This PDF file contains the front matter associated with SPIE Proceedings Volume 9033, including the Title Page, Copyright information, Table of Contents, Invited Panel Discussion, and Conference Committee listing.

Contrast enhancement on cardiac CT provides valuable information about myocardial perfusion and methods have been proposed to assess perfusion with static and dynamic acquisitions. There is a lack of knowledge and consensus on the appropriate approach to ensure 1) sufficient diagnostic accuracy for clinical decisions and 2) low radiation doses for patient safety. This work developed a thorough dynamic CT simulation and several accepted blood flow estimation techniques to evaluate the performance of perfusion assessment across a range of acquisition and estimation scenarios. Cardiac CT acquisitions were simulated for a range of flow states (Flow = 0.5, 1, 2, 3 ml/g/min, cardiac output = 3,5,8 L/min). CT acquisitions were simulated with a validated CT simulator incorporating polyenergetic data acquisition and realistic x-ray flux levels for dynamic acquisitions with a range of scenarios including 1, 2, 3 sec sampling for 30 sec with 25, 70, 140 mAs. Images were generated using conventional image reconstruction with additional image-based beam hardening correction to account for iodine content. Time attenuation curves were extracted for multiple regions around the myocardium and used to estimate flow. In total, 2,700 independent realizations of dynamic sequences were generated and multiple MBF estimation methods were applied to each of these. Evaluation of quantitative kinetic modeling yielded blood flow estimates with an root mean square error (RMSE) of ~0.6 ml/g/min averaged across multiple scenarios. Semi-quantitative modeling and qualitative static imaging resulted in significantly more error (RMSE = ~1.2 and ~1.2 ml/min/g respectively). For quantitative methods, dose reduction through reduced temporal sampling or reduced tube current had comparable impact on the MBF estimate fidelity. On average, half dose acquisitions increased the RMSE of estimates by only 18% suggesting that substantial dose reductions can be employed in the context of quantitative myocardial blood flow estimation. In conclusion, quantitative model-based dynamic cardiac CT perfusion assessment is capable of accurately estimating MBF across a range of cardiac outputs and tissue perfusion states, outperforms comparable static perfusion estimates, and is relatively robust to noise and temporal subsampling.

A new motion estimation and compensation method for cardiac computed tomography (CT) was developed. By combining two motion estimation (ME) approaches the proposed method estimates the local and global cardiac motion and then preforms motion compensated reconstruction. The combined motion estimation method has two parts: one is the local motion estimation, which estimates the coronary artery motion by using coronary artery tree tracking and registration; the other one is the global motion estimation, which estimates the entire cardiac motion estimation by image registration. The final cardiac motion is the linear combination of the coronary artery motion and entire cardiac motion the. We use the backproject-then-warp method proposed by Pack et al. to perform motion compensation reconstruction (MCR). The proposed method was evaluated with 5 patient data and improvements in sharpness of both coronary arteries and heart chamber boundaries were obtained.

We investigated the use of an advanced hybrid iterative reconstruction (IR) technique (iDose4, Philips Health- care) for low dose dynamic myocardial CT perfusion (CTP) imaging. A porcine model was created to mimic coronary stenosis through partial occlusion of the left anterior descending (LAD) artery with a balloon catheter. The severity of LAD occlusion was adjusted with FFR measurements. Dynamic CT images were acquired at end-systole (45% R-R) using a multi-detector CT (MDCT) scanner. Various corrections were applied to the acquired scans to reduce motion and imaging artifacts. Absolute myocardial blood flow (MBF) was computed with a deconvolution-based approach using singular value decomposition (SVD). We compared a high and a low dose radiation protocol corresponding to two different tube-voltage/tube-current combinations (80kV p/100mAs and 120kV p/150mAs). The corresponding radiation doses for these protocols are 7.8mSv and 34.3mSV , respectively. The images were reconstructed using conventional FBP and three noise-reduction strengths of the IR method, iDose. Flow contrast-to-noise ratio, CNR_f, as obtained from MBF maps, was used to quantitatively evaluate the effect of reconstruction on contrast between normal and ischemic myocardial tissue. Preliminary results showed that the use of iDose to reconstruct low dose images provide better or comparable CNR_f to that of high dose images reconstructed with FBP, suggesting significant dose savings. CNR_f was improved with the three used levels of iDose compared to FBP for both protocols. When using the entire 4D dynamic sequence for MBF computation, a 77% dose reduction was achieved, while considering only half the scans (i.e., every other heart cycle) allowed even further dose reduction while maintaining relatively higher CNR_f.

Purpose: To examine the potential for dose reduction in chest CT studies where lesion volume is the primary output (e.g. in therapy-monitoring applications). Methods: We added noise to the raw sinogram data from 15 chest exams with lung lesions to simulate a series of reduced-dose scans for each patient. We reconstructed the reduced-dose data on the clinical workstation and imported the resulting image series into our quantitative imaging database for lesion contouring. One reader contoured the lesions (one per patient) at the clinical reference dose (100%) and 8 simulated fractions of the clinical dose (50, 25, 15, 10, 7, 5, 4, and 3%). Dose fractions were hidden from the reader to reduce bias. We compared clinical and reduced-dose volumes in terms of bias error and variability (4x the standard deviation of the percent differences). Results: Averaging over all lesions, the bias error ranged from -0.6% to 10.6%. Variability ranged from 92% at 3% of clinical dose to 54% at 50% of clinical dose. Averaging over only the smaller lesions (<1cm equivalent diameter), bias error ranged from -9.2% to 14.1% and variability ranged from 125% at 3% dose to 33.9% at 50% dose. Conclusions: The reader’s variability decreased with dose, especially for smaller lesions. However, these preliminary results are limited by potential recall bias, a small patient cohort, and an overly-simplified task. Therapy monitoring often involves checking for new lesions, which may influence the reader’s clinical dose threshold for acceptable performance.

In recent years, iterative algorithms have become popular in diagnostic CT imaging to reduce noise or radiation dose to the patient. The non-linear nature of these algorithms leads to non-linearities in the imaging chain. However, the methods to assess the performance of CT imaging systems were developed assuming the linear process of filtered backprojection (FBP). Those methods may not be suitable any longer when applied to non-linear systems. In order to evaluate the imaging performance, a phantom is typically scanned and the image quality is measured using various indices. For reasons of practicality, cost, and durability, those phantoms often consist of simple water containers with uniform cylinder inserts. However, these phantoms do not represent the rich structure and patterns of real tissue accurately. As a result, the measured image quality or detectability performance for lesions may not reflect the performance on clinical images. The discrepancy between estimated and real performance may be even larger for iterative methods which sometimes produce “plastic-like”, patchy images with homogeneous patterns. Consequently, more realistic phantoms should be used to assess the performance of iterative algorithms. We designed and constructed a biological phantom consisting of porcine organs and tissue that models a human abdomen, including liver lesions. We scanned the phantom on a clinical CT scanner and compared basic image quality indices between filtered backprojection and an iterative reconstruction algorithm.

