Two-Color X-ray Imaging Techniques

• Two-color X-ray imaging is a method that acquires dual energy channels in a single exposure to improve contrast by differentiating low- and high-energy photons.
• Advanced implementations include photon-counting with direct conversion, scintillator stacking for spectral color separation, and spectral-spatial cameras with sub-pixel resolution.
• Applications range from radiographic material discrimination and element-specific XRF imaging to ultrafast pump-probe experiments, while challenges involve photon pile-up and data disentanglement.

Two-color X-ray imaging denotes the acquisition of two energy-distinguished X-ray image channels from the same object, typically in a single exposure or single shot. In different subfields, the two channels may be produced by photon-by-photon energy thresholding in a direct-conversion detector (Bellazzini et al., 2012), by converting different X-ray energy ranges into different scintillation colors in a stacked scintillator flat-panel architecture (Ran et al., 2022), by assigning detected fluorescence energies to distinct spatial maps in full-field spectral cameras (Nowak et al., 2015), or by illuminating a sample with two synchronized free-electron-laser pulses at different photon energies and disentangling the superposed signals computationally (Hecht et al., 27 Aug 2025). The shared objective is to move beyond conventional energy-integration black-white X-ray imaging, whose contrast is proportional to the integrated X-ray intensity and contains no spectral information (Ran et al., 2022).

1. Spectral contrast as the defining principle

In conventional energy-integration flat-panel X-ray imaging, the detector records the summed light from all transmitted photons, so the image contrast depends only on the integrated X-ray intensity. This suppresses information about whether the detected signal originated from low- or high-energy photons. Because X-ray attenuation depends strongly on energy through photoelectric absorption and Compton scattering, separating the spectrum improves material discrimination and can reveal details that are invisible or ambiguous in conventional energy-integration imaging (Ran et al., 2022).

A direct detector implementation of this principle is chromatic photon counting. In the CdTe-CMOS architecture described for PIXIRAD, each pixel contains two discriminators and two 15-bit counters/registers, so photons are counted individually and sorted in real time into two threshold-defined bands. In the simplest mode, the LOW counter records all photons above the low threshold, the HIGH counter records only photons above the higher threshold, and the low-energy-only image is reconstructed by subtraction,

$I_{\text{low}} = I_{\text{LOW}} - I_{\text{HIGH}}.$

This yields a total/low-threshold image, a high-energy image, and a derived low-energy image from one exposure (Bellazzini et al., 2012).

A different realization is stacked scintillation. Here, energy discrimination is translated into color discrimination: spectrally distinct scintillator layers are arranged along the beam direction, each chosen to absorb a different part of the X-ray spectrum and emit in a different visible band. In the schematic formulation, layers $R_1 \sim R_4$ correspond to energy bins $E_1 \sim E_4$, with the requirement that the radioluminescence spectra are non-overlapping so that each energy channel can be read out independently (Ran et al., 2022).

In fluorescence and spectral microscopy, “two-color” frequently refers to simultaneous spatial mapping of distinct X-ray lines or energy windows. In the SLcam(R)-type pnCCD camera, different X-ray lines such as Au L and Cu K can be recorded in the same exposure and mapped separately to produce element- or energy-specific images (Nowak et al., 2015). In a large-area triple-GEM XRF imager, the final rendering uses hue to encode X-ray energy and brightness to encode intensity, so color has a direct correspondence to the energy in each pixel (Souza et al., 2019).

In ultrafast coherent diffraