Open Access
Add to BibSonomyAbstract
A wide range of X-ray imaging applications demand micrometer spatial resolution. In material science and biology especially, there is a great interest in material determination and material separation methods. Here we present a new detector design that allows the recording of a low- and a high-energy radiography image simultaneously with micrometer spatial resolution. The detector system is composed of a layered scintillator stack, two CCDs and an optical system to image the scintillator responses onto the CCDs. We used the detector system with a standard laboratory microfocus X-ray tube to prove the working principle of the system and derive important design characteristics. With the recorded and registered dual-energy data set, the material separation and determination could be shown at an X-ray tube peak energy of up to 160 keV with a spatial resolution of 12 μm. The detector design shows a great potential for further development and a wide range of possible applications.
© 2017 Optical Society of America
1. Introduction
X-ray imaging is a widely used non-destructive testing method to produce 2D and 3D images. The demand of micrometer spatial resolution applications is increasing in material science and biology [1]. As the usage of solid-state detectors is not feasible at a resolution in the range of several microns, there are two common designs to perform micro-CT. The conventional design, used in desktop applications, consists of a micro focus X-ray source and a flat panel detector. The geometric magnification, resulting from the cone beam, is then used to achieve a micron resolution of the sample.
In the second approach, mainly developed at synchrotrons, luminescence screens are used together with an optical magnification and focusing systems, as the parallel beam at synchrotrons does not allow for geometric magnification. This well established technique for micron-resolution X-ray imaging uses thin-film single-crystal scintillators for X-ray to light conversion. The achievable resolution can reach close to the diffraction limit of visible light and is mainly limited by the scintillator thickness. Therefore, the thickness must be adapted to the depth of focus (DOF) of the used microscope objective, described in detail by Koch et al. [2]. A spatial resolution of 1 μm was demonstrated by Martin et al. [3], using a 6 μm thick LSO(Tb) screen.
However, as thin-film scintillators have only a limited X-ray stopping power, the weak X-ray absorption in the scintillator leads to low performance in detection efficiency. Martin et al. [3] used two scintillator layers to overcome this problem for monochromatic synchrotron X-ray beams. If the emission spectra of the scintillator layers can be separated, a dichroic mirror can be used to divide both spectra and direct them to two unique lens-CCD systems. For each recorded image, the resolution depends only on the thickness of the scintillator, from which the emission light is captured. By adding the images, the effective detected X-ray count is increased which corresponds to the increase in effective absorption through the addition of a second scintillator layer. Martin et al. [3] showed for this system, that the detection efficiency results from the combined thickness of all scintillator layers, whilst maintaining the resolution of a single scintillator layer. Modgil et al. showed for synchrotron sources that “a dual-layer, dual-color scintillator construct can also be used to perform material quantification and classification when coupled with polychromatic illumination” [4] and derived the mathematical model needed.
Here we show that a dual-layer, dual-color system, modified and illuminated with a microfocus laboratory X-ray tube, can be used to perform energy resolved material detection with a micrometer spatial resolution. The concept is similar to the dual-layer detectors used in clinical CT setups [5].
2. Methods
2.1. X-ray imaging with thin-film scintillators
X-ray imaging with a thin-film luminescence screen (scintillator) and a microscope objective is an established technique, mainly at synchrotrons, in scientific research and is also commercially realized in desktop applications. The visible light emitting scintillator is imaged via a microscope system onto a 2D light sensitive sensor array (CCD). The imaging system is shown in Fig. 2. The resolution of the X-ray image, recorded with such a scintillator-microscope system, depends on the one side on the X-ray interactions in the scintillator and on the other side on the performance of the optical system. Therefore, characterizing properties of the components are for example scintillator thickness, magnification, numerical aperture and CCD pixel size. Below 100 keV, photoelectric effect is the dominant process for the spatial resolution in a luminescent screen [6]. For the scintillators used in this work, the interaction probabilities of X-rays at different energies have been calculated. Fig. 1 shows, that for the used scintillators this holds true around the expected effective energies.
Fig. 1 Interaction probabilities of the predominant interactions of X-rays with the scintillator materials. The solid grey line marks the position of the expected mean energy as shown in Fig. 6. The values are calculated from the cross-sections calculated with xraylib [7].
