistical iterative reconstruction (SIR) methods without strong constraints, for example, the maximum likelihood estimation, could induce excessive noise into reconstructions. The noise significantly degrades image quality. In this case, the traditional method of iterating till convergence is no longer feasible. In this work, we propose a structural similarity index (SSIM) based stopping criterion for SIR. We define an indicator, referred as mSSIM, of the turning point of noise amplification based on SSIM map of reconstructed images from two adjacent iterations. The mSSIM is computed from the average of SSIM map within regions of interest (ROI). A threshold of the mSSIM is set to be the stopping criterion of iterative reconstruction. We applied this strategy to the cases of two different data noise models and iterative step sizes. Experimental tests are done on two practical datasets. Result shows that we could successfully and stably obtain images of similar quality by applying this SSIM-based stopping criterion in different cases.

Lens-coupled micro-CT is a common type of high-resolution micro-CT with spatial resolution up to sub-micron. Due to diffraction limit of optical system, mechanical inaccuracy and noise effects in lens-coupled micro-CT, its spatial resolution is difficult to further improve, and its scope of application is also greatly restricted. In this paper, we propose a super-resolution model-based iterative reconstruction (SR-MBIR) algorithm to improve the spatial resolution of micro- CT. The algorithm firstly designs a scanning scheme, in which the scanned object is placed on a high-precision nano stage mounted on the rotation stage, and when the sample is moved in a fixed trajectory, a plurality of projection images with sub-pixel displacement information are obtained. Then based on the MAP reconstruction framework, a sub-pixel displacement model, a blur model and a noise statistical model are incorporated in the forward projection process in reconstruction procedures. By upsampling the data grid and using the sub-pixel displacement information of the projection images in the back projection process, the system resolution can be improved. In the experiment, a bamboo fiber is scanned and reconstructed by the proposed method. Compared with other reconstruction approaches, the result demonstrates that the reconstructed images by the proposed method have sharper boundaries and more details.

As an alternative to conventional sources, field emission x-ray cold cathodes of nanomaterials have been developed in recent years. Many different imaging geometries with this kind of source have been proposed, which has the merits of fast response, low energy consumption, and individually addressable switching ability. In this work, we proposed a novel digital tomosynthesis (DTS) geometry based on field emission flat-panel X-ray source array (FEF X-ray source array) and a reconstruction method based on this new geometry. The new DTS with designed lighting mode has shorter acquisition time and lower dose compared with the traditional DTS scheme. Due to the designed lighting mode, it cannot use a traditional reconstruction algorithm. The proposed reconstructed algorithm builds the relation on photons to solve the reconstruction problem. The simulated result shows that the proposed method can obtain 5pl / mm in the X-Y plane and 2pl / mm in the Z plane which indicates the potential of the proposed reconstruction modality.

This work concerns automatic geometric calibration in 3D parallel geometry. The aim is to develop analytical formulas to identify the view direction and the in-plane angle only from projections of the object. We consider that two view directions are known and treat all others once at a time, repeating the same three-projection procedure. We identify the unknown view direction of the third projection image and the in-plane angles of the three projections.

In this work, we present a novel model-based material decomposition (MBMD) approach for x-ray CT that includes system blur in the measurement model. Such processing has the potential to extend spatial resolution in material density estimates - particularly in systems where different spectral channels exhibit different spatial resolutions. We illustrate this new approach for a dual-layer detector x-ray CT and compare MBMD algorithms with and without blur in the reconstruction forward model. Both qualitative and quantitative comparisons of performance with and without blur modeling are reported. We find that blur modeling yields images with better recovery of high-resolution structures in an investigation of reconstructed line pairs as well as lower cross-talk bias between material bases that is ordinarily found due to mismatches in spatial resolution between spectral channels. The extended spatial resolution of the material decompositions has potential application in a range of high-resolution clinical tasks and spectral CT systems where spectral channels exhibit different spatial resolutions.

In a modern CT system, the localizer radiograph (LR), which makes a visualized scan range planning and provides information for individual patient oriented automatic exposure control, is now a standard procedure before a major diagnostic scan. We proposed a rotating projection based localizer radiograph (RPLR), in contrast to current realization as static tube positioned scanning. With this approach, the localizer Radiograph can be realized to improve the whole clinical workflow by providing no geometry deformation from multiple viewing directions within one scan, with balanced image performance and controllable x-ray dose, . Somehow with this feasibility check on current 16-row CT gantry, the scanned dose is still higher than traditional situation, but with dedicated dose reduction techniques proposed with the RPLP, the exposure dose for the scanning can be dramatically decreased.

Compared with conventional CT imaging, dual energy CT (DECT) provides more material information using two sets of projection data with different x-ray spectra. Current implementations of DECT typically require two CT scans with different x-ray energies or costly upgrades of x-ray tube or detector. In this work, we propose a single-scan DECT method for conebeam CT (CBCT) imaging by inserting a rotating metal filter between x-ray source and the scanned object.
CBCT systems have a relatively slow scanning speed, which offers a possibility of altering the x-ray spectrum via a metal filter for every other projection. We insert a rotating metal filter between the x-ray source and the detector, with the filter rotation synchronized with the x-ray pulse and the detector readout. As such, one single-scan acquires interlacing low- and high-energy CT projections. We then reconstruct high and low CT images using a combined approach of iterative reconstruction and decomposition with similarity matrix and TV based regularization to preserve the image structure and denoising.
We have implemented the proposed method on a benchtop CBCT system, and compared the performance with that of a photon-counting detector based system on physical phantoms. On an electron density phantom, our proposed algorithm decreases the noise from a root mean square of error of 2.105 to 0.017(10²³ 𝑒/𝑐𝑐) compared with a direct decomposition from FBP reconstructed images. On a resolution gauge phantom, our method maintained the resolution of 15 line pairs per mm compared with FBP reconstruction and decomposition results and subdue the image aliasing artifacts under same noise level with that of a TV regularization results without losing spatial resolution. The performance is further validated on an anthropomorphic head phantom.
We propose a practical and accurate DECT method on CBCT with insertion of a rotating metal filter. Compared with those of the photon-counting detector based approaches, our method achieves better performances on spatial resolution of decomposed material images and similar performance on electron density measurement, with attractive features of no requirement of scan time increase or costly imaging component upgrades.

Dynamic imaging (such as computed tomography (CT) perfusion, dynamic CT angiography, dynamic positron emission tomography, four-dimensional CT, etc.) is widely used in the clinic. The multiple-scan mechanism of dynamic imaging results in greatly increased radiation dose and prolonged acquisition time. To deal with these problems, low-mAs or sparse-view protocols are usually adopted, which lead to noisy or incomplete data for each frame. To obtain high-quality images from the corrupted data, a popular strategy is to incorporate the composite image that reconstructed using the full dataset into the iterative reconstruction procedure. Previous studies have tried to enforce each frame to approach the composite image in each iteration, which, however, introduces mixed temporal information into each frame. In this paper, we propose an average consistency (AC) model for dynamic CT image reconstruction. The core idea of AC is to enforce the average of all frames to approach the composite image in each iteration, which preserves image edges and noise characteristics while avoids the invasion of mixed temporal information. Experiment on a dynamic phantom and a patient for CT perfusion imaging shows that the proposed method obtains the best qualitative and quantitative results. We conclude that the AC model is a general framework and a superior way of using the composite image for dynamic CT reconstruction.

Many improvements have been made to amorphous silicon (a-Si) flat panel detectors (FPDs) to meet the market needs for different x-ray imaging applications. With the current generation of a-Si FPDs the performance is limited by the a-Si thin film transistors (TFTs). The low electron mobility of a-Si necessitates large TFT’s with large parasitic dataline capacitance, which increases electronic noise and reduces the pixel fill factor (FF). In other words, large TFT’s negatively impact Signal-to-Noise Ratio (SNR). CMOS FPDs were introduced to provide improved low dose imaging performance and faster readout times, but the increase in cost can be prohibitive. IGZO TFTs have an electron mobility that is <10x higher than a-Si, which facilitates a reduction in the size of the TFT while also reducing the pixel discharge time, resulting in an increase to both the detector readout rate and the SNR. Reducing the TFT size is particularly important in achieving adequate low dose performance in dynamic detectors with pixels approaching 100μm. A 31cm x 31cm (100μm) FPD using IGZO TFTs was evaluated at 25 frames/second (fps) in 1x1, 2x2, 3x3, and 4x4 binning. In the 1x1 standard noise configuration, the noise equivalent dose (NED) was 24nGy with a max linear dose (MLD) of 10uGy. The NED was reduced to 6.6nGy in the 2x2, 3.4nGy in the 3x3, and 2.4nGy in the 4x4 mode. The linearity of the IGZO imager was comparable to a-Si imager. The 1x1 MTF was 57.5% at 1 lp/mm and 28.5% at 2lp/mm. The quantum limited DQE in the 1x1 binning mode was 79% at 0 lp/mm and 47% at 1 lp/mm. The 1x1 DQE measured at NED was 71% at 0 lp/mm, 29% at 1 lp/mm. This paper will explore how to optimally employ IGZO and present data from a first IGZO imager, showing that IGZO is an excellent technology for the future of FPDs.