Iterative image reconstruction methods have been proposed in computed tomography to address two major challenges: one is to reduce radiation dose while maintaining image quality and the other is to reconstruct diagnostic quality images from angularly sparse projection datasets. A variety of regularization models have been introduced in these iterative image reconstruction methods to incorporate the desired image features. To address the sparse view angle image reconstruction problem in four-dimensional cone-beam CT (4DCBCT), Prior Image Constrained Compressed Sensing (PICCS) was proposed. In the past in 4DCBCT, as well as other applications of the PICCS algorithm, the PICCS regularization was formulated using the 1 norm as the means to promote image sparsity. The 1 norm in the objective function is not differentiable and thus may pose challenges in numerical implementations. When the norm deviates from 1.0, the differentiability of the objective function improves, however, the imaging performance may degrade in image reconstruction from sparse datasets. In this paper, we study how the performance of PICCS-4DCBCT changes with norm selection and whether the introduction of a reweighted scheme in relaxed norm PICCS reconstruction helps improve the imaging performance.

The aim of this work is to implement and evaluate a 3D image-based approach for the estimation of scattered radiation in multi-detector CT. Based on a reconstructed CT image volume, the scattered radiation contribution is calculated in 3D fan-beam geometry in the framework of an extended point-scatter kernel (PSK) model of scattered radiation. The PSK model is based on the calculation of elemental scatter contributions propagating the rays from the focal spot to the detector across the object for defined interaction points on a 3D fan beam grid. Each interaction point in 3D leads to an individual elemental 2D scatter distribution on the detector. The sum of all elemental contributions represents the total scatter intensity distribution on the detector. Our proposed extended PSK depends on the scattering angle (defined by the interaction point and the considered detector channel) and the line integral between the interaction point on a 3D fan beam ray and the intersection of the same ray with the detector. The PSK comprises single- and multiple scattering as well as the angular selectivity characteristics of the anti-scatter grid on detector. Our point-scatter kernels were obtained from a low-noise Monte-Carlo simulation of water-equivalent spheres with different radii for a particular CT scanner geometry. The model allows obtaining noise-free scatter intensity distribution estimates with a lower computational load compared to Monte-Carlo methods. In this work, we give a description of the algorithm and the proposed PSK. Furthermore, we compare resulting scatter intensity distributions (obtained for numerical phantoms) to Monte-Carlo results.

X-ray phase-contrast (XPC) imaging methods are well-suited for lung imaging applications due to the weakly absorbing nature of lung tissue and the strong refractive effects associated with tissue-air interfaces. Until recently, XPC lung imaging had only been accomplished at synchrotron facilities. In this work, we investigate the manifestation of speckle in propagation-based XPC images of mouse lungs acquired in situ by use of a benchtop imager. The key contributions of the work are: a) the demonstration that lung speckle can be observed by use of a benchtop XPC imaging system employing a polychromatic tube-source; and b) a systematic experimental investigation of how the texture of the speckle pattern depends on the parameters of the imaging system. Our analyses consists of image texture characterization based on the statistical properties of pixel intensity distributions. Results show how image texture measures of lung regions are strongly dependent on imaging system parameters associated with XPC sensitivity.

Grating-based x-ray differential phase contrast imaging (DPCI) often uses a phase stepping procedure that involves sequential grating motion and multiple x-ray exposures to obtain x-ray phase information. Such a data acquisition process breaks the continuous data acquisition into several step-and-shoot data acquisition sequences. Between two neighboring x-ray pulses, the acquisition will have to be stopped for the grating to translate into the next phase stepping position. This setup also requires that the grating not be fixed. If the gratings are to be mounted onto a fast-rotating gantry (such as those used in x-ray CT), this translation of the grating would add another potential source of mechanical instability. To accelerate the data acquisition speed and improve the mechanical stability of of DPCI data acquisitions, a new grating design was developed. In this method, one of the gratings used in DPCI was divided into four-row groups, within each group, grating structures have a designed offset with respect to their neighboring rows. This design allows the acquired data from any adjacent four detector rows to be combined in order to retrieve the needed x-ray differential phase information from a single x-ray exposure. Both numerical simulations and initial phantom experiments have demonstrated that the new interferometer design can enable DPCI image acquisitions without this well-known overhead in data acquisition time.

The demands for a large field-of-view (FOV) and the stringent requirements for a stable acquisition geometry rank among the major obstacles for the translation of grating-based, differential phase-contrast techniques from the laboratory to clinical applications. While for state-of-the-art Full-Field-Digital Mammography (FFDM) FOVs of 24 cm x 30 cm are common practice, the specifications for mechanical stability are naturally derived from the detector pixel size which ranges between 50 and 100 μm. However, in grating-based, phasecontrast imaging, the relative placement of the gratings in the interferometer must be guaranteed to within micro-meter precision. In this work we report on first experimental results on a phase-contrast x-ray imaging system based on the Philips MicroDose L30 mammography unit. With the proposed approach we achieve a FOV of about 65 mm x 175 mm by the use of the slit-scanning technique. The demand for mechanical stability on a micrometer scale was relaxed by the specific interferometer design, i.e., a rigid, actuator-free mount of the phase-grating G₁ with respect to the analyzer-grating G₂ onto a common steel frame. The image acquisition and formation processes are described and first phase-contrast images of a test object are presented. A brief discussion of the shortcomings of the current approach is given, including the level of remaining image artifacts and the relatively inefficient usage of the total available x-ray source output.

In this work, we report on the development of an advanced multi-channel (MC) image reconstruction algorithm for grating-based X-ray phase-contrast computed tomography (GB-XPCT). The MC reconstruction method we have developed operates by concurrently, rather than independently as is done conventionally, reconstructing tomographic images of the three object properties (absorption, small-angle scattering, refractive index). By jointly estimating the object properties by use of an appropriately defined penalized weighted least squares (PWLS) estimator, the 2nd order statistical properties of the object property sinograms, including correlations between them, can be fully exploited to improve the variance vs. resolution tradeoff of the reconstructed images as compared to existing methods. Channel-independent regularization strategies are proposed. To solve the MC reconstruction problem, we developed an advanced algorithm based on the proximal point algorithm and the augmented Lagrangian method. By use of experimental and computer-simulation data, we demonstrate that by exploiting inter-channel noise correlations, the MC reconstruction method can improve image quality in GB-XPCT.

We present a development of the laboratory-based implementation of edge-illumination (EI) x-ray phase contrast imaging (XPCI) that simultaneously enables low-dose and high sensitivity. Lab-based EI-XPCI simplifies the set-up with respect to other methods, as it only requires two optical elements, the large pitch of which relaxes the alignment requirements. Albeit in the past it was erroneously assumed that this would reduce the sensitivity, we demonstrate quantitatively that this is not the case. We discuss a system where the pre-sample mask open fraction is smaller than 50%, and a large fraction of the created beamlets hits the apertures in the detector mask. This ensures that the majority of photons traversing the sample are detected i.e. used for image formation, optimizing dose delivery. We show that the sensitivity depends on the dimension of the part of each beamlet hitting the detector apertures, optimized in the system design. We also show that the aperture pitch does not influence the sensitivity. Compared to previous implementations, we only reduced the beamlet fraction hitting the absorbing septa on the detector mask, not the one falling inside the apertures: the same number of x-rays per second is thus detected, i.e. the dose is reduced, but not at the expense of exposure time. We also present an extension of our phase-retrieval algorithm enabling the extraction of ultra-small-angle scattering by means of only one additional frame, with all three frames acquired within dose limits imposed by e.g. clinical mammography, and easy adaptation to lab-based phase-contrast x-ray microscopy implementations.