Fig. 2 X-ray imaging with transparent luminescent screens (scintillators). Identical visible-light images are created by the X-ray beam in different planes of the scintillator. An image in plane z0 is focused onto the CCD (solid curves). An image in plane z0 + δz is out of focus at the CCD (dashed curves). Image adapted from [2].
To investigate the influence in more detail Koch et al. [2] performed a Monte Carlo simulation of photon and electron transport in a YAG crystal triggered by a pencil X-ray beam. The simulation shows a very high absorption dose in 100 nm FWHM range. This can be explained by the short interaction path of Auger electrons. The energy deposition outside the 100 nm region is due to secondary electrons and characteristic fluorescence radiation. The response function of the scintillator with a FWHM of 100 nm indicates that the lateral resolution is not limited by the scintillator energy absorption but by the resolving power of the optical system.
The scintillator is placed in front of the objective in a way that the focus plane is positioned inside the luminescent screen at distance z0 (see Fig. 2). All planes in front of and behind the focus plane are out of focus, but contribute to the total light intensity recorded by the CCD. The resolution strongly depends on the defect of focus δz and therefore of the scintillator thickness z. The resolution is therefore governed by the defect of focus: rdefect_of_focus = δzNA with the numerical aperture NA and the diffraction of the objective, which is given by the Rayleigh criterion: rDiffraction = (0.61λ)/(NAobj). Koch et al. [2] derived the formula for the spatial resolution by fitting numerical simulated spatial resolution data with the function
and specified the parameters p = 0.18 and q = 0.075 (parameters for the full width, which covers 50 % of the integrated line spread function (LSF)). The first term expresses the diffraction and the second term is due to the defect of focus. The spatial resolution as a function of the numerical aperture NA for different scintillator thicknesses is plotted in Fig. 3.Fig. 3 Spatial resolution as a function of the numerical aperture NA of an optical system for different scintillator thicknesses z (see Eq. (1)). FW50%Int represents the full width, which covers 50% of the integrated line spread function (LSF).
The image quality can be described by the spatial resolution and by the detector quantum efficiency, which relates to the input and the output signal-to-noise ratio [6],
where SNRin and SNRout are the input and output signal-to-noise ratios of the detector, ηabs is the absorption for X-rays in the scintillator, ην/e is the quantum efficiency of the CCD, ηcoll = NA2/(4n2) is the collection efficiency of the light by an objective with numerical aperture NA in combination with a scintillator with refractive index n, Ex is the X-ray energy, Eν is the photon energy of the visible light photons and ηx/ν is the conversion efficiency of X-rays to visible-light photons.2.2. Dual-layer, dual-color X-ray imaging
Placing a second thin-film scintillator layer below the first gives the opportunity to gain spectral information. The first scintillator (top) will predominately absorb the low-energy X-ray photons in the polychromatic X-ray beam. Due to the beam hardening effect, more high-energy photons will be converted by the second scintillator (bottom). To distinguish between the top and bottom scintillator signal, the emission wavelengths need to be different. Using this configuration (single imaging path as in Fig. 2) and a color CCD one can record both signals simultaneously and derive them afterwards, as done by Modgil et al. [4]. The use of only one imaging path and a color CCD comes with the drawback of a stronger limitation of the spatial resolution as the DOF need to match the added thickness of both scintillators.
2.3. Evaluation of dual-energy data
There is a wide variety of dual-energy data processing techniques and they are subject of current research. The processing methods can be categorized in pre- and post-processing approaches. To show the dual-energy characteristic of the measured data sets, the post-processing concept of image-based decomposition, valid for simple radiography images or reconstructed CT volumes, is used, and briefly described in the following.
The first step of post-processing is an alignment of the datasets using image registration. We used our own implementation of the Lucas-Kanade algorithm for affine transformations [8, 9]. After this pixel wise alignment, the μ values in corresponding pixel pairs can be used as coordinates in a 2D-plane. The x-axis represents the attenuation coefficient for the low-energy Elow and the y-axis for the attenuation coefficient for the high-energy Ehigh. Plotting all pixel pairs in the attenuation plane leads to a 2D-histogram (Energy Map) [10]. The slope of a linear region in the Energy Map is given by the ratio of low- to high-energy attenuation coefficient, μ(Elow, Zeff, ρe)/μ(Ehigh, Zeff, ρe) = F(Zeff), and depends only on the effective atomic numbers Zeff of the material. Materials with different Zeff are represented in the Energy Map by different slopes. The density ρe or concentration c defines the position on the slope line [5]. One material with a certain Zeff and a defined density is characterized by one specific point on the Attenuation Map. In a real measurement, variations in Zeff and ρe combined with noise lead to blurring in the plane (Fig. 4).