In this paper, the behavior of the amorphous indium–gallium–zinc-oxide (a-IGZO) thin film transistors (TFTs) after X-ray irradiation is analyzed. With the increase of irradiation dose, the a-IGZO TFTs show more negative shifts of threshold voltage (Vth), while both the field effect mobility and subthreshold swing (S.S.) keep fairly constant. Not like being a switch in passive pixel sensor (PPS), the Vth shift of the TFT in active pixel sensor (APS) can result in variation in the output signal. With techniques of compensating Vth shift, a-IGZO TFTs are more suitable for the application of Xray image sensors from the viewpoint of stability.

CZT detectors are gaining popularity in the medical field. A numerical study of the response of CZT detector pixels to parallel and oblique x rays was conducted. The average k-fluorescence energy generated in a given pixel is 26.0 keV. It is assumed that such x rays travel a mean free path mfp = 0.161 mm in CZT before being absorbed. For applications in nuclear medicine, incident 140 keV photons were used. Each pixel had an area 1.5 mm x 3.3 mm and thickness 5 mm. The spacing between pixels was Δ = 0:3 mm and since Δ >mfp, the model predicts no cross talk between pixels. For parallel photons, the probability of escape is PF_esc = 4:0% and the quantum efficiency η = 0:88. Next consider an isotropic source located at a distance 20 cm from the detector plane. For incident angles of 0° to 23° η varied from 0.88 to 0.5 and the fluorescence escape from 3.8 % to 2.0 %. For breast CT applications a typical pixel size is 0.2 mm by 0.2 mm and thickness 0.75 mm. The corresponding values for photons of energy 60 keV are as follows: (i) parallel η = 0:94, PF_esc = 30%, oblique η = 0:94, PF_esc = 30% at 0°; η = 0:55, PF_esc = 18% at 23°. This work provides some quantification of the response of CZT detector pixels to x rays. Cross talk would need to be examined for the breast CT detector.

X-ray detectors increasingly utilize active pixel CMOS instead of amorphous silicon technology because of its superior noise, pixel lag, readout speed and offset stability. We already demonstrated [1] that adding an a additional pixel capacitance to a standard 3T CMOS pixel architecture allows operating the detector in either high sensitivity (HS) or high saturation dose (HD) mode. Since the charge capacity is limited in HS mode, a large signal will saturate the pixel, causing a loss of information. In HD mode, a very low exposure will lead to a loss of contrast-to-noise ratio (CNR) due to the inherently higher noise floor in this mode. Using the same 3T pixel architecture, we propose a dual readout method to combine the benefits of HS and HD modes. After a single exposure the pixel signal is read twice, respectively in HS and HD mode. The two linear signal values are then combined to create the pixel value of the final image. In this paper proof of concept is demonstrated using images acquired separately and combined offline. The benefits of this method are demonstrated for different x-ray imaging modalities such as mammography, extra-oral dental, interventional and non-destructive testing. Using different detector models, results show that extended dynamic range combined with low noise leads to better image quality without introducing artifacts. It is expected that implementing the fast CMOS-sensor dual readout and image synthesis inside the detector will preserve important application requirements such as frame rate, data bandwidth and power consumption.

Radiation dose is of an important consideration for x-ray fluoroscopy imaging of interventional C-arm systems. Low-dose imaging is always expected, but it also results in noisy images. Noise reduction is one of the important topics for fluoroscopic images. Recently, the advances in deep learning have achieved outstanding denoising results for x-ray images. However, most existing methods in the field focus only on 2D image denoising from frame-by-frame independently, and removing temporal noise in image sequence remains a challenging problem. Our goal is simultaneously to reduce both spatial and temporal noises for fluoroscopic image sequences in a unified framework. In this paper, we propose a deep learning algorithm that extensively utilizes temporal information to maximize the efficiency of noise reduction. The proposed convolutional neural network (CNN) is based on DenseNet¹ and DnCNN² but with improved multi-channel input layers for image sequence. That network architecture not only enables spatial domain deep learning from the input of every individual frame, but also is able to make full use of temporally correlative information among adjacent frames for temporal domain learning. In order to further suppress temporal noise resulting in visual flickers of image sequence, an additional term is introduced to the network loss function. Besides two conventional terms of L2 and perceptual losses, the new proposed loss calculates the statistical variance of the network performance caused by random influence of temporal imaging. The developed algorithm is evaluated with fluoroscopic phantom images and clinical patient data, showing superior performance for spatio-temporal denoising.

Reducing the radiation dose is always an important topic in modern computed tomography (CT) imaging. As the dose level reduces, the conventional analytical filtered backprojection (FBP) reconstruction algorithm becomes inefficient in generating satisfactory CT images for clinical applications. To overcome such difficulties, in this study we developed a novel deep neural network (DNN) for low dose CT image reconstruction by exploring the simultaneous sinogram domain and CT image domain denoising capabilities. The key idea is to jointly denoise the acquired sinogram and the reconstructed CT image, while reconstructing CT image in an end-to-end manner with the help of DNN. Specifically, this new DNN contains three compartments: the sinogram domain denoising compartment, the sinogram to CT image reconstruction compartment, and the CT image domain denoising compartment. This novel sinogram and image domain based CT image reconstruction network is named as ADAPTIVE-NET. By design, the first and third compartments of ADAPTIVE-NET can mutually update their parameters for CT image denoising during network training. Clearly, one advantage of using ADAPTIVE-NET is that the unique information stored in sinogram can be accessed directly during network training. Validation results obtained from numerical simulations demonstrate that this newly proposed ADAPTIVE-NET can effectively improve the quality of CT images acquired with low radiation dose levels.

Low-dose computed tomography (LDCT) examinations are of essential usages in clinical applications due to the lower radiation-associated cancer risks in CT imaging. Reductions in radiation dose can produce severe noise and artifacts that can affect the diagnostic accuracy of radiologists. Although deep learning networks have been widely proposed, most of these networks rely on a large number of annotated CT image pairs (LDCT images/high-dose CT (HDCT) images). Moreover, it is challenging for these networks to cope with the growing amount of CT images, especially large amount of medium-dose CT (MDCT) images that are easily to collect and have lower radiation dose than the HDCT images and higher radiation dose than the LDCT images. Therefore, in this work, we propose a progressive transfer-learning network (PETNet) for low-dose CT image reconstruction with limited annotated CT data and abundant corrupted CT data. The presented PETNet consists of two phases. In the first phase, a network is trained on a large amount of LDCT/MDCT image pairs, similar to the Noise2Noise network that has shown potential in yielding promising results with corrupted data for network training. It should be noted that this network would inevitably introduce undesired bias in the results due to the complex noise distribution in CT images. Then, in the second phase, we combined the pre-trained network and another simple network to construct the presented PETNet. In particular, the parameters of the pre-trained network are frozen and transferred directly to the presented PETNet, and the presented PETNet is trained on a small amount of LDCT/HDCT image pairs. Experimental results on Mayo clinic data demonstrate the superiority of the presented PETNet method both qualitatively and quantitatively compared with the network trained on LDCT/HDCT images pairs, and Noise2Noise method trained on LDCT/MDCT image pairs.

Recently, deep neural networks (DNNs) have been widely applied in low-dose computed tomography (LDCT) imaging field. Their performances are highly related to the number of the pre-collected training data. Meanwhile, the training data is usually hard to obtain, especially for the high-dose CT (HDCT) images. And HDCT images sometimes contain undesired noises, which easily result in network overfitting. To address the two issues, we proposed a cooperative meta-learning strategy for CT image reconstruction (CmetaCT) combining the metalearning strategy and Co-teaching strategy. The meta-learning (teacher/student model) strategy allows for training network with a large number of LDCT images without the corresponding HDCT images and only a small number of labeled CT data in a semi-supervised learning manner. And the Co-teaching strategy is able to make a trade-off between overfitting and introducing extra errors, which includes a part of samples in every minibatch for updating model parameters. Due to the capacity of meta-learning, the presented CmetaCT method is flexible enough to utilize any existing CT restoration/reconstruction network in meta-learning framework. Finally, both quantitative and visual results indicated that the proposed CmetaCT method achieves a superior performance on low-dose CT imaging compared with the DnCNN method.