Unlike conventional x-ray absorption imaging, x-ray differential phase contrast imaging (DPCI) uses a phase retrieval algorithm to obtain x-ray phase information from a group of x-ray intensity measurements. As a result, the noise performance of DPCI is expected to differ from that of x-ray absorption imaging. Given the total number of x-ray photons used in imaging, lower noise variance in estimated phase contrast images suggests superior dose efficiency, which is one of the most desirable feature in x-ray imaging. When an algorithm is used to retrieve the phase information, it is important to understand what the lowest possible noise variance would be and whether the algorithm used to retrieve the phase information yields the lowest possible noise variance. To address these questions for differential phase contrast imaging, we studied the noise performance of DPC imaging using the powerful Cramér-Rao lower bound (CRLB) in statistical signal estimation method. Results demonstrated that the noise variances in DPCI images obtained by the algorithmic phase retrieval are always higher than the CRLB, which implies a possible sub-optimality of current phase estimation method. The results also call for the need to apply statistical signal estimation theory to DPCI in order to further improve its noise performance and dose efficiency.

In x-ray differential phase contrast imaging (DPCI), a simple sinusoidal model is used to describe the measured signal intensities. When there is no noise added to the measurements, one is able to estimate the three unknown parameters, absorption contrast, phase-contrast, and small angle scatter, directly from the signal model. However, for x-ray detections, noise is always present and the noise level increases with a decrease in x-ray flux. When noise is involved, an experimental ensemble average is needed to make the above parameter estimation method accurate. Unfortunately, an ensemble average requires a large number of measurements acquired under identical experimental conditions. Such repetition procedures would significantly prolong the data acquisition time and require additional radiation dose to the image object. Therefore, it is desirable to develop a reliable parameter estimation method from a group of single x-ray signal intensity measurements. This paper addresses this concern by proposing a statistical signal estimation method. This method maximizes the joint probability of measurements in a complete phase stepping period and is able to estimate the three unknowns in DPCI simultaneously. Our numerical simulations demonstrate that the statistical signal estimation method outperformed the current blind estimation method by minimizing the noise variance of each parameter. As a result, this method has the potential to further improve the dose efficiency of DPCI.

In-line x-ray phase-contrast (XPC) tomosynthesis combines the concepts of tomosynthesis and in-line XPC imaging to utilize the advantages of both for biological imaging applications. Tomosynthesis permits reductions in acquisition times compared with conventional tomography scans while in-line XPC imaging provides high contrast and resolution in images of weakly absorbing materials. In this work, we develop an advanced iterative algorithm as an approach for dealing with the incomplete (and often noisy) data inherent to XPC tomosynthesis. We also investigate the depth resolution properties of XPC tomosynthesis and demonstrate that the z-resolution properties of XPC tomosynthesis is superior to that of conventional absorption-based tomosynthesis. More specifically, we find in-plane structures display strong boundary-enhancement while out-of-plane structures do not. This effect can facilitate the identification of in-plane structures.

With modern CT scanners, detection and classification of coronary artery disease has become a routine applica- tion in cardiac CT. It poses a desirable non–invasive alternative to the invasive coronary angiography, which is the current clinical gold standard. However, the accuracy of cardiac CT depends on the spatial resolution of the imaging system. The limited spatial resolution leads to blooming artifacts, arising from hyper–dense calcification deposits in the arterial walls. This blooming leads to an overestimation of the degree of luminal narrowing and to loss of the morphology of the calcified region. We propose an image–based algorithm, which aims at removing the blooming and estimating the correct CT–value and morphology of the calcification. The method is based on the assumption, that each calcification consists of a compact region which has an almost constant density and attenuation. This knowledge is incorporated into an iterative deconvolution algorithm in image space. We quantitatively assess the accuracy of the proposed algorithm on analytically simulated phantom data. Qualita- tive results of clinical patient data are presented as well. In both cases, the proposed method outperforms the compared algorithms. The initial patient data results are promising. However, an ex vivo study has to be done to confirm the quantitative results of the simulation study with real specimen.

Cardiac C-arm computed tomography (CT) imaging using interventional C-arm systems can be applied in various areas of interventional cardiology ranging from structural heart disease and electrophysiology interventions to valve procedures in hybrid operating rooms. In contrast to conventional CT systems, the reconstruction field of view (FOV) of C-arm systems is limited to a region of interest in cone-beam (along the patient axis) and fan-beam (in the transaxial plane) direction. Hence, highly X-ray opaque objects (e.g. cables from the interventional setup) outside the reconstruction field of view, yield streak artifacts in the reconstruction volume. To decrease the impact of these streaks a cable tracking approach on the 2D projection sequences with subsequent interpolation is applied. The proposed approach uses the fact that the projected position of objects outside the reconstruction volume depends strongly on the projection perspective. By tracking candidate points over multiple projections only objects outside the reconstruction volume are segmented in the projections. The method is quantitatively evaluated based on 30 simulated CT data sets. The 3D root mean square deviation to a reference image could be reduced for all cases by an average of 50 % (min 16 %, max 76 %). Image quality improvement is shown for clinical whole heart data sets acquired on an interventional C-arm system.

In digital radiography imaging, the antiscatter grid is usually employed to obtain clear projected images by absorbing scattered x-ray beams. However, due to the shadow of the stationary antiscatter grid, the grid artifact appears in the obtained x-ray image as stripes if a linear grid is considered. In order to alleviate the grid artifact, band-stop filters (BSFs) can be used and may cause annoying ringing artifact especially for metallic materials. In this paper, in order to reduce the ringing artifact while applying BSFs to alleviate the grid artifact, a spatial prefiltering technique, which is called the gradient-reduction filter, is proposed. The gradient-reduction filter can remove steep edges in the obtained x-ray images especially for metallic materials and can prevent the grid- artifact BSFs from yielding serious ringing artifact. The structure of the proposed filter is similar to that of the Crawford noise-reduction filter. However, the nonlinear function in the proposed filter has decreasing slopes as its input increases contrary to the Crawford filter case. Through extensive experiments for real digital x-ray images, which are obtained from a direct digital radiography detector, we can see the superior ringing-artifact reduction performance comparing to the conventional filtering approaches especially when the metallic materials present in or on the patients.

The ability to customize the incident x-ray fluence in CT via beam-shaping filters or mA modulation is known to improve image quality and/or reduce radiation dose. Previous work has shown that complete control of x-ray fluence (ray-by-ray fluence modulation) would further improve dose efficiency. While complete control of fluence is not currently possible, emerging concepts such as dynamic attenuators and inverse-geometry CT allow nearly complete control to be realized. Optimally using ray-by-ray fluence modulation requires solving a very high-dimensional optimization problem. Most optimization techniques fail or only provide approximate solutions. We present efficient algorithms for minimizing mean or peak variance given a fixed dose limit. The reductions in variance can easily be translated to reduction in dose, if the original variance met image quality requirements. For mean variance, a closed form solution is derived. The peak variance problem is recast as iterated, weighted mean variance minimization, and at each iteration it is possible to bound the distance to the optimal solution. We apply our algorithms in simulations of scans of the thorax and abdomen. Peak variance reductions of 45% and 65% are demonstrated in the abdomen and thorax, respectively, compared to a bowtie filter alone. Mean variance shows smaller gains (about 15%).