Fig. 4 Energy map or Attenuation map; a) Materials with different effective Zeff values are represented by different slopes. The position on the linear region is defined by the density ρ or the concentration c of the specific material. The variation in density and the induced noise result in a blurring of the actual position. b) A mixture of two materials lies in a geometrically constructed area of the two materials.
3. Dual-layer, dual-color setup
To overcome the restrictions on the DOF, introduced by a single imaging path, we use in our setup two separated imaging paths, see Fig. 5. The image path separation requires the DOF to only match the relevant single scintillator thickness but introduces the need for a higher degree of alignment work with additional image registration. The path split is realized by a dichroic mirror placed in the infinity region of the microscope components, between microscope objective and tube lens. An image grade mirror with a reflection from 350 – 500 nm of > 95% and a transmission from 518 – 950 nm > 93% to match the scintillator emission wavelength as mentioned in Table 1. An optical system with microscope standard components was chosen for high resolution and high image quality. Standard microscope components are aberration corrected and come in a variety of magnifications. We insert a dividing layer between top and bottom scintillator to increase the beam hardening for the bottom scintillator, increasing the energy splitting of the dual-energy data set. Different thicknesses of the dividing layer can be used to tune the beam-hardening and therefor the energy splitting to optimize for different material combinations expected in the sample.
Fig. 5 Layout of the dual-layer dual-color X-ray imaging setup with separated imaging paths and the idealized source and absorption spectra of the different layers. Where T, D, B, ODD and WD represent Top scintillator layer, Dividing layer, Bottom scintillator layer, Object Detector Distance and Working Distance of the microscope objective.
Table 1. The physical properties of a selection of inorganic scintillators.
The setup design shown in Fig. 5 allows a short object-detector (sOD) and working distance. Therefore the sample can be placed near the detector (≈ 20 mm). This allows a reduction of the effective tube spot size and a high spatial resolution. The unrestricted microscope working distance (WD) allows a wide choice of different microscope objectives with high NAs and different magnifications, allowing the setup to be used both in a high-magnification configuration and in a large field of view configuration. This flexibility potentially allows the usage of the setup both in material and non-destructive testing as well as in pre-clinical, biomedical imaging applications. However, the scintillators must meet a number of requirements: Firstly, they must meet the requirements for dual-energy separation. The used scintillator materials must therefore be chosen with the maximum absorption and stopping power in the range of the desired spectra. Secondly, for the scintillator thicknesses there has to be made a compromise between resolution and light yield as explained further in section 3.2. Thirdly, the geometrical design demands the separation of the scintillators’ emission wavelength spectra with least possible overlap. The peak emission should be above 400 nm to enable the use of standard microscope optical components and should match the CCD quantum efficiency. Furthermore it is required that the dividing layer and the bottom scintillator are transparent for the emission light of the top scintillator and a similar refractive index in all three layers is preferable to reduce the loss of intensity due to scattering and refraction at the interfaces. Fourthly, a high light yield is favored to improve image quality and reduce exposure time.
3.1. Scintillator materials
For the top scintillator a zinc selenide scintillator doped with tellurium (ZnSe(Te)) was chosen due to its high wavelength of emission and its high light yield (compare Table 1). The ZnSe(Te) material also shows strong absorption for low (10 keV – 40 keV), and weak absorption for high X-ray energies (80 keV – 160 keV; compare Fig. 6 (d)). The maximum emission wavelength of 640 nm matches well the QE of the used CCD sensor (camera: FLI PL1001; sensor: ON Semi KAF-1001E; compare [11]). Its emission characteristic also enables a wide choice of scintillators for the bottom scintillator as most scintillators show an emission spectrum with lower maximum emission wavelengths than ZnSe(Te), therefore, the combination with other scintillator materials show weak overlap of the emission spectra. Furthermore, do most scintillators have a good transmittance in the 640 nm wavelength regime.