Fully supervised deep learning (DL) methods have been widely used in low-dose CT (LDCT) imaging field and can usually achieve high accuracy results. These methods require a large labeled training set which consists of pairs of LDCT images as well as their corresponding high-dose CT (HDCT) ones. They successfully learn intermediate concept of features describing important components in CT images, such as noise distribution, and structure details, which is important to capture dependencies from LDCT image to HDCT ones. However, it should be noted that it is quite time-consuming and costly to obtain such a large of labeled CT images especially the HDCT images are limited in clinics. In comparison, lots of unlabeled LDCT images are usually easily accessible and massive critical information latent in the unlabeled LDCT can be leveraged to further boost restoration performance. Therefore, in this work, we present a semi-supervised noise distribution learning network to suppress noise-induced artifacts in the LDCT images. For simplicity, the presented network in termed as "SNDL-Net". The presented SNDL-Net consists of two sub-networks, i.e., supervised network, and unsupervised network. In the supervised network, the LDCT/HDCT image pairs are used for network training. And the unsupervised network considers the complex noise distribution in the LDCT images, and model the noise with a Gaussian mixture framework, then learns the proper gradient of LDCT images in a purely unsupervised manner. Similar with the supervised network training, the gradient information in a large of unlabeled LDCT images can be used for unsupervised network training. Moreover, to learn the noise distribution accurately, the discrepancy between the learned noise distribution in the supervised network and learned noise distribution in the unsupervised network can be modeled by a Kullback-Leibler (KL) divergence. Experiments on the Mayo clinic dataset verify the method is effective in low-dose CT image restoration with only a small amount of labeled data compared to previous supervised deep learning methods.

According to an increased use of computed tomography (CT) in medicine, the risk caused by radiation exposure has been considered as one of the major issues. In order to reduce the risk, low-dose CT imaging has attracted attention. However, the low-dose CT imaging causes low spatial resolution (LR) and high noise in reconstructed images. Recently, deep learning-based models have shown a feasibility for reducing noise and improving spatial resolution. However, these models have the drawbacks such as complex structures, large sample size and computational costs. In this study, a simple denoising and super-resolution convolutional neural network (SDSRCNN) was proposed to overcome the limitations of conventional methods. Two networks were trained for the denoising and super-resolution imaging separately, and the trained networks were linearly combined as a single network with a simple architecture. In comparison with conventional methods, denoise-autoencoder (DAE) and super-resolution convolutional neural network (SRCNN) were also implemented. We evaluated the performance of the SDSRCNN in terms of peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). The results showed that the proposed model could efficiently reduce noise and preserve spatial resolution information comparing the conventional methods. Therefore, the proposed model has the potential for improving the quality of CT images and rejecting the complexity of the conventional methods.

Based on the “as low as reasonably practicable” principle, the amount of radiation exposure of CT should be decreased without impairing image quality. However, the excessive reduction of radiation exposure in CT results in the degraded images with noise and artifacts. The purpose of this study was to construct virtual normal dose CT images from ultra-low dose CT images by using dilated residual networks (DRN) that can expand the receptive field. Our database consisted of 1,860 pairs of normal dose and ultra-dose chest CT images obtained from 12 patients. In our proposed method, the DRN which consisted of seven dilated convolutional layers were trained the relationship of signal patterns between ultra-low dose CT patches (small region) and the corresponding normal dose CT patches. The trained DRN was employed to construct the virtual normal dose CT images from the ultra-low dose CT images. The root mean squared error, peak signal to noise ratio, and structural similarity index for the ultra-low dose CT images to the normal dose CT images were 59.3, 32.6dB, and 0.897, whereas those for the constructed images from ultra-low dose CT with a model-based iterative reconstruction (MBIR) were 49.1, 34.1dB, and 0.956. Those indices for virtual normal dose CT images were 44.4, 35.0dB, and 0.964, showing a significant improvement when compared with the ultra-low dose CT images and the constructed images with MBIR. The virtual normal dose CT images achieved higher image quality as compared with the ultra-low dose CT images and the constructed images with MBIR.

Decreasing administered activity directly reduces radiation exposure to patients and medical staff, but meanwhile has adverse impacts on image quality and PET quantification accuracy. In this work, we propose to integrate multi-modality images and self-attention strategy into a cycle-consistent adversarial network (CycleGAN) framework to generate the full count PET image from low count PET and CT images. During the training stage, deep features are extracted by 3D patch fashion from low count PET and CT images, and are automatically highlighted with the most informative features by self-attention strategy. Then, the deep features are mapped to the full count PET image by using 3D CycleGAN. During the testing stage, the paired patches are extracted from a new arrival patient’s low count PET and CT images, and are fed into the trained networks to obtain the synthetic full count PET image. This proposed algorithm was evaluated using 16 patients’ data. Four-fold cross-validation was used to test the performance of the proposed method. The proposed method suppressed image noise significantly, and obtained images close to the diagnostic PET images. The organ boundaries can be better visualized on the PET images generated with the proposed method. We have investigated a method to estimate diagnostic PET image from low dose data. Experimental validation has been performed to demonstrate its clinical feasibility and accuracy. This technique could be a useful tool for low dose PET imaging.

Deriving accurate attenuation maps for PET/MRI remains a challenging problem because MRI voxel intensities are not related to properties of photon attenuation and bone/air interfaces have similarly low signal. This work presents a learning-based method to derive patient-specific pseudo computed tomography (PCT) maps from routine T1-weighted MRI in their native space for attenuation correction of brain PET. We developed a machine-learning-based method using a sequence of alternating random forests under the framework of an iterative refinement model. Anatomical feature selection is included in both training and predication stages to achieve excellent performance. To evaluate its accuracy, we retrospectively investigated 17 patients, each of which has been scanned by PET/CT and MR for brain. The PET images were corrected for attenuation on CT images as ground truth, as well as on PCT images generated from MR images. The side-by-side image comparisons and joint histograms demonstrated very good agreement of PET images after correction by PCT and CT. The mean differences of voxel values in selected VOIs were less than 4%, the mean absolute difference of all active area is around 2.5%. This work demonstrates a novel learning-based approach to automatically generate CT images from routine T1-weighted MR images based on a random forest regression with patch-based anatomical signatures to effectively capture the relationship between the CT and MR images. Reconstructed PET images using the PCT exhibit errors well below accepted test/retest reliability of PET/CT indicating high quantitative equivalence.

The accuracy of attenuation correction on whole-body PET images is subject to inter-scan motion, image artifacts such as truncation and distortion, and erroneous transformation of structural voxel-intensities to PET mu-map values. We proposed a deep-learning-based method to derive synthetic CT (sCT) images from non-attenuation corrected PET (NAC PET) images for AC on whole-body PET imaging. We utilized a 3D cycle-consistent generative adversarial networks (CycleGAN) to synthesize CT images from NAC PET. The model learns a transformation that minimizes the difference between sCT, generated from NAC PET, and true CT. It also learns an inverse transformation such that cycle NAC PET image generated from the sCT is close to true NAC PET image. Both generators are implemented by a fully convolutional attention network (FCAN), and followed by a discriminator which is structured as a fully convolutional network. A retrospective study was performed with a total of 60 sets of whole-body PET/CT, 40 sets for training and 20 sets for testing. The sCT images generated with proposed method show great contrast on lung, soft tissue and bony structures. The mean absolute error of sCT over true CT is less than 110 HU. Using sCT for whole-body PET AC, the mean error of PET quantification is less than 1% and normalized mean square error is less than 1.4%. We proposed a deep learning-based approach to generate synthetic CT from whole-body NAC PET for PET AC, which demonstrates excellent synthetic CT estimation accuracy and PET quantification accuracy.

Machine learning, especially convolutional neural network (CNN) approach has been successfully applied in noise suppression in natural image. However, shifting from natural image to medical image filed remains challenging due to specific difficulties such as training samples limitation, clinically meaningful image quality requirement and so on. To address this challenge, one possible solution is to incorporate our human prior knowledge into the machine learning model to better benefit its power. Therefore, in this work, we propose one prior knowledge driven machine learning based approach for positron emission tomography (PET) sinogram data denoising. Two main properties of PET sinogram data were considered in CNN architecture design, which are the Poisson statistics of the data and different correlation strength in the detector and view directions. Specially, for the statistical property, the sparse non-local method was used. For the correlation property, separate convolution was applied in two directions respectively. Experimental results showed the proposed model outperform the CNN model without prior knowledge. Results also demonstrate our insight of applying human knowledge strength the power of machine learning in medical imaging field.

Deep learning-based algorithms have been widely used in the low-dose CT imaging field, and have achieved promising results. However, most of these algorithms only consider the information of the desired CT image itself, ignoring the external information that can help improve the imaging performance. Therefore, in this study, we present a convolutional neural network for low-dose CT reconstruction with non-local texture learning (NTL-CNN) approach. Specifically, different from the traditional network in CT imaging, the presented NTL- CNN approach takes into consideration the non-local features within the adjacent slices in 3D CT images. Then, both low-dose target CT images and the non-local features feed into the residual network to produce desired high-quality CT images. Real patient datasets are used to evaluate the performance of the presented NTL-CNN. The corresponding experiment results demonstrate that the presented NTL-CNN approach can obtain better CT images compared with the competing approaches, in terms of noise-induced artifacts reduction and structure details preservation.