Flat field calibration methods are commonly used in computed tomography (CT) to correct for system imperfections. Unfortunately, they cannot be applied in energy-resolving CT when using bow-tie filters owing to spectral distortions imprinted by the filter. This work presents a novel semi-analytical calibration method for photon-counting spectral CT systems, which is applicable with a bow-tie filter in place and efficiently compensates pile-up effects at fourfold increased photon flux compared to a previously published method without degradation of image quality. The achieved reduction of the scan time enabled the first K-edge imaging in-vivo. The method employs a calibration measurement with a set of flat sheets of only a single absorber material and utilizes an analytical model to predict the expected photon counts, taking into account factors such as x-ray spectrum and detector response. From the ratios of the measured x-ray intensities and the corresponding simulated photon counts, a look-up table is generated. By use of this look-up table, measured photon-counts can be corrected yielding data in line with the analytical model. The corrected data show low pixel-to-pixel variations and pile-up effects are mitigated. Consequently, operations like material decomposition based on the same analytical model yield accurate results. The method was validated on a experimental spectral CT system equipped with a bow-tie filter in a phantom experiment and an in-vivo animal study. The level of artifacts in the resulting images is considerably lower than in images generated with a previously published method. First in-vivo K-edge images of a rabbit selectively depict vessel occlusion by an ytterbium-based thermoresponsive polymer.

Nearly all reconstruction methods are controlled through various parameter selections. Traditionally, such parameters are used to specify a particular noise and resolution trade-off in the reconstructed image volumes. The introduction of reconstruction methods that incorporate prior image information has demonstrated dramatic improvements in dose utilization and image quality, but has complicated the selection of reconstruction parameters including those associated with balancing information used from prior images with that of the measurement data. While a noise-resolution tradeoff still exists, other potentially detrimental effects are possible with poor prior image parameter values including the possible introduction of false features and the failure to incorporate sufficient prior information to gain any improvements. Traditional parameter selection methods such as heuristics based on similar imaging scenarios are subject to error and suboptimal solutions while exhaustive searches can involve a large number of time-consuming iterative reconstructions. We propose a novel approach that prospectively determines optimal prior image regularization strength to accurately admit specific anatomical changes without performing full iterative reconstructions. This approach leverages analytical approximations to the implicitly defined prior image-based reconstruction solution and predictive metrics used to estimate imaging performance. The proposed method is investigated in phantom experiments and the shift-variance and data-dependence of optimal prior strength is explored. Optimal regularization based on the predictive approach is shown to agree well with traditional exhaustive reconstruction searches, while yielding substantial reductions in computation time. This suggests great potential of the proposed methodology in allowing for prospective patient-, data-, and change-specific customization of prior-image penalty strength to ensure accurate reconstruction of specific anatomical changes.

In CT imaging, a variety of applications exist where reconstructions are SNR and/or resolution limited. However, if the measured data provide redundant information, composite image data with high SNR can be computed. Generally, these composite image volumes will compromise spectral information and/or spatial resolution and/or temporal resolution. This brings us to the idea of transferring the high SNR of the composite image data to low SNR (but high resolution) ‘source’ image data. It was shown that the SNR of CT image data can be improved using iterative reconstruction [1] .We present a novel iterative reconstruction method enabling optimal dose usage of redundant CT measurements of the same body region. The generalized update equation is formulated in image space without further referring to raw data after initial reconstruction of source and composite image data. The update equation consists of a linear combination of the previous update, a correction term constrained by the source data, and a regularization prior initialized by the composite data. The efficiency of the method is demonstrated for different applications: (i) Spectral imaging: we have analysed material decomposition data from dual energy data of our photon counting prototype scanner: the material images can be significantly improved transferring the good noise statistics of the 20 keV threshold image data to each of the material images. (ii) Multi-phase liver imaging: Reconstructions of multi-phase liver data can be optimized by utilizing the noise statistics of combined data from all measured phases (iii) Helical reconstruction with optimized temporal resolution: splitting up reconstruction of redundant helical acquisition data into a short scan reconstruction with Tam window optimizes the temporal resolution The reconstruction of full helical data is then used to optimize the SNR. (iv) Cardiac imaging: the optimal phase image (‘best phase’) can be improved by transferring all applied over radiation into that image. In all these cases, we show that - at constant patient dose - SNR can efficiently be transferred from the composite data to the source data while maintaining spatial, temporal and contrast resolution properties of the source data.

Statistical iterative image reconstruction has become the subject of strong, active research in X-ray computed tomography (CT), primarily because it may allow a significant reduction in dose imparted to the patient. However, developing an algorithm that converges fast while allowing parallelization so as to obtain a product that can be used routinely in the clinic is challenging. In this work, we present a novel algorithm that combines the strength of two popular methods. A preliminary investigation of this algorithm was performed, and strongly encouraging initial results are reported.

Iterative image reconstruction gains more and more interest in clinical routine, as it promises to reduce image noise, to reduce artifacts, or to improve spatial resolution. Among vendors and researchers, however, there is no consensus of how to best achieve these aims. The general approach is to incorporate a priori knowledge into iterative image reconstruction for example by adding additional constraints to the cost function, which penalize strong variations between neighboring voxels. However this approach to regularization in general poses a resolution noise trade–off because the stronger the regularization, and thus the noise reduction, the stronger the loss of spatial resolution and thus loss of anatomical detail. We propose a method which tries to improve this trade-off. One starts with generating basis images, which emphasize certain desired image properties, like high resolution or low noise. The proposed reconstruction algorithm reconstructs voxel–specific weighting coefficients that are applied to combine the basis images. By combining the desired properties of each basis image one can generate an image with lower noise and maintained high contrast resolution thus improving the resolution noise trade–off.

Polychromatic statistical reconstruction algorithms have very high computational demands due to the difficulty of the optimization problems and the large number of spectrum bins. We want to develop a more practical algorithm that has a simpler optimization problem, a faster numerical solver, and requires only a small amount of prior knowledge. In this paper, a modified optimization problem for polychromatic statistical reconstruction algorithms is proposed. The modified optimization problem utilizes the idea of determining scanned materials based on a first pass FBP reconstruction to fix the ratios between photoelectric and Compton scattering components of all image pixels. The reconstruction of a density image is easy to solve by a separable quadratic surrogate algorithm that is also applicable to the multi-material case. In addition, a spectrum binning method is introduced so that the full spectrum information is not required. The energy bins sizes and attenuations are optimized based on the true spectrum and object. With these approximations, the expected line integral values using only a few energy bins are very closed to the true polychromatic values. Thus both the problem size and computational demand caused by the large number of energy bins that are typically used to model a full spectrum are reduced. Simulation showed that three energy bins using the generalized spectrum binning method could provide an accurate approximation of the polychromatic X-ray signals. The average absolute error of the logarithmic detector signal is less than 0.003 for a 120 kVp spectrum. The proposed modified optimization problem and spectrum binning approach can effectively suppress beam hardening artifacts while providing low noise images.