Fig. 6 Results of the absorption simulation with a Tungsten X-ray tube with 160 keV, 10 mm water as target, 200 μm ZnSe(Te) top scintillator, 200 μm ZrO dividing layer and 600 μm LYSO(Ce) as bottom scintillator. a) The X-ray spectra at different positions in the beam path; b) The spectrum absorbed by the top scintillator (Red line; 21.2 % of the intensity after sample), the dividing layer (Gray line; 17.6 % of the intensity after sample) and the bottom scintillator (Green line; 46.0 % of the intensity after sample); The corresponding mean energies are: Ēlow = 46.13 keV; Ēdiv = 55.24 keV; Ēhigh = 72.97 keV; c) The calculated DQE for each scintillator; d) The linear attenuation coefficient μ for ZnSe, ZrO and LYSO in the corresponding energy range.
For the bottom scintillator the cerium-doped rare-earth scintillator lutetium-yttrium oxyorthosilicate (LYSO(Ce)) was chosen. LYSO has a high density (compare Table 1) and therefore a low radiation length X0, which is an important attribute, since the bottom scintillator has to absorb the high energy X-ray photons. Additionally, Lutetium has its K-edge at 63.3 keV (Fig. 6 (d)) which can be used to increase the absorption for energies above the edge. It also shows a good light yield with a matching emission spectrum compared with ZnSe(Te).
For the dividing layer zirconium oxide (ZrO2, ρ = 5.9 g/cm3, n = 2.15) was chosen due to its high attenuation from 19 – 80 keV (Fig. 6 (d)) and its constant decrease to higher energies. Furthermore, zirconium oxide shows a good transmittance of used wavelengths.
3.2. Scintillator thickness
After the material definition, each layer’s thickness has to be determined. The thickness defines the spectrum, the intensity and the mean energy which is absorbed by each layer. Therefore, the thickness defines the spectral dual-energy separation. Additionally a trade-off has to be found between a thick scintillator with a high absorption, resulting in a high light output but worse spatial resolution, and a thin scintillator with weak absorption and light output but high spatial resolution (compare Fig. 3). For optimizing the layer thicknesses a computer simulation with the pyXSFW package [16] was performed, therefor, simulated X-ray tube spectra (simulated with pyPENELOPE [17]) were used.
The total recording efficiency of the scintillators and therefore the image quality were estimated by calculating the DQE (see Eq. (2)) for each scintillator layer. The absorption efficiency ηabs is calculated by the ratio of the absorption intensity and the X-ray tube intensity, ην/e represents the light conversion efficiency and was determined by the CCD QE [11] and the wavelength of maximum emission (see Table 1), the light collection efficiency ηcoll was calculated with (NA2)/(4n2), with the refractive index n of the scintillator and a NA of 0.13. The X-ray conversion efficiency is proportional to the light yield of the respective scintillator (see Table 1). For Ex and Eν, the mean energy of the absorbed spectrum and the wavelength of maximal emission were used. A variety of different thickness combinations was simulated.
For our setup we used the combinations:
- Top scintillator layer: 200 μm of ZnSe(Te);
- Dividing layer: 0/200/400 μm of ZrO stabilized with yttria;
- Bottom scintillator layer: 600 μm of LYSO(Ce).
In Fig. 6 the result for the absorption simulation with the calculated DQE is displayed. The used scintillator thicknesses limit the spatial resolution (see Fig. 3), but for the enhancement of light output and DQE, the reduced resolution was accepted.
3.3. Critical setup components
During the characterisation oft the Dual-Layer Dual-Color Setup, we could distinct four critical components which need to be considered with special care. Using a CCD camera in the environment of an X-ray tube leads to bright, sharp spots in the image, created by direct hits (in astronomy: cosmic ray artifact). The size of these spots are in the range of a few pixels or more, rarely in bright streaks if the high energy particle hits the sensor in a shallow angle. A large number of direct hits reduces the SNR significantly. Therefore, the camera CCD has to be shielded with lead and lead glass. The dichroic mirror has to be of image quality and requires stress free mounting, otherwise the curvature of the mirror produces astigmatism in the reflected image. Due to the limited X-ray photon flux, of the used X-RAY WorkX 190SE microfocus X-ray tube [19], exposure times from 15 to 600 second were needed, depending on the setup configuration. Finally, imperfections of the scintillator crystals like impurities, microbubbles and scratches are strongly visible at the used magnifications and are a challenging task to correct for.