As a quantitative CT imaging technique, the dual-energy CT (DECT) imaging method attracts a lot of research interests. However, material decomposition from high energy (HE) and low energy (LE) data may suffer from magnified noise, resulting in severe degradation of image quality and decomposition accuracy. To overcome these challenges, this study presents a novel DECT material decomposition method based on deep neural network (DNN). In particular, this new DNN integrates the CT image reconstruction task and the nonlinear material decomposition procedures into one single network. This end-to-end network consists of three compartments: the sinogram domain decomposition compartment, the user-defined analytical domain transformation operation (OP) compartment, and the image domain decomposition compartment. By design, both the first and third compartments are responsible for complicated nonlinear material decomposition, while denoising the DECT images. Natural images are used to synthesized the dual-energy data with assumed certain volume fractions and density distributions. By doing so, the burden of collecting clinical DECT data can be significantly reduced, therefore the new DECT reconstruction framework becomes more easy to be implemented. Both numerical and experimental validation results demonstrate that the proposed DNN based DECT reconstruction algorithm can generate high quality basis images with improved accuracy.

Deep learning (DL) networks show a great potential in computed tomography (CT) imaging field. Most of them are supervised DL network greatly based on their capability and the amount of CT training data (i.e., low-dose CT measurements/high-quality ones). However, collection of large-scale CT datasets are time-consuming and expensive. In addition, the training and testing CT datasets used for supervised DL network are highly desired similarities in CT scan protocol (i.e., similar anatomical structure, and same kVp setting). These two issues are particularly critical in spectral CT imaging. In this work, to address the issues, we presents an unsupervised data fidelity enhancement network (USENet) to produce high-quality spectral CT images. Specifically, the presented USENet consists of two parts, i.e., supervised network and unsupervised network. In the supervised network, the spectral CT image pairs at 140 kVp (low-dose CT images/high-dose ones) are used for network training. It should be noted that there is a great difference of CT value between spectral CT images at 140 kVp and 80 kVp, and the supervised network trained with CT images at 140 kVp cannot be directly used for CT image reconstruction at 80 kVp. Then unsupervised network enrolls physical model and the spectral CT measurements at 80 kVp for fine-tuning the supervised network, which is the major contribution of the presented USENet method. Finally, accurate spectral CT reconstructions are achieved for the sparse-view and low-dose cases, which fully demonstrate the effectiveness of the presented USENet method.

Flat-panel radiography detectors employ the thin film transistor (TFT) panels to acquire high-quality x-ray images. Pixel defects in the TFT panel can degrade the image quality and lower the production yield of the panel, and ultimately increase the production cost. Hence, developing an appropriate defect correction algorithm for acquired images is important. Conventional algorithms are based on statistical learning and hence optimizing their performances is difficult especially for image edge parts. To alleviate this problem, a template matching technique can be used. In this paper, we considered various pixel-defect correction algorithms based on deep learning techniques, such as the artificial neural network (ANN), convolutional neural network (CNN), and generative adversarial networks, and compared their performances. The defect-correction performances are compared using practical x-ray images acquired from general radiography detectors. A concatenate CNN showed the best defect-correction performance. We also showed that a single-layer ANN could conduct an efficient defect correction in terms of both correction and computational complexity performances.

This study aimed to evaluate a denoising convolutional neural network reconstruction (DnCNNR) algorithm for reducing metal objects on digital tomosynthesis when using projection data for arthroplasty. For metal artifact reduction (MAR), we implemented a DnCNNR algorithm based on a training network (i.e., mini-batch stochastic gradient-descent algorithm with momentum) to estimate residual reference (140 keV virtual monochromatic [VM]) and object (70 kV with metal artifacts) images using projection data, and subtracted the estimated residual images from the object images using hybrid and subjectively reconstructed images (back projection and maximum likelihood expectation maximization [MLEM]). This DnCNNR algorithm was compared with a dual-energy material decomposition reconstruction algorithm (DEMDRA), VM, MLEM, established and commonly used filtered back projection (FBP) methods, and simultaneous algebraic reconstruction technique-total variation (SART-TV) with MAR processing. MAR was then compared using artifact index (AI) and texture analyses. For images that were in-focus were evaluated using a prosthesis phantom. The derived images yielded better results that were not influenced by the metal type (e.g., the AI was almost equal to the best value for the DEMDRA). The DnCNNR algorithm also yielded the best performance with regard in the texture analysis. The proposed algorithm is particularly useful for not affected by tissue misclassification.

Flat-panel detector based cone-beam CT systems have been widely used in image-guided interventions and image-guided radiation therapy. However, several notoriously difficult challenges persist in these cone-beam CT systems: given the relatively large cone angles used in data acquisition, scatter induced artifacts significantly degrade image quality and thus the algorithms to reduce these artifacts have remained an active area of research through out the past decade. To accommodate for the limited detector dynamic range, these systems often use auto-exposure control to homogenize the noise distribution, and as a result, both kV and mA are modulated in some systems making beam hardening correction extremely difficult. Additionally, when data acquisition time is long, inadvertent motion artifacts often exacerbate the situation. In this paper, we develop a deep learning method to empirically correct these most often observed artifacts in flat-panel based CBCT images.

Stationary inverse-geometry digital tomosynthesis (s-IGDT) has advantages in terms of motion artifact reduction and diagnostic efficiency improvement. However, truncation artifacts are caused in reconstructed images owing to the geometric characteristics of s-IGDT systems, and this drawback degrades the diagnostic accuracy. In order to overcome this limitation, we proposed a convolutional neural network (CNN)-based truncation artifact reduction method. We simulated a s-IGDT system with stationary X-ray source array and small detector. Also, we acquired s-IGDT images using 70 volumetric phantoms based on the SPIE-AAPM lung CT challenge dataset. The U-Net was used as the CNN architecture, and we trained the network through 207 s-IGDT images. We confirmed that the truncation artifacts with various patterns included in the prior images were clearly removed in the prediction images obtained by the trained network. Moreover, the quantitative evaluation showed that both of the structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) improved when using the proposed method. The averaged SSIM and PSNR of the prediction images were approximately 6% and 25% higher than those of the prior images, respectively. In conclusion, the proposed model based on the CNN has superior performance in removing the truncation artifacts of s-IGDT images.

Concerns on the risks of radiation dose in the cone-beam breast CT (CBBCT) motivated the development of low dose CBBCT (LdCBBCT). Due to the noisy and inadequate data acquisition in LdCBBCT, the conventional analytical Filtered Back Projection (FBP) algorithm tends to result in severe image artifacts and overwhelming noise. Model-based iterative reconstruction methods managed to reduce artifacts and enhance the signal-to-noise ratio but were unable to recover many fine structures and low contrast objects pertinent to diagnosis and treatment. To maintain the strengths of the model-based optimization framework and overcome its limitations in signal recovery, we adapted a CNN-based iterative reconstruction framework, termed Plugand-Play (PnP) proximal gradient descent (PGD) framework, that incorporated state-of-the-art deep-learningbased denoising algorithms with model-based image reconstruction. The PnP-PGD framework is achieved by combining a least-square data fidelity term for data consistency with a non-local regularization for image smoothness, which was solved via PGD. A deep convolutional neural network (DCNN) was plugged in to substitute the proximal operator of the regularization term. The PnP-PGD was evaluated on LdCBBCT scans of a breast phantom and was compared with Filtered Back Projection (FBP), Total Variation (TV), the BlockMatching 3D-transform shrinkage (BM3D), and the DCNN based post-processing method. Compared with FBP, iterative reconstruction, and BM3D, the proposed PnP-PGD substantially reduced image noise and artifacts. Compared with the DCNN based post-processing method, the PnP-PGD improved image contrast-tonoise ratio (CNR). The proposed PnP-PGD takes advantage of both model-based reconstruction and deeplearning-based denoisers, showing improved image quality.

Proton radiation therapy has shown highly conformal distribution of prescribed dose in target with outstanding normal tissue sparing stemming from its steep dose gradient at the distal end of the beam. However, the uncertainty in everyday patient setup can lead to a discrepancy between treatment dose distribution and the planning dose distribution. Conebeam CT (CBCT) can be acquired daily before treatment to evaluate such inter-fraction setup error, while a further evaluation on resulted dose distribution error is currently not available. In this study, we developed a novel deep-learning based method to predict the relative stopping power maps from daily CBCT images to allow for online dose calculation in a step towards adaptive proton radiation therapy. 20 head-and-neck patients with CT and CBCT images are included for training and testing. Our CBCT RSP results were evaluated with RSP maps created from CT images as the ground truth. Among all the 20 patients, the averaged mean absolute error between CT-based and CBCT-based RSP was 0.04±0.02, the averaged mean error was -0.01±0.03 and the averaged normalized correlation coefficient was 0.97±0.01. The proposed method provides sufficiently accurate RSP map generation from CBCT images, possibly allowing for CBCT-guided adaptive treatment planning for proton radiation therapy.