The short-scan Feldkamp David Kress (FDK) method for C-arm CT reconstruction involves a heuristic raybased weighting scheme to handle data redundancies. This scheme is known to be approximate under general circumstances and it often creates low frequency image artifacts in regions away from the central axial plane. Alternative algorithms, such as the one proposed by Defrise and Clack (DC),1 can handle data redundancy in a theoretically exact manner and thus notably improve image quality. The DC algorithm, however, is computationally more complex than FDK, as it requires a shift-variant 2D filtering of the data instead of a efficient 1D filtering. In this paper, a modification of the original DC algorithm is investigated, which applies the efficient FDK filtering scheme whereever possible and the DC filtering scheme only where it is required. This modification leads to a more efficient implementation of the DC algorithm, in which filtering effort can be reduced by up to about 70%, dependent on the specific geometry set-up. This gain in computation speed makes the DC method even more attractive for use in an interventional environment, where fast and interactive X-ray imaging is a crucial requirement.

Statistical Image Reconstruction (SIR) often involves a balance of two requirements: the first requirement is enforcing a minimal difference between the forward projection of the reconstructed image with the measured projection data and the second requirement enforcing some kind of image smoothness, which depends on the specific selection of regularizer, to reduce the noise in the reconstructed image. The needed delicate balance between these two requirements in the numerical implementations often slow down the reconstruction speed due to either a degradation in convergence rate of the algorithm or a degradation of parallellizability of the numerical implementation algorithms. In this work, a general numerical implementation strategy has been proposed to allow the SIR algorithms to be implemented in two decoupled and alternating steps. The first step using SIR without any regularizer which allows for the use of the well-known ordered subset (OS) strategy to accelerate the image reconstruction. The second step solves a denoising problem without involving the data fidelity term. The alternation of these two decoupled steps enable one to perform SIR with both high convergence rate and high parallellizability. The total variation norm of the image has been used as an example of regularizers to illustrate the proposed numerical implementation strategy. Numerical simulations have been performed to validate the proposed algorithm. The noise-spatial resolution tradeoff curve and convergence speed of the algorithm have been investigated and compared against the conventional gradient descent based implementation strategy.

Model-based iterative reconstruction (MBIR) is an emerging technique for several imaging modalities and appli- cations including medical CT, security CT, PET, and microscopy. Its success derives from an ability to preserve image resolution and perceived diagnostic quality under impressively reduced signal level. MBIR typically uses a cost optimization framework that models system geometry, photon statistics, and prior knowledge of the recon- structed volume. The challenge of tomosynthetic geometries is that the inverse problem becomes more ill-posed due to the limited angles, meaning the volumetric image solution is not uniquely determined by the incom- pletely sampled projection data. Furthermore, low signal level conditions introduce additional challenges due to noise. A fundamental strength of MBIR for limited-views and limited-angle is that it provides a framework for constraining the solution consistent with prior knowledge of expected image characteristics. In this study, we analyze through simulation the capability of MBIR with respect to prior modeling components for limited-views, limited-angle digital breast tomosynthesis (DBT) under low dose conditions. A comparison to ground truth phantoms shows that MBIR with regularization achieves a higher level of fidelity and lower level of blurring and streaking artifacts compared to other state of the art iterative reconstructions, especially for high contrast objects. The benefit of contrast preservation along with less artifacts may lead to detectability improvement of microcalcification for more accurate cancer diagnosis.

We design an iterative image reconstruction (IIR) algorithm for enhancing tissue structure contrast. The algorithm takes advantage of a data fidelity term, which compares the derivative of the DBT projections with the derivative of the estimated projections. This derivative data fidelity is sensitive to the edges of tissue structure projections, and as a consequence minimizing the corresponding the data-error term brings out structure information in the reconstructed volumes. The method has the practical advantages that few iterations are required and that direct region-of-interest (ROI) reconstruction is possible with the proposed derivative data fidelity term. Both of these advantages reduce the computational burden of the IIR algorithm and potentially make it feasible for clinical application. The algorithm is demonstrated on clinical DBT data.

Compressed sensing (CS) [1] is a novel sensing (acquisition) paradigm that applies to discrete-to-discrete system models and asserts exact recovery of a sparse signal from far fewer measurements than the number of unknowns [1- 2]. Successful applications of CS may be found in MRI [3, 4] and optical imaging [5]. Sparse reconstruction methods exploiting CS principles have been investigated for CT [6-8] to reduce radiation dose, and to gain imaging speed and image quality in optical imaging [9]. In this work the objective is to investigate the applicability of compressed sensing principles for a faster brain imaging protocol on a hybrid collimator SPECT system. As a proofof- principle we study the null space of the fan-beam collimator component of our system with regards to a particular imaging object. We illustrate the impact of object sparsity on the null space using pixel and Haar wavelet basis functions to represent a piecewise smooth phantom chosen as our object of interest.

Graphical analysis is employed in the research setting to provide quantitative estimation of PET tracer kinetics from dynamic images at a single bed. Recently, we proposed a multi-bed dynamic acquisition framework enabling clinically feasible whole-body parametric PET imaging by employing post-reconstruction parameter estimation. In addition, by incorporating linear Patlak modeling within the system matrix, we enabled direct 4D reconstruction in order to effectively circumvent noise amplification in dynamic whole-body imaging. However, direct 4D Patlak reconstruction exhibits a relatively slow convergence due to the presence of non-sparse spatial correlations in temporal kinetic analysis. In addition, the standard Patlak model does not account for reversible uptake, thus underestimating the influx rate Ki. We have developed a novel whole-body PET parametric reconstruction framework in the STIR platform, a widely employed open-source reconstruction toolkit, a) enabling accelerated convergence of direct 4D multi-bed reconstruction, by employing a nested algorithm to decouple the temporal parameter estimation from the spatial image update process, and b) enhancing the quantitative performance particularly in regions with reversible uptake, by pursuing a non-linear generalized Patlak 4D nested reconstruction algorithm. A set of published kinetic parameters and the XCAT phantom were employed for the simulation of dynamic multi-bed acquisitions. Quantitative analysis on the K_i images demonstrated considerable acceleration in the convergence of the nested 4D whole-body Patlak algorithm. In addition, our simulated and patient whole-body data in the postreconstruction domain indicated the quantitative benefits of our extended generalized Patlak 4D nested reconstruction for tumor diagnosis and treatment response monitoring.

Scatter in cone-beam computed tomography (CBCT) is a significant problem that degrades image contrast, uniformity and CT number accuracy. One means of estimating and correcting for detected scatter is through an iterative deconvolution process known as scatter kernel superposition (SKS). While the SKS approach is efficient, clinically significant errors on the order 2-4% (20-40 HU) still remain. We have previously shown that the kernel method can be improved by perturbing the kernel parameters based on reference data provided by limited Monte Carlo simulations of a first-pass reconstruction. In this work, we replace the Monte Carlo modeling with a deterministic Boltzmann solver (AcurosCTS) to generate the reference scatter data in a dramatically reduced time. In addition, the algorithm is improved so that instead of adjusting kernel parameters, we directly perturb the SKS scatter estimates. Studies were conducted on simulated data and on a large pelvis phantom scanned on a tabletop system. The new method reduced average reconstruction errors (relative to a reference scan) from 2.5% to 1.8%, and significantly improved visualization of low contrast objects. In total, 24 projections were simulated with an AcurosCTS execution time of 22 sec/projection using an 8-core computer. We have ported AcurosCTS to the GPU, and current run-times are approximately 4 sec/projection using two GPU’s running in parallel.