4. Results
4.1. Spatial resolution
The resolution power of the setup was specified by using an Xradia test pattern. The center of the test pattern consists of a Siemens star with a decreasing line width from 64 – 2 μm. On the edges, there are line spacing gratings with a half period from 32 – 4 μm. Using the microscope objective Nikon CFI Plan Fluor with a NA = 0.13 and the scintillator thicknesses of 200 μm and 600 μm in Eq. (1) the optical resolution is expected to be 2.4 μm and 6.0 μm. The total magnification equals the geometrical times the optical magnification. The geometrical magnification is in the range from 1.07 − 1.11, depending on the source-detector distance (sSD from 200 – 300 mm). This leads to a mean total magnification of 4.2 and an effective pixel size of 5.7 μm (12 μm physical sensor pixel size). Therefore the camera and overall resolution is expected to be 11.4 μm. In Fig. 7 (a) and (b) the recorded images with the best resolution achieved are presented. For the low-energy image the expected resolution of ∼ 12 μm could be confirmed. The measured effective pixel size in the image was (5.72 ± 0.02) μm. For the high-energy image a resolution of 24 μm was measured. Reasons for not achieving the expected resolution are: aberration error produced by the mirror, a weak absorption contrast in the Xradia test pattern for the high-energies detected, scatter events in the top scintillator or the scintillator response function.
Fig. 7 Dual energy images with Nikon objective (Flat-field and dark-frame corrected; exposure time = 360 s; sSD = 305 mm; sOD = 15 mm; Upeak = 80 keV; P = 30 μW). a) Low-energy image with an effective pixel size of 5.72 μm. b) High-energy image with an effective pixel size of 5.73 μm. c) Low energy image with Zeiss objective (Flat-field and dark-frame corrected; exposure time = 600 s; sSD = 305 mm; sOD = 15 mm; Upeak = 80 keV; P = 30 W). Low-energy image with an effective pixel size of 1.99 ± 0.02 μm. No dividing layer in both arrangements.
To specify the scintillator resolution behavior, measurements with the Zeiss A Plan 10X objective were performed. Zeiss designs their objectives for a tube lens focal length of 165 mm, therefore, the Zeiss 10X objective with the Nikon tube lens has an effective magnification of 12.1. With this and the numerical aperture of 0.25, an optical resolution of 3.8 μm and 11.3 μm were expected. The effective pixel size of 2.0 μm and thus a detector resolution of 4.0 μm are assumed. Due to a weak luminescent intensity of the scintillators at this configuration an exposure time of 10 min was necessary. In Fig. 7 (c) the recorded image of the low-energy image is displayed. Despite the long exposure time the achieved resolution of 10 μm seems to be limited by the recording statistics. For the high-energy image only blurred images with a resolution of ∼20 – 30 μm could be recorded, caused by the scintillator thickness and the resulting weak optical resolution.
4.2. Spectral information
4.2.1. Radiographic material separation
The fundamental material separation ability of the setup was first investigated with radiography images of a sample composed of three quadratic metal plates with a thickness of 1 mm. The metal plates made of aluminum, iron and copper are placed in a checkered pattern [Fig. 8 (c)]. The low- and high-energy images can be partitioned into four areas: aluminum, iron, copper and air [Fig. 8 (a, b)]. The measurements were taken with an X-ray tube peak voltage of 160 keV and 200 μm dividing layer. The linear attenuation coefficient of the flat homogeneous objects with the thickness x were calculated with the equation
which follows directly from Lambert-Beer’s law. The intensity fraction I/I0 equals the recorded images with flat-field and dark-frame correction. When the negative logarithm of the low- and high-energy images is taken and thickness corrected, the corresponding pixel pairs (if aligned) represent μ(Ēlow, ZZeff, ρ) and μ(Ēhigh, ZZeff, ρ). The results are plotted in Fig. 8. The four areas and the different materials can nicely be separated.Fig. 8 Energy map of a sample composed of 1 mm aluminum, 1 mm iron and 1 mm copper. The total field of view is used. Low- (a) and high-energy (b) image of the aluminum-iron-copper sample. (Flat-field and dark-frame corrected; exposure time = 30 s; sSD = 305 mm; sOD = 20 mm; Upeak = 160 keV; P = 120 W). c) Image of sample. A dividing layer of 200 μm Zr0 was used.