By exploiting the energy dependence of photoelectric and Compton interactions, dual-energy CT (DECT) can be used to derive a number of parameters based on physical properties, such as relative stopping power map (RSPM). The accuracy of dual-energy CT (DECT)-derived parametric maps relies on image noise levels and the severity of artifacts. Suboptimal image quality may degrade the accuracy of physics-based mapping techniques and affect subsequent processing for clinical applications. In this study, we propose a deep-learning-based method to accurately generate relative stopping power map (RSPM) based on the virtual monoenergetic images as an alternative to physics-based dual-energy approaches. For the training target of our deep-learning model, we manually segmented head-and-neck DECT images into brain, bone, fat, soft-tissue, lung and air, and then assigned different RSP values into the corresponding tissue types to generate a reference RSPM. We proposed to integrate a residual block concept into a cycle-consistent generative adversarial network (cycleGAN) framework to learn the nonlinear mapping between DECT 70keV/140keV monoenergetic image pairs and reference RSPM. We evaluated the proposed method with 18 head-and-neck cancer patients. Mean absolute error (MAE) and mean error (ME) were used to quantify the differences between the generated and reference RSPM. The average MAE between generated and reference RSPM was 3.1±0.4 % and the average ME was 1.5±0.5 % for all patients. Compared to the physics-based method, the proposed method could significantly improve RSPM accuracy and had comparable computational efficiency after training.

Filtered back projection (FBP) reconstruction is simple and computationally efficient and is used in many commercial CT (tomography) imaging products. However, higher Poisson noise levels or metal objects in the imaged area can lead to severe artifacts. Iterative reconstruction employs stochastic models for the imaging process and the characteristics of the medical images and can reduce Poisson noise and metal related artifacts. But it is computation-intensive and furthermore, its image models are relatively simple and cannot quite capture the highly complex nature of the medical images, leaving rooms for further improvement. Recent advances in neural networks and deep learning could offer potential solutions to overcome these two problems. Towards that end, most of the neural networks proposed so far for CT image reconstruction are feed-forward networks with CNN (convolutional neural network) and fully connected layers, attempting to learn the mapping from the projections or the FBP output to the reconstructed image. While these networks have demonstrated some promising reconstruction or post-processing results, their architectures are somewhat arbitrary and the question remains as to what would be a more principled way to find a good architecture, thereby further improving reconstruction results. One promising idea is to design the network structure based on signal processing principles such as MAP (maximum a posteriori) estimation and iterative optimization. In this work, we developed a novel RNN (recurrent neural network) based on an accelerated iterative MAP estimation algorithm. This network makes use of, rather than learn, the forward image model such that the learning can be focused on the image or prior model and acceleration. This has led to good reconstruction results where Poisson noise and metal artifacts are greatly reduced.

The overall goal of this work is to develop a rapid, accurate and fully automated software tool to estimate patient-specific organ doses from CT scans using a deterministic Linear Boltzmann Transport Equation (LBTE) solver (Acuros CTD) and deep-learning CT segmentation algorithms. This study evaluated the accuracy of deep learning segmentation for estimating organ dose from dose maps generated by Acuros. The study focused on pediatric CT due to increased radiation concerns and segmentation challenges for the pediatric population. Organs relevant to CT dosimetry were manually contoured by experts in 246 pediatric chest-abdomen-pelvis CT datasets to serve as ground truth. A fully convolutional network based on a modified V-net architecture was trained and tuned using 226 pediatric datasets ranging in age from 1 to 16 years. An additional twenty datasets were used for preliminary evaluation. The accuracy of organ dose estimates obtained from deep learning segmentation was evaluated relative to doses obtained from the ground truth contours. The deep learning segmentation algorithm resulted in low dose errors for all organs, with a mean absolute error across test patients of 1% or less and a maximum error of 3.5% for the heart. There was high similarity between deep learning and expert contours, with mean Dice coefficients across patients greater than or equal to 0.95. There was no correlation between organ dose error or Dice coefficient with the patient age. Based on statistical analysis of students paired T-test, there was no statistically significant difference between organ doses estimated using the deep learning contours as compared to the expert ground truth contours (p>0.2). Overall, the deep learning segmentation models applied to dose maps generated by the LBTE solver (Acuros CTD) resulted in high organ dose accuracy. Additional evaluation is planned for more organ structures and patient datasets.

We developed a machine-learning-based method generate good quality low dose CT using a residual block concept and a self-attention strategy with a cycle-consistent adversarial network framework. A fully convolution neural network with residual blocks and attention gates is used in the generator to enable end-to-end transformation. We have collected CT images from 30 patients treated with frameless brain stereotactic radiosurgery (SRS) for this study. These full dose images were used to generate projection data, which were then added with noise to simulate the low mAs scanning scenario. Low dose CT images were reconstructed from this noise-contaminated projection data, and were fed into our network along with the original full dose CT images for training. The performance of our network was evaluated by quantitatively comparing the high quality CT images generated by our method with the original full dose images. When mAs is reduced to 0.5% of the original CT scan, the mean square error of the CT images obtained by our method is ~1.6%, with respective to the original full dose images. The proposed method successfully improved the noise, CNR and non-uniformity level to be close to those of full dose CT images, and outperforms a state-of-art iterative reconstruction method. Dosimetric studies shows that the average differences of DVH metrics are less than 0.1 Gy (p>0.05). These quantitative results strongly indicate that the denoised low dose CT images using our method maintains image accuracy and quality, and are accurate enough for dose calculation in current CT simulation of brain SRS treatment. This study also demonstrates the great potential for low dose CT in the process of simulation and treatment planning.

The eye lens is a very radiosensitive organ and is at risk for cataractogenesis during neuro-interventional procedures. It is paramount that the lens is exposed to the x-ray beam as little as possible while still being able to complete the clinical task. In this preliminary investigation, a convolutional neural network (CNN) has been created in order identify if the lens is within the x-ray projection image and where it is located with the intent to facilitate lens dose estimation. The model was trained using a database of patient cases of radiographic skull images, which had different views, in order to generalize the data. The size of the dataset was increased by rotating the images at various angles and masks were created for each corresponding image by hand-contouring the eye socket in the image. For image segmentation, a U-Net model was used which consisted of a down-block, bottleneck, and up-block. Different network parameters were tested and receiver operating curves (ROCs), with Jaccard indices, were assessed to identify the best model. The end goal of this project is model implementation into the real-time Canon Dose Tracking System (DTS) during interventional fluoroscopic procedures. This will allow the DTS to have a more accurate identification of where the lens is, whether fully in the beam or only partially. With this information, a more accurate calculation of the eye lens dose can be done which allows for patients’ dose to be more carefully monitored.

Proliferation of vasa vasorum, the microvasculature within artery walls, is an early marker of atherosclerosis. Detection of subtle changes in the spatial density of vasa vasorum using contrast-enhanced CT is challenging due to the limited spatial resolution and blooming effects. We report a forward model-based blooming correction technique to improve vasa vasorum detection in a porcine model imaged using an ultra-high resolution photon-counting detector CT. Six weeks preceding the CT study the animal received autologous blood injections in its left carotid artery to stimulate vasa vasorum proliferation within the arterial wall (right carotid served as control). The forward model predicted radial extent and magnitude of the luminal blooming affecting the wall signal by using prior data acquired with a vessel phantom of known dimensions. The predicted contamination from blooming was then subtracted from the original wall signal measurement to recover the obscured vasa vasorum signal. Attenuation measurements made on a testing vessel phantom before and after blooming corrections revealed a reduction in mean squared error by ~99.9% when compared to the ground truth. Applying corrections to contrast-enhanced carotid arteries from in vivo scan data demonstrated consistent reductions of blooming contamination within the vessel walls. An unpaired student t-test applied to measurements from the uncorrected porcine scan data revealed no significant difference between the vessel walls (p=0.26). However, after employing blooming correction, the mean enhancement was significantly greater in the injured vessel wall (p=0.0006).

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

Spectroscopic x-ray detectors are under development in academic and industry laboratories, and have been receiving attention for their ability to perform single-shot, multi-material decomposition. However, a number of physical processes, including charge sharing and characteristic emission, cause spectral distortion which results in image degradation. In this study, the effects of the system response of cadmium telluride x-ray detectors on basis-material decomposition were investigated. A spatio-energetic model of the system response of spectroscopic x-ray detectors was developed and incorporated into material decomposition of simulated flat-field images. The spatio-energetic decomposition was compared against two other decomposition methods: one which incorporated the energy response and one which assumed an ideal system response. The results were also compared against the decomposition of images simulated assuming ideal conditions (i.e. no charge sharing or characteristic emission). All decomposition methods investigated here were shown to preserve linearity of the iodine signal with respect to the background. However, inclusion of the non-ideal system response resulted in a 3-fold reduction in SNR. Investigating the effects of the spatio-energetic system response on the spatial resolution of basis-material images will be a focus of future work.

Energy-resolving CT (ErCT) with a photon counting detector (PCD) is able to generate multi-energy data with high spatial resolution, and it can be used to improve contrast-to-noise ratio (CNR) of iodinated tissues and to reduce beam hardening artifacts. In addition, ErCT allows for generating virtual mono-energetic CT images with improved CNR. However, most of ErCT scanners are lab-built, but little used in clinical research. Deep learning based methods can help to generate ErCT images from energy-integrating CT (EiCT) images via convolution neural networks (CNNs) because of its capability in learning features of the EiCT images and ErCT images. Nevertheless, current CNNs usually generate ErCT images at one energy bin at a time, and there is large room for improvement, such as, generating multi-energy ErCT images at a time. Therefore, in this work, we investigate to leverage a deep generative model (IuGAN-ErCT) to simultaneously generate ErCT images at multiple energy bins from existing EiCT images. Specifically, a unified generative adversarial network (GAN) is employed. With a single generator, the generative network learns the latent correlation between the EiCT images and ErCT images to estimate ErCT images from EiCT images. Moreover, to maintain the value accuracy of different ErCT images, we introduced a fidelity loss function. In the experiment, 1384 abdomen and chest images collected from 22 patients were utilized to train the proposed IuGAN-ErCT method and 130 slices were used for test. Result shows that the IuGAN-ErCT method can generate more accurate ErCT images than the uGAN-ErCT method both in quantitative and qualitative evaluation.