This paper provides a fast and patient-specific scatter artifact correction method for cone-beam computed tomography (CBCT) used in image-guided interventional procedures. Due to increased irradiated volume of interest in CBCT imaging, scatter radiation has increased dramatically compared to 2D imaging, leading to a degradation of image quality. In this study, we propose a scatter artifact correction strategy using an analytical convolution-based model whose free parameters are estimated using a rough estimation of scatter profiles from the acquired cone-beam projections. It was evaluated using Monte Carlo simulations with both monochromatic and polychromatic X-ray sources. The results demonstrated that the proposed method significantly reduced the scatter-induced shading artifacts and recovered CT numbers.

The performance of a novel designed x-ray CT scanning geometry is investigated. Composed of a specially designed tungsten collimation mask and a high resolution flat panel detector, this scanning geometry provides high efficient data acquisition allowing dose reduction potentially up to 50%. In recent years a special type of scanning geometry has been proposed. A first prototype of this geometry called CTDOR( CT with Dual Optimal Reading) has already been built. Despite many drawbacks, resulting images have shown promising potential of dual reading. The approach of gaining two subsets of data has anew been picked up and come to terms with a novel designed CT scanner for breast imaging. The main idea consists of collimating the X-ray beam through a specially designed shielding mask thereby reducing radiation dose without compromising image quality. This is achieved by hexagonally sampled Radon transform and image reconstruction with the especially suitable OPED (orthogonal polynomial expansion on disk) algorithm. This work now presents the development and evaluation of the novel designed breast CT system. Therefore simulated phantom data were obtained to test the performance of the scanning device and compared to a standard 3rd generation scanner. Retaining advantages such as scatter-correction potential and 3D-capability, the proposed CT system yields high resolution images for breast diagnostics in low energy ranges. Assuming similar sample size, it is expected that the novel designed breast CT system in conjunction with OPED outperforms the standard 3rd generation CT system combined with FBP (filtered back projection).

We previously introduced three different fiducial marker-based correction methods (2D projection shifting, 2D projection warping, and 3D image warping) for patients’ involuntary motion in the lower body during weight-bearing Carm CT scanning. The 3D warping method performed better than 2D methods since it could more accurately take into account the lower body motion in 3D. However, as the 3D warping method applies different rotational and translational movement to the reconstructed image for each projection frame, distance-related weightings were slightly twisted and thus result in overlaying background noise over the entire image. In order to suppress background noise and artifacts (e.g. metallic marker-caused streaks), the 3D warping method has been improved by incorporating bilateral filtering and a Landwebertype iteration in one step. A series of projection images of five healthy volunteers standing at various flexion angles were acquired using a C-arm cone-beam CT system with a flat panel. A horizontal scanning trajectory of the C-arm was calibrated to generate projection matrices. Using the projection matrices, the static reference marker coordinates in 3D were estimated and used for the improved 3D warping method. The improved 3D warping method effectively reduced background noise down below the noise level of 2D methods and also eliminated metal-generated streaks. Thus, improved visibility of soft tissue structures (e.g. fat and muscle) was achieved while maintaining sharp edges at bone-tissue interfaces. Any high resolution weight-bearing cone-beam CT system can apply this method for motion compensation.

Projection images from digital breast tomosynthesis acquisitions can contain a large fraction of scattered x-rays due to the absence of an anti-scatter grid in front of the detector. In order to produce quantitative results, this should be accounted for in reconstruction algorithms. We examine the possible improvement in signal difference to noise ratio (SDNR) for low contrast spherical densities when applying a scatter correction algorithm. Hybrid patient data were created by combining real patient data with attenuation profiles of spherical masses acquired with matching exposure settings. Scatter in these cases was estimated using Monte-Carlo based scatter- ing kernels. All cases were reconstructed using filtered backprojection (FBP) with and without beam hardening correction and two maximum likelihood methods for transmission tomography, with and without quadratic smoothing prior (MAPTR and MLTR). For all methods, images were reconstructed without scatter correction, and with scatter precorrection, and for the iterative methods also with an adjusted update step obtained by including scatter in the physics model. SDNR of the inserted spheres was calculated by subtracting the recon- structions with and without inserted template to measure the signal difference, while noise was measured in the image containing the template. SDNR was significantly improved by 3.5% to 4.5% (p < 0.0001) at iteration 10 for both correction methods applied to the MLTR and MAPTR reconstructions. For MLTR these differences disappeared by iteration 100. For regular FBP SDNR remained the same after correction (p = 0.60) while it dropped slightly for FBP with beam hardening correction (-1.4%, p = 0.028). These results indicate that for the iterative methods, application of a scatter correction algorithm has very little effect on the SDNR, it only causes a slight decrease in convergence speed, which is similar for precorrection and correction incorporated in the update step. The FBP results were unchanged because the scatter being corrected is a low frequency component in the projection images, and this information is mostly ignored in the reconstruction due to the high pass filter.

Digital breast tomosynthesis (DBT) is a technique developed to overcome the limitations of conventional digital mammography by reconstructing slices through the breast from projections acquired at different angles. In developing and optimizing DBT, the x-ray scatter reduction technique remains a significant challenge due to projection geometry and radiation dose limitations. The most common approach for scatter reduction technique is a beam-stop-array (BSA) algorithm while this method has a concern of additional exposure to acquire the scatter distribution. The compressed breast is roughly symmetry and the scatter profiles from projection acquired at axially opposite angle are similar to mirror image from each other. The purpose of this study was to apply the BSA algorithm acquiring only two scans with a beam stop array, which estimates scatter distribution with minimum additional exposure. The results of scatter correction with angular interpolation were comparable to those of scatter correction with all scatter distributions at each angle and exposure increase was less than 13%. This study demonstrated the influence of scatter correction by BSA algorithm with minimum exposure which indicates the practical application in clinical situations.

In cone beam computed tomography (CT), it is common practice to use the Defrise phantom for image quality assessment. The phantom consists of a stack of plastic plates with low frequency spacing. Because the x-ray beam may traverse multiple plates, the spacing between plates can appear blurry in the reconstruction, and hence modulation provides a measure of image quality. This study considers the potential merit of using the Defrise phantom in digital breast tomosynthesis (DBT), a modality with a smaller projection range than CT. To this end, a Defrise phantom was constructed and subsequently imaged with a commercial DBT system. It was demonstrated that modulation is dependent on position and orientation in the reconstruction. Modulation is preserved over a broad range of positions along the chest wall if the input frequency is oriented in the tube travel direction. By contrast, modulation is degraded with increasing distance from the chest wall if the input frequency is oriented in the posteroanterior (PA) direction. A theoretical framework was then developed to model these results. Reconstructions were calculated in an acquisition geometry designed to improve modulation. Unlike current geometries in which the x-ray tube motion is restricted to the plane of the chest wall, we consider a geometry with an additional component of tube motion along the PA direction. In simulations, it is shown that the newly proposed geometry improves modulation at positions distal to the chest wall. In conclusion, this study demonstrates that the Defrise phantom is a tool for optimizing DBT systems.