4.2.2. CT material separation
To perform a material separation in a CT reconstructed volume a gold ore sample (3 mm × 4 mm × 5 mm) originating from the Witwatersrand area (South Africa) was chosen. The measurements were performed at an X-ray tube peak voltage of 160 keV with a 0.05 mm thick copper filter and a 400 μm dividing layer was used. For the CT, 809 angles with an exposure time of 15 s were recorded. A binning of 2 was used, resulting in a pixel size of 24 μm. The energy map of the separately reconstructed low- and high-energy volumes is shown in Fig. 9. The results of the statistical-iterative-reconstructions (SIR) are shown in the Stone-CT Dataset 1 [18]. Two prominent regions in the energy map which are clearly different in the stone matrix were selected to perform material separation. In Fig. 10 (a) a 3D rendered (Software FEI Avizo) stone model with a material separation using the pixel selection according to the marked areas in Fig. 9 is plotted. Fig. 10 (b) and (c) show the same stone volume, where the prominent material (red area) was thresholded separately in the low- [Fig. 10(b)] and the high-energy [Fig. 10(c)] data set. The red area material can’t neither be clearly separated in the low- nor in the high-energy data set. Only by using the spectral information and defining the corresponding area in the energy map, it is possible to make a clear separation of the two highly absorbing materials. Due to the origin of the sample and previous investigations of similar materials we assume that in Fig. 10 (a) the red area represents gold (Z = 79) and the blue area represents lead (Z = 82) or uranium (Z = 92).
Fig. 9 Energy map of the reconstructed gold ore sample. The total reconstructed volume is used. Blue and Red areas correspond to the manual attenuation value selection which was used for material separation. Lines are drawn in for illustratory purposes and show possible slopes of the selected materials. The CT reconstructions are shown in the Stone-CT Dataset 1 [18].
Fig. 10 3D rendering of the gold ore sample with a) the Energy map selection shown in Fig. 9 and the stone segmentation yielded by thresholding in the low energy CT [18] b) and high energy CT [18] c).
5. Discussion and conclusion
The chosen scintillator thicknesses of 200 μm and 600 μm with the Nikon CFI Plan Flour objective (4X, NA = 0.13) and a sensor pixel size of 24 μm do not exhaust the possible resolution potential of the system. To decrease the absorbed spectral overlap and increase the energy separation, a dividing layer was placed in between the scintillators.
Dual-energy data sets were collected by taking radiography images and CT measurements.
For the radiography images, the reference attenuation in the unobscured detector area showed an unexpected behavior. The air segment was expected to have an attenuation value of zero for both energies. However, it was observed that the air attenuation seems to be dependent on the absorption behavior of the surrounding materials. The air attenuation values were centered close to the expected zero values in images with weak absorbing materials and at an increased distance with strong absorbing materials such as iron and copper present. A possible explanation is luminescent and stray light behavior taking place inside and outside the scintillator. The luminescence of a certain area of the scintillator depends not only on the radiation deposit in this area, but also on the radiation of the surrounding scintillator areas. Due to impurities in the scintillator and total reflection at the scintillator interface, light of a certain area of the scintillator can emerge at another. Additionally it is possible that emerged scintillator light is reflected by the surrounding components and reenters the scintillator or enters the objective, leading to stray light and therefore to a false signal which is dependent on the total illumination of the scintillator.
For the CT measurements of the gold ore sample we could prove that, with the two recorded data sets, a usage of the collected spectral information is possible. In this case, we were able to separate gold and lead which would not be possible in a single reconstruction. It showed, that registration of the datasets prior to dual-energy evaluation is a crucial step, which can be easily done by registration of the projection images before reconstruction.
Finally, we showed, that the simultaneous acquisition of dual-energy datasets is possible with the presented setup with a resolution down to a few tens of micrometers and without additional dose compared to the mono-energetic case.