We developed a simple method of many (more than 2) material decomposition by combining 2-material decomposition by using energy information obtained with Photon Counting CT. We started from a theoretical approach and led the equation of the attenuation caused by 3 basis materials. As a result, we confirmed that the equations were degenerated when we used the same combination of the energy ranges among multiple 2-material decompositions, and the right lengths of individual basis materials cannot be obtained. On the other hand, we found that a set of equations can be solved when we used different combinations. To demonstrate our method, we simulated how 3-material decomposition (basis materials: water, iodine and calcium) is carried by 2-material decompositions with the same energy data and the different energy data. As a result, we cannot obtain proper solutions in the case of the same energy ranges as found in the theoretical approach, and we confirmed that the right lengths were provided in the case of the different energy ranges. In conclusion, we demonstrated that 3- material decomposition can be implemented by combining multiple 2-material decompositions by using different sets of the energy ranges.

Implemented with energy-integration x-ray detector, dual energy spectral CT has been playing increasingly important roles in the clinic for diagnostic imaging. Encouraged by the clinical value added by dual energy spectral CT in oncological, cardiovascular and neurovascular applications, the technological community is investing more resource and effort on photon-counting spectral CT that is anticipated to outperform its energy-integration counterpart in many advanced clinical applications. Based on the preliminary data of phantom and animal studies acquired using a prototype photoncounting spectral CT system, we revisit the underlying physics, mathematics, dimensionality of material space and selection of basis materials for spectral imaging. Focused on three-material decomposition for virtual monochromatic imaging/analysis, we investigate the feasibility and performance of spectral imaging without contrast agents in photoncounting CT, along with an in-depth analysis of the pros and cons of three-material decomposition over two-material decomposition for spectral imaging.

With technological advancement in x-ray photon-counting detection, the research and development in photon-counting spectral CT is gaining momentum towards clinical applications. Compared to its energy-integration counterpart, the photon-counting spectral CT is of freedom in energy binning to support three-material (or multiple-material) decomposition for spectral imaging without contrast agent, in which the selection of basis materials and energy binning (spectral channelization), including the cutting-off at low and high energy ends, the number of energy bins, the allocation of bin center and width and the adoption of spectral filtration, plays pivotal roles in determining the imaging performance. Via phantom studies carried out with a photon-counting spectral CT simulation software package, we investigate the potential performance, including contrast, noise and contrast-to-noise ratio (CNR), of three-material decomposition for spectral imaging (mainly virtual monochromatic imaging/analysis) without contrast agents and its variation as functions over base material selection and energy binning, with emphasis on soft tissue differentiation towards clinical and preclinical applications. The preliminary data show that, given identical x-ray dose, the three-material decomposition based virtual monochromatic imaging/analysis can outperform its two-material decomposition based counterpart in the best case CNR and in the scenarios while the energy goes down to approach 18 keV.

X-ray photon counting is a novel imaging technology gaining increasing academic and industrial interest due to its potential for high-resolution computed tomography (CT) and spectral CT. These photon-counting detector systems could have numerous benefits in medical imaging because they provide a substantial increase in the amount of available raw imaging information (e.g., multi-energy information). However, new algorithms are needed for using this information to increase the accuracy of diagnostic decisions. In this work, we focused on addressing technical challenges to consider when designing and building a laboratory benchtop for photon counting CT systems. We developed a photon counting cone-beam CT system containing a photon counting detector with a small detective area based on a translate-rotate geometry and a step-and-shoot acquisition mode. To control and synchronize the CT system components, we developed a desktop computer program with a user-friendly graphical user interface. Furthermore, we established a fully automated method for estimating some of the parameters that describe the geometry of the cone-beam CT system. This method is based on an iterative optimization technique in which the reconstructed image of a small sphere is evaluated to find the required geometrical parameters for better image reconstruction. Finally, we used images reconstructed from different scans to confirm the methodology used in the study.

Phase-contrast X-ray provides attenuation, phase-shift and small-angle-scatter in tissue in same scan yielding multi-contrast information about object, which has greatly benefitted breast-imaging, pre-clinical lung-imaging and bone imaging. A primary barrier for clinical adaptation of interferometric X-ray/CT for torso imaging is the manufacturing difficulty of large gratings. Large-gratings have to be stitched from smaller elements introducing errors such as gaps, errors in pitch, phase-jumps, tilts, causing imaging artifacts. Removing these artifacts will be an advancement towards clinical adaptability this multi-contrast modality. In this work we focus on the Talbot-Lau X-ray Interferometer and investigate effects of different grating defects in 1-D simulations. The grating spot defects include gaps, pitch errors, phase-height errors. We quantify the sum-squared error in reconstructed phase for different types of defects, showing most egregious artifacts for the pitch-errors. We developed two artifact correction methods in interference fringe patterns (i.e. before reconstruction) – an analytical and a neutral network approach. The analytical method (SWFT) uses Short- Window-Fourier-Transform to estimate the local phase-shift and attenuation due to the defect in the blank scans and then applies the correction for the with-objects scans. We also proposed a Regression Convolution Neural Network (R-CNN) to learn these errors and correct for them. Distinct sets of pitch artifacts were used each for training (300 datasets) and testing (300 datasets) with variety of levels of severity of artifacts for three different objects – sphere, ramp and slab. The algorithms performed well, reducing the artifacts from initial average normalized-mean-squared-error of 44.7% to 6.3% for SWFT and 7% for SWFT+R-CNN.

Analyzer-based Imaging (ABI) belongs to a broader family of phase-contrast (PC) X-ray techniques. PC measures X-ray deflection phenomena when interacting with tissue, which is known to provide higher contrast images of soft tissue than other X-ray methods. This has a very important interest in the medical field, specially for mammogram applications. This paper presents a simulation study to compare the sensitivity of filtered back projection of absorption based imaging (implemented by FDK - a method usually used in the conventional X-ray computed tomography) and the phase contrast tomographic imaging. These results are based on previously proposed simulation tool for a table-top ABI systems using a conventional polychromatic X-ray source. In our experiments we evaluated images obtained under different errors in reconstruction parameters namely: the detector’s offset in position, the distance source to the detector (DSD) or source to the object (DSO).

Propagation-based phase-contrast X-ray imaging is the earliest developed phase-contrast X-ray imaging method, however, it requires mathematically intensive algorithms to retrieve phase information. For quantitative analysis, an algorithm called phase retrieval must be applied to the phase-contrast image to retrieve the phase information because the phase and the attenuation coefficient are encoded in the intensity values obtained by the detector. Phase retrieval is a nonlinear inverse technique used to estimate the object X-ray phase shift, thickness or electron density. Many of these phase retrieval methods have been developed assuming a monochromatic X-ray source, although some have been reported to work for a polychromatic X-ray source with some modifications. In this work, we compare seven reported phase retrieval methods for polychromatic sources using a weighted average to calculate the wavelength dependent parameters. Six of the methods compared are single distance approaches and one is an iterative approach that requires an absorption and phase-contrast image. These seven phase retrieval methods are compared for varying object thicknesses in the presence of a polychromatic source in simulation of different materials. The materials investigated in this work are polytetra uorethylene, polystyrene and Kapton. Overall, Paganin's method performed with the lowest relative error for all materials when a polychromatic source is applied to object thicknesses less than 400 microns.

The goal of this study was to perform a characterization study for an in-line phase contrast x-ray imaging prototype with a mid-energy source. Compared to similar prototypes that use high energies, the mid-energy system offers better balancing between the attenuation and phase induced contrasts. An inline phase sensitive prototype acquired all images for this study. The prototype utilizes a microfocus x-ray tube and a flat panel detector, aligned on an optical rail. The source-to-object distance (SOD) was set to 68.58cm while the source-to-image distance (SID) was set to 150.876cm for a magnification of 𝑀 = 2.2. The modulation transfer function (MTF), noise power spectrum (NPS), and detective quantum efficiency (DQE) were calculated for the prototype with source potentials of 60, 90, and 120kV. The oversampled MTF was calculated for each setting. NPS experiments were conducted with a virtual detector set at the SOD. Exposure at the SID was approximately the same for all NPS experiments. The 90 and 120kVp beams were directed through a 2.5mm Al filter, while 60kVp beams were sent through a 1.2mm Al filter. Results indicate that 60kV imaging yields lower amplitude noise than high energy imaging, while maintaining the same resolving power. The cutoff frequency for each source potential was approximately 14 line pairs per mm (lp/mm). The DQE(0) for 60kV, 90kV, and 120kV were 0.757, 0.564, and 0.571 respectively. The study confirmed the hypothesis that 60kVp phase sensitive x-ray imaging yields a higher DQE than those found at 90 and 120kVp.