Current digital breast tomosynthesis (DBT) systems have been shown to have diminished microcalcification (MC) visibility compared to 2D mammography systems. Rotating gantry DBT systems require mechanical motion of the X-ray source which causes motion blurring of the focal spot, thus reducing spatial resolution. We have developed a stationary DBT (s-DBT) technology that uses a carbon nanotube (CNT) based X-ray source array in order to acquire all the projections images without any mechanical motion. It is capable of producing full tomosynthesis datasets with zero motion blur. It has been shown to have significantly higher spatial resolution than continuous motion DBT systems. An s-DBT system also allows for a wider angular span without increasing the acquisition time. A larger angular span covers a larger portion of the Fourier domain, thus decreasing the tissue overlap. In this study, we compare tomosynthesis imaging of MCs, in lumpectomy specimens, between an s-DBT system and a rotating gantry DBT system. Results show that s-DBT produces better MC sharpness and reduced tissue overlap compared to continuous motion DBT systems.

We have recently demonstrated the feasibility of stationary digital chest tomosynthesis (s-DCT) using a dis- tributed carbon nanotube x-ray source array. The technology has the potential to increase the imaging resolution and speed by eliminating source motion. In addition, the flexibility in the spatial configuration of the individual sources allows new tomosynthesis imaging geometries beyond the linear scanning mode used in the conventional systems. In this paper, we report the preliminary results on the effects of the tomosynthesis imaging geometry on the image quality. The study was performed using a bench-top s-DCT system consisting of a CNT x-ray source array and a flat-panel detector. System MTF and ASF are used as quantitative measurement of the in-plane and in-depth resolution. In this study geometries with the x-ray sources arranged in linear, square, rectangular and circular configurations were investigated using comparable imaging doses. Anthropomorphic chest phantom images were acquired and reconstructed for image quality assessment. It is found that wider angular coverage results in better in-depth resolution, while the angular span has little impact on the in-plane resolution in the linear geometry. 2D source array imaging geometry leads to a more isotropic in-plane resolution, and better in-depth resolution compared to 1D linear imaging geometry with comparable angular coverage.

Possible advantages of energy-selective photon counting detectors compared to dual energy CT shall be evaluated in the case of a typical dual energy application: Image-based material decomposition into an iodine and a water material image. Apart from a possibly smaller spectral overlap between the low and the high energy information, a photon counting detector will probably offer more than the two necessary energy bins. In this case additional degrees of freedom are gained that allow minimizing the noise in the material images. We propose an image-based method that determines optimal bin image weighting factors for material decomposition with respect to minimal material image noise. Results indicate that a perfect photon counting detector with four bins outperforms the dual energy CT technique by a noise reduction of 22% in the water image and 43% in the iodine image. Limited detector energy resolution has only a small impact on the achieved noise reduction, the photon counting detector performs still better than the dual energy technique. However, if degrading detector effects like charge sharing and K-escape peaks are taken into account, the performance of our proposed method drops below the one of conventional dual energy CT.

We present a fast, noise-efficient, and accurate estimator for material separation using photon-counting x-ray detectors (PCXDs) with multiple energy bin capability. The proposed targeted least squares estimator (TLSE) improves a previously proposed A-Table method by incorporating dynamic weighting that allows noise to be closer to the Cramér- Rao Lower Bound (CRLB) throughout the operating range. We explore Cartesian and average-energy segmentation of the basis material space for TLSE, and show that iso-average-energy contours require fewer segments compared to Cartesian segmentation to achieve similar performance. We compare the iso-average-energy TLSE to other proposed estimators - including the gold standard maximum likelihood estimator (MLE) and the A-Table1 - in terms of variance, bias and computational efficiency. The variance and bias of this estimator between 0 to 6 cm of aluminum and 0 to 50 cm of water is simulated with Monte Carlo methods. Iso-average energy TLSE achieves an average variance within 2% of CRLB, and mean of absolute error of (3.68 ± 0.06) x 10^-6 cm. Using the same protocol, MLE showed variance-to- CRLB ratio and average bias of 1.0186 ± 0.0002 and (3.10 ± 0.06) x 10^-6 cm, respectively, but was 50 times slower in our simulation. Compared to the A-Table method, TLSE gives a more homogenous variance-to-CRLB profile in the operating region. We show that variance-to-CRLB for TLSE is lower by as much as ~36% than A-Table method in the peripheral region of operation (thin or thick objects). The TLSE is a computationally efficient and fast method for implementing material separation technique in PCXDs, with performance parameters comparable to the MLE.

Photon counting detectors used in spectroscopic CT are often based on small pixels and therefore offer only limited space to include energy discriminators and their associated counters in each pixel cell. For this reason, it is important to make efficient use of the available energy discriminators in order to achieve an optimized material contrast at a radiation dose as low as possible. Unfortunately, the complexity of evaluating every possible combination of energy thresholds, given a fixed number of counters, rapidly increases with the resolution at which this search is performed, and makes brute-force approaches to this problem infeasible. In this work, we introduce methods from machine learning, in particular sparse regression, to perform a feature selection to determine optimal combinations of energy thresholds. We will demonstrate how methods enforcing row-sparsity on a linear regression’s coefficient matrix can be applied to the multiple response problem in spectroscopic CT, i.e. the case in which a single set of energy thresholds is sought to simultaneously retrieve concentrations pertaining to a multitude of materials in an optimal way. These methods are applied to CT images experimentally obtained with a Medipix3RX detector operated in charge summing mode and with a CdTe sensor at a pixel pitch of 110μm. We show that the least absolute shrinkage and selection operator (lasso), generalized to the multiple response case, chooses four out of 20 possible threshold positions that allow discriminating PMMA, iodine and gadolinium in a contrast agent phantom at a higher accuracy than with equally spaced thresholds. Finally, we illustrate why it might be unwise to use a higher number of energy thresholds than absolutely necessary.

Spectral CT with photon-counting detectors has the potential to improve material decomposition and contrastto- noise ratio (CNR) compared to conventional CT. This work compared the noise properties of two general energy-bin acquisition methods: (1) energy bins acquired from the same spectrum noise realization, and (2) energy bins acquired from different spectrum noise realizations. For both types of acquisitions, the detected number of counts per bin was simulated and measured on a bench-top system. The energy-bin noise standard deviation was compared for both acquisition methods. Simulations were performed to compare both methods with respect to noise in material decomposition estimates and the CNR in image-based weighted images. Both the experimental and simulation results indicated increased energy-bin noise when energy measurements were acquired from different spectrum realizations. The noise increased by a factor of 2 for the lowest energy bin, with the noise penalty decreasing with increasing bin energy. The simulation results demonstrated a factor of 1.2 to 2 increase in noise in material decomposition estimates when acquiring from different spectrum realizations. Despite the increased energy-bin noise, energy measurements from different spectrum realizations increased the CNR in image-based-weighted images by 10%, potentially due to noise correlations across bins. Overall, the investigated acquisition methods demonstrated differences in noise standard deviation, affecting material decomposition estimates and CNR.