Funding
We acknowledge financial support through the European Research Council (ERC, H2020, AdG 695045), the DFG Cluster of Excellence Munich-Centre for Advanced Photonics (MAP), the DFG Gottfried Wilhelm Leibniz program and the support of the TUM Institute for Advanced Study, funded by the German Excellence Initiative.
References and links
1. K. C. Cheng, X. Xin, D. P. Clark, and P. La Riviere, “Whole-animal imaging, gene function, and the Zebrafish Phenome Project,” Cur. Opin. Genet. Dev. 21(5), 620–629 (2011). [CrossRef]
2. A. Koch, C. Raven, P. Spanne, and A. Snigirev, “X-ray imaging with submicrometer resolution employing transparent luminescent screens,” J. Opt. Soc. Am. 15(7), 1940–1951 (1998). [CrossRef]
3. T. Martin, P. A. Douissard, M. Couchaud, A. Cecilia, T. Baumbach, K. Dupre, and A. Rack, “LSO-based single crystal film scintillator for synchrotron-based hard X-ray micro-imaging,” IEEE T. Nucl. Sci. 3(56), 1412–1418 (2009). [CrossRef]
4. D. Modgil, D. S. Rigie, Y. Wang, X. Xiao, P. A. Vargas, and P. J. La Rivière, “Material identification in x-ray microscopy and micro CT using multi-layer, multi-color scintillation detectors,” Phys. Med. Biol. 60(20), 8025 (2015). [CrossRef] [PubMed]
5. B. J. Heismann, J. Leppert, and K. Stierstorfer, “Density and atomic number measurements with spectral x-ray attenuation method,” Appl. Phys. 94(3), 2073–2079 (2003). [CrossRef]
6. T. Martin and A. Koch, “Recent developments in X-ray imaging with micrometer spatial resolution,” J. Synchrotron Radiat. 13(2), 180–194 (2006). [CrossRef] [PubMed]
7. T. Schoonjans, A. Brunetti, B. Golosio, M. Sanchez del Rio, V. Armando Solé, C. Ferrero, and L. Vincze, “The xraylib library for X-ray-matter interactions. Recent developments,” Spectrochim. Acta B 66(11–12), 776–784 (2001). [CrossRef]
8. B. D. Lucas and T. Kanade, “An Iterative Image Registration Technique with an Application to Stereo Vision,” in Proc. 7th Intl. Joint Con. on Artificial Intelligence, (1981), 674–679.
9. S. Baker and I. Matthews, “Lucas-Kanade 20 Years On: A Unifying Framework,” Int. J. Comput. Vision 56(3), 221–255 (2004) [CrossRef]
10. T. Johnson, C. Fink, S. O. Schönberg, and M. F. Reiser, “Dual Energy CT in Clinical Practice,” in Medical Radiology (SpringerBerlin Heidelberg, 2011). [CrossRef]
11. Finger Lakes Instrumentation, “PL1001E” (2016).
12. V. Gaysinskiy, B. Singh, L. Ovechkina, S. Miller, S. Thacker, and V. Nagarkar, “Luminescence properties and morphology of ZnSe:Te films,” IEEE T. Nucl. Sci. 55(3), 1556–1560 (2008). [CrossRef]
13. P. Schotanus, P. Dorenbos, and V. D. Ryzhikov, “Detection of CdS(Te) and ZnSe(Te) scintillation light with silicon photodiodes,” IEEE T. Nucl. Sci. 39(4), 546–550 (1992). [CrossRef]
14. Saint-Gobain Crystals, “Physical Properties of Common Inorganic Scintillators” (2016).
15. CRYTUR, “Scintillation Materials Data” (2016).
16. A. Malecki, G. Potdevin, and F. Pfeiffer, “Quantitative wave-optical numerical analysis of the dark-field signal in grating-based X-ray interferometry,” Europhys. Lett. 99(4), 48001 (2012). [CrossRef]
17. P. T. Pinard, “pyPENELOPE,” (2012).
18. J. Schock, “Dual Energy Micro CT of a Stone Sample,” figshare (2016) [retrieved 15 December 2016], https://dx.doi.org/10.6084/m9.figshare.4322465.v1.
19. X-RAY WorX GmbH, “PRODUCT LINE SE,” http://www.x-ray-worx.com (2016).