X-ray phase imaging can significantly improve image contrast in soft-tissue imaging applications such as mammography where the inherent contrast in conventional attenuation-based imaging is low. A system that employs a stainless steel wire mesh to produce a high-contrast structured illumination pattern reduces the need for source coherence and complex alignments. Phase is reconstructed from distortions in this pattern due to phase-based x-ray deflection. A computational model of this system has been developed that allows exploration of design parameters including source size, mesh period, sample structure, detector pixel size and locations of these system elements. Simulation results are presented and compared with experiment for validation.

X-ray phase imaging has found limited clinical use due to requirements on x-ray coherence that may not be easily translated to clinical practice. Instead, this work employs a conventional source to create structured illumination with a simple wire mesh. A mesh-shifting algorithm that incorporates deconvolution of the source spot width is used to enhance resolution. Polycapillary optics are employed to enhance coherence. The effects of incorporating optics with two different focal lengths are compared. Measurements of fat embedded artificial tumors have been performed.

Purpose: Current dedicated cone-beam breast CT (CBBCT) systems typically adopt a pendant geometry for patient comfort and mechanical compatibility with a breast biopsy device. In this work, we design and construct a prototype system of upright CBBCT with a compact size, such that mammography devices can be readily replaced in current clinical rooms. The system performance on spatial resolution, CT accuracy and field-of-view (FOV) size are evaluated via preliminary phantom studies.

Materials and Methods: The prototype system consists of a mono-block x-ray source, a flat panel detector, a slip ring for signal transmission, a servo motor and a rotating gantry. The relative positions and angles of x-ray source and detector are carefully designed to ensure a large FOV with almost zero dead space on the chest side. The slip ring is used to accommodate a breast stretching device designed to stretch the breast length to match the same imaging volume size as in a prone gesture.

Results: The designed prototype system has a size of 1950mm, 1200mm, and 660mm in height, length and width, respectively. On the Catphan®600 phantom, the acquired CBBCT images have an average CT number error of less than 0.5% and an image non-uniformity of 0.19%. An in-house water phantom with inserted thin tungsten wires is designed to evaluate the system capability on imaging small calcifications. The results show that our system successfully images small high-contrast objects with a diameter of 30 μm.

Conclusion: We have developed a prototype upright CBBCT system with a compact size, a high imaging accuracy and spatial resolution.

Simultaneous digital breast tomosynthesis (DBT) and mechanical imaging (MI) offer the potential to combine anatomic information from DBT with functional information from MI. This makes it possible to associate tissue stiffness with specific anatomic structures in the breast, a combination that can reduce false-positive findings by using the MI data to discriminate between ambiguous lesions in DBT. This, in turn, will reduce the frequency of negative biopsies. Simultaneous imaging requires that the MI sensor array be present during DBT acquisition. This introduces artifacts, since the sensor is attenuating. Previously, we demonstrated that the DBT reconstruction could be modified to reduce sensor conspicuity in DBT images. In this paper, we characterize the relative attenuation of the breast and the sensor, to calculate the artifact reduction in DBT reconstruction. We concentrate on pre-processing DBT projections prior to reconstruction. Using commercially available a DBT system, we have confirmed that the sensor array does not completely attenuate the x-rays. This suggests that a pre-processing method based upon flat fielding can be used to reduce artifacts. In a proof-ofconcept study, we performed flat fielding by combining DBT projections of the MI sensor with and without an anthropomorphic breast phantom. Visual evaluation confirmed substantially improved image quality. The artifacts were reduced throughout the image for all sensor elements. Few residual artifacts are noticeable where the phantom thickness decreases. The investigation of additional pre-processing, including beam hardening correction is ongoing. Future work includes quantitative validation, noise stabilization, and method optimization in virtual clinical trials and subsequent patient studies.

In this study, two digital breast tomosynthesis (DBT) systems were evaluated: Siemens Mammomat Inspiration TOMO (Siemens Healthineers, Erlangen, Germany) and GE Senographe Pristina (GE, Buc, France). Along with differences such as angular range and detectors type, the systems use different reconstruction algorithms. One was available for the GE system, based on iterative reconstruction (IR). Two algorithms were available for the Siemens system: TOMO_STANDARD, using filtered back projection (FBP) and EMPIRE, FBP with statistically based artifact reduction. Two commercially available DBT phantoms (CIRS model 020 & 021), with heterogeneous and homogenous background respectively, were used to calculate signal-difference-to-noise-ratio (SDNR) in key structures for varying phantom thickness (30, 45 & 70 mm) and average glandular dose (AGD). Key phantom structures include calcifications and lesion masses of different sizes. Results show a positive correlation between SDNR and AGD except for the EMPIRE algorithm where there was a negative SDNR/AGD trend for one of the microcalcification specks in the heterogeneous phantom. The highest overall SDNR was acquired using the EMPIRE algorithm. Both systems are well within the recommended dose limits but could increase their dose levels in order to achieve higher SDNR. This indicates that there may be room for dose optimization in DBT systems used in screening programs, confirming the importance of continuous evaluation and optimization.

We are developing a platform for virtual (in silico) clinical trials in x-ray breast imaging and dosimetry fully based on Monte Carlo simulations and which adopts patient-like digital phantoms. We report the initial steps and first results of this project. Our collaboration was established for producing a “user-friendly” computational platform which reproduces both 2D (full-fields digital mammography) and 3D (digital breast tomosynthesis, CT dedicated to the breast) breast imaging examinations for technique optimization and for developing protocols and computed aided diagnostic applications. Dose optimization strategies and technique performance comparisons will be explored. A dataset of anatomically realistic, 2D and 3D digital breast phantoms have been produced by means of voxel classification and digital compression of clinical breast CT scans acquired at UC Davis. Monte Carlo simulations, based on the Geant4 toolkit, have been developed for insilico FFDM, DBT and BCT examinations.

The detection of microcalcification on mammograms is known as the most important feature. Microcalcifications are difficult to distinguish because they have low contrast in mammography due to size and breast density. The purpose of this study was to evaluate the feasibility of virtual monochromatic image (VMI) for quantitative assessment of the appropriate energy region for detection of malignant microcalcifications. The photon-counting spectral mammography system was modeled using Geant4 Application for Tomographic Emission (GATE) simulation tools. The breast phantom used a 50/50 ratio of adipose/glandular tissue and microcalcifications used calcium hydroxyapatite (Ca₅(PO₄)₃(OH)), which is mainly malignant. Microcalcifications with various sizes ranging from 150 μm to 550 μm were embedded into the breast phantom. In this study, projection based VMI was used. This study quantitatively evaluates the appropriate energy region in terms of image quality using VMI technique. The results showed that VM images were optimized at an energy range of approximately 26 to 27 keV. In order to verify the usefulness of the results obtained from the VM images, the CNR was evaluated according to the microcalcification size using the bin images obtained by setting various energy thresholds based on the photon-counting detector. Compared to the results of VM images, the results of bin images showed a similar tendency. In this study, we investigated the optimum energy of monochromatic images for breast diagnostic applications. By setting the optimal energy range using VMI, we can identify microcalcifications better in mammography and expect to reduce the frequency of additional examination.

The coherent nature of X-ray scattering at small angles produce scattering interference which generates a unique energy spectrum characteristic of the intra-molecular structure of the irradiated sample. The phase difference introduced by the scattering centers will depend upon the wavelength of the radiation and the scattering angle. The resulting interference effects can be studied very effectively using a mono energetic source and scanning an angular range. However, when using a poly energetic source the interference happen at different energies simultaneously, which generates a much more complex and obscured interference pattern due to the spread of wavelength. The spectral pattern still contains valuable information about the electron density distribution of materials of interest. In this study we propose the use of spectral xray measurements with an advanced photon counting detector Timepix3 (300μm Si) as well as an X-123 (1mm CdTe) Amptek spectrometer to measure the small angle x-ray scattering generated by samples illuminated by a polychromatic X-ray source. We show that the resultant scatter interference can be effectively characterized with this setting. In this study, we also examine the photon counting profiles obtained from different metals and plastics as well as normal and malignant breast tissue. The measured photon counting spectrum for different materials shows reproducible features. We were able to characterize benign and malignant breast tissue using spectral scatter signature. Momentum transfer when scattered from benign breast tissue is observed to be lower in comparison to that from carcinoma. These features could be used for reliable tissue discrimination and material identification. Future work will assess the potential and limitations of this method in improving mammographic diagnostic capabilities.

Based on the breast imaging and reporting data system (BI-RADS) for mammography (MMG) and types of cancer cells detected, a patient is listed into various categories which determine whether they should undergo biopsy or not. Generally, patients under the BI-RADS category 4 or 5 have to go through surgery. During the surgery, a pathological examination is performed with the help of a microscope and additional X-ray images of the removed tissue or breast specimen are taken to determine the positive and negative surgical margins. Although the pathological examination is the best way to determine carcinoma at the inked margin, it consumes a significant amount of time and makes the duration of the surgery longer. In this study, we propose the open-type carbon nanotube (CNT)-based X-ray system, which can be helpful to determine the carcinoma on breast specimen during breast surgery. The technique proposed in this study successfully obtained X-ray images of a breast specimen with visibly clear cancer masses. These results could pave the way for efficient determination of surgical margins by eliminating the time-consuming histological procedures.