The energy-selectivity of photon counting detectors provides contrast enhancement and enables new material-identification techniques for clinical Computed Tomography (CT). Patient dose considerations and the resulting requirement of efficient X-ray detection suggest the use of CdTe or CdZnTe as detector material. The finite signal pulse duration of several nanoseconds present in those detectors requires strong reduction of the pixel size to achieve feasible count rates in the high-flux regime of modern CT scanners. Residual pulse pile-up effects in scans with high X-ray fluxes still can limit two key properties of the counting detector, namely count-rate linearity and spectral linearity. We have used our research prototype scanner with CdTe-based counting detector and 225μm small pixels to investigate these effects in CT imaging scenarios at elevated X-ray tube currents. We present measurements of CT images and provide a detailed analysis of contrast stability, image noise and multi-energy performance achieved with different phantom sizes at various X-ray tube settings.

We present investigations into direct, calibration-free recovery of distorted spectral x-ray measurements with the Medipix 2 detector. Spectral x-ray measurements using pixelated photon counting spectral x-ray detectors are subject to significant spectral distortion. For detectors with small pixel size, charge sharing between adjacent electrodes often dominates this distortion. In material decomposition applications, a popular spectral recovery technique employs a calibration phantom with known spectral properties. This works due to the similarity of the attenuation properties of the phantom and the material to be studied. However, this approach may be too simplistic for clinical imaging applications as it assumes the homogeneity (and knowledge) of exactly the properties whose variation accounts entirely for the diagnostic content of the spectral data obtained by the photon counting detector. It may also be difficult to find the right calibration phantom for varying patient size and tissue densities on a case-by-case basis. Thus, it is desirable to develop direct correction strategies, based on the objectively measurable response of the detector. We model analytically the distortion of a spectral signal in a PCSXD by applying Gaussian broadening and a charge-sharing model. The model parameters are fitted to the measured fluorescence of several metals. While we are investigating the methodology using Medipix detectors, it should be applicable to other PCXDs as well.

In x-ray imaging, contrast information content varies with photon energy. It is therefore possible to improve image quality by weighting photons according to energy. We have implemented and evaluated so-called energy weighting on a commercially available spectral photon-counting mammography system. A practical formula for calculating the optimal weight from pixel values was derived. Computer simulations and phantom measurements revealed that the contrast-tonoise ratio was improved by 3%–5%, and automatic image analysis showed that the improvement was detectable in a set of screening mammograms.

Spectral X-ray imaging allows to differentiate between two given tissue types, provided their spectral absorption characteristics differ measurably. In mammography, this method is used clinically to determine a decomposition of the breast into adipose and glandular tissue compartments, from which the glandular tissue fraction and, hence, the volumetric breast density (VBD) can be computed. Another potential application of this technique is the characterization of lesions by spectral mammography. In particular, round lesions are relatively easily detected by experienced radiologists, but are often difficult to characterize. Here, a method is described that aims at discriminating cystic from solid lesions directly on a spectral mammogram, obtained with a calibrated spectral mammography system and using a hypothesis-testing algorithm based on a maximum likelihood approach. The method includes a parametric model describing the lesion shape, compression height variations and breast composition. With the maximum likelihood algorithm, the model parameters are estimated separately under the cyst and solid hypothesis. The resulting ratio of the maximum likelihood values is used for the final tissue characterization. Initial results using simulations and phantom measurements are presented.

We have developed a high resolution amorphous selenium (a-Se) direct detection imager using a large-area compatible back-end fabrication process on top of a CMOS active pixel sensor having 25 micron pixel pitch. Integration of a-Se with CMOS technology requires overcoming CMOS/a-Se interfacial strain, which initiates nucleation of crystalline selenium and results in high detector dark currents. A CMOS-compatible polyimide buffer layer was used to planarize the backplane and provide a low stress and thermally stable surface for a-Se. The buffer layer inhibits crystallization and provides detector stability that is not only a performance factor but also critical for favorable long term cost-benefit considerations in the application of CMOS digital x-ray imagers in medical practice. The detector structure is comprised of a polyimide (PI) buffer layer, the a-Se layer, and a gold (Au) top electrode. The PI layer is applied by spin-coating and is patterned using dry etching to open the backplane bond pads for wire bonding. Thermal evaporation is used to deposit the a-Se and Au layers, and the detector is operated in hole collection mode (i.e. a positive bias on the Au top electrode). High resolution a-Se diagnostic systems typically use 70 to 100 μm pixel pitch and have a pre-sampling modulation transfer function (MTF) that is significantly limited by the pixel aperture. Our results confirm that, for a densely integrated 25 μm pixel pitch CMOS array, the MTF approaches the fundamental material limit, i.e. where the MTF begins to be limited by the a-Se material properties and not the pixel aperture. Preliminary images demonstrating high spatial resolution have been obtained from a frst prototype imager.

In order to predict and improve the performance of pixelated detectors, it is important to understand the optical properties of the basic unit of the scintillating structure in the detector. To measure one of the essential optical properties, reflectance, we have used a device composed of a laser and photodiode array. We have also developed an analytical model of the optical phenomena based on Snell's law and the Fresnel equations to simply analyze measured results and reflectance parameters at the interface. The computed and experimentally measured results typically have good agreement, validating the analytical model and measurements. The optical parameters are used as inputs to GEANT4 [1]. The simulations are then leveraged to optimize an imager design before a prototype is built. The optical reflectance was measured by using relatively inexpensive samples. A sample has scintillator, glue, and septum (reflector) layers, and each sample has a different scintillator surface (polished/rough) and/or reflector [ESR film/aluminum-sputtered (coated) ESR film] condition. A high-refractive-index hemisphere was attached on the top surface of a sample to increase the maximum incidence angle at the scintillator-glue interface from 27° to 52°. The sample including ESR film demonstrated average reflectance approximately 1.3 times higher than that from the sample with aluminum-sputtered ESR film as a reflector, and the polished surface condition showed higher reflectance than the rough-cut surface condition.

The thin-film semiconductor processing methods that enabled creation of inexpensive liquid crystal displays based on amorphous silicon transistors for cell phones and televisions, as well as desktop, laptop and mobile computers, also facilitated the development of devices that have become ubiquitous in medical x-ray imaging environments. These devices, called active matrix flat-panel imagers (AMFPIs), measure the integrated signal generated by incident X rays and offer detection areas as large as ~43×43 cm². In recent years, there has been growing interest in medical x-ray imagers that record information from X ray photons on an individual basis. However, such photon counting devices have generally been based on crystalline silicon, a material not inherently suited to the cost-effective manufacture of monolithic devices of a size comparable to that of AMFPIs. Motivated by these considerations, we have developed an initial set of small area prototype arrays using thin-film processing methods and polycrystalline silicon transistors. These prototypes were developed in the spirit of exploring the possibility of creating large area arrays offering single photon counting capabilities and, to our knowledge, are the first photon counting arrays fabricated using thin film techniques. In this paper, the architecture of the prototype pixels is presented and considerations that influenced the design of the pixel circuits, including amplifier noise, TFT performance variations, and minimum feature size, are discussed.