Purpose: To compare the visibility of microcalcifications and image quality of specimen mammograms obtained with the digital magnification algorithm(TruviewMAG) view and standard magnification view

Materials and Methods: Nineteen women (median age, 49 years; ranges, 38-71 years) who underwent stereotactic biopsy or ultrasound-guided vacuum-assisted biopsy for suspicious microcalcifications detected at mammography were prospectively enrolled. After biopsy, specimen mammograms were obtained to confirm the removal of microcalcifications. A pair of specimen mammogram were obtained with both TruviewMAG view (DRTECH Corp., Seongnam, Korea) and standard magnification view. Four blinded readers independently evaluated visibility of microcalcifications, contrast and resolution of specimen mammograms using a 10-point scale (1, poor; 10, excellent), and preference. In addition to specimen mammograms, we performed pilot clinical test to evaluate the visibility of microcalcifications in two patients.

Results: The visibility of microcalcifications on specimen mammograms obtained with TruviewMAG view was comparable to those obtained with standard magnification view in overall reading (mean, 8.2±1.8 vs. 8.1±1.4, P = .893). The contrast of specimen mammograms obtained with TruviewMAG view was comparable to those obtained with standard magnification view in overall reading (mean, 7.8±1.6 vs. 7.8±1.4, P = .827). The resolution of specimen mammograms obtained with TruviewMAG view was comparable to those obtained with standard magnification view in overall reading (mean, 7.7±1.7 vs. 7.5±1.3, P = .173). Four readers preferred TruviewMAG view and standard magnification view at equal rates. The mean entrance surface dose for TruviewMAG view was 2.88mGy, which was 58.3% reduction compared to 4.94 mGy in standard magnification view. The visibility of microcalcifications were similar in two patients (one with benign fibrocytic change, and the other with invasive ductal carcinoma)

Conclusion: The visibility of microcalcifications and image quality of the specimen mammograms obtained with TruviewMAG view was comparable with those obtained with standard magnification view. In clinical patients data, microcalcifications were well visualized on TruviewMAG view, even though prospective study with larger study population is needed to confirm our finding.

The purpose of this work is to introduce a new analytical inversion method for three-dimensional (3D) sub-surface imaging beneath the skin from time sequenced infrared (IR) thermography images. The work was motivated by advances in transient thermal nondestructive evaluation methods. Using relationships between wave propagation and thermal diffusive propagation, transformation from diffusive propagation into an equivalent wave field is performed. This transformation results in well-defined reflections with time delay proportional to the distance. We apply the algorithm to a dynamic thermogram of the breasts obtained from the Brazilian breast thermal imaging set. Inversion of the raw data reveals intensities that correspond to heat conduction. Equivalent wave field transform (EWFT) serves as a computationally efficient method of extracting depth resolved anatomical and physiological information from skin surface thermogram data.

Purpose. Report advances being made in synthetic mammography applied to carbon nanotube-enabled stationary digital breast tomosynthesis (sDBT). Methods. The potential value of adding Laplacian decomposition, feature-enhancement algorithms, and weighted recombination to the tunable forward-projection steps developed previously to generate synthetic mammograms for sDBT was studied in this phantom-based comparison of sDBT to full field digital mammography (FFDM) and moving-source or conventional DBT. Contrast-to-noise ratio (CNR) and the full-width-at-half-maximum (FWHM) of the signal intensity were used to compare the display of microcalcification and mass features in the FFDM image and the synthetic images generated by sDBT and DBT. These findings guided modifications in the sDBT image processing chain, seeking to maximize the display of clinically-important features in the sDBT-based synthetic image. Results. Decomposing each reconstructed image slice into its high, mid, and low-frequency components yielded images emphasizing a different feature of clinical importance: microcalcifications, masses, and background density. Applying feature-enhancement algorithms to these images followed by weighted recombination during forward projection yielded an sDBT-based synthetic image that displayed masses with a higher CNR than the FFDM image and the synthetic image generated by DBT. Additionally, microcalcifications that could be visualized in all three modalities were displayed with a higher CNR in the synthetic images generated by DBT and sDBT compared to the FFDM image. Conclusion. Adding Laplacian decomposition, feature-enhancement, and weighted recombination steps to the image processing chain that generates a synthetic image from information collected by sDBT improved the display of clinicallyimportant features. Advancing the synthetic mammography capability of sDBT is important, as it will help complete the evolution of this promising technology to a viable clinical tool.

Our previous work showed that digital breast tomosynthesis (DBT) systems are capable of super-resolution, or subpixel resolution relative to the detector. Using a bar pattern phantom, it is possible to demonstrate that there are anisotropies in super-resolution throughout the reconstruction. These anisotropies are lessened in acquisition geometries with narrow spacing between source positions. This paper demonstrates that by re-arranging the source positions in the scan, the anisotropies can be minimized even further. To this end, a theoretical model of the reconstruction of a high-frequency sinusoidal test object was developed from first principles. We modeled the effect of clustering additional source positions around each conventional source position in fine increments (submillimeter). This design can be implemented by rapidly pulsing the source during a continuous sweep of the x-ray tube. It is shown that it is not possible to eliminate the anisotropies in a conventional DBT system with uniformly-spaced source positions, even if the increments of spacing are narrower than those used clinically. However, super-resolution can be achieved everywhere if the source positions are re-arranged in clusters with submillimeter spacing. Our previous work investigated a different approach for optimizing super-resolution through the use of detector motion perpendicular to the breast support. The advantage of introducing rapid source pulsing is that detector motion is no longer required; this mitigates the need for a thick detector housing, which may be cumbersome for patient positioning.

Breast cancer is the most common form of cancer diagnosed with women. To reduce its mortality rate, early diagnosis is important. In the past, this has led to the introduction of national screening programs using mammography. The disadvantages of mammography (application of ionizing radiation and low detection rate in dense breast) resulted in the demand for an alternative. This demand has led to the development of ultrasonic water bath scanning systems. Those systems scan the breast from all sides and aim for reconstructing the acoustic tissue properties from the measured pressure fields by employing among others full-waveform inversion methods. However, full-wave inversion is computationally expensive, especially in 3-D, and scales almost linear with the size of the spatial domain. To reduce the computational load, we propose a method that reduces the size of the spatial computational domain by back-propagating the field measured on the surface of the 3-D scanning geometry to a surface enclosing a reduced volume. To this end, the measured field is first decomposed into spherical Hankel functions with complex coefficients and subsequently redatumed to a new surface closer by the object. The proposed redatuming method is tested successfully for 3-D synthetic examples.

The purpose of this study is to investigate deep learning model observer (DLMO) using classification network for various sizes of mass detection tasks in digital breast tomosynthesis (DBT). Different tube current and sweep angular range acquisition settings were repeated to acquire different DBT images. As a result, a total number of 66 cases of four alternative forced choice (4AFC) reading was undertaken to the human observer and DLMO. The images of spheroidal mass with different sizes of 1.8, 2.3, 3.1, 3.9, 4.7, and 6.3 mm in the target slab of CIRS breast phantom were cropped by 200x200 size for their region of interest (ROI). A percentage of correct responses (𝑃_c) was measured at the end of each human observer test and compared by the accuracy of prediction using DLMO. The results indicated that our proposed DLMO showed the averaged training accuracy 93% using 60 testing datasets after 50 epochs of training using 204 input training datasets. Both the accuracy and loss function in training session of categorical cross-entropy reached a plateau after 35 epochs. The sensitivity and specificity of the testing data from DLMO showed 78% and 98%, respectively, which is comparable to the 𝑃_c, 0.89, on 4AFC test from the human observer. Although the current number of datasets is too small to apply in clinical trials, the authors considered that our DLMO on the phantom study is useful for initial trial of analysis on DBT images.

Mammography, which is the most common screening technique for breast cancer, is known to be less sensitive for women with dense breast tissue. Breast thermography, which records an image from the temperature pattern at the surface of the skin has been used as an adjunct to detect breast pathologies. The working principle of breast thermography relies on the increase in vascularity and metabolic heat generation of breast tumors compared to healthy tissue, which generates an increase in temperature that can be detected by current infrared imaging equipment. In this work the effect of breast density on the temperature pattern of the breast is investigated through finite element simulations. A three-dimensional scan of a real patient was imported into COMSOL Multiphysics along with structures resembling breast lobes, by changing the volume ratio of breast lobes and the whole breast an approximation of different breast densities was simulated. The thermal parameters of the different breast tissues were obtained from the literature and finite element simulations of heat transfer in biological tissue was performed using COMSOL Multiphysics. Results show that the breast density does not impact significantly the temperature pattern of the breast, therefore it can be concluded that thermography does not dependent on breast density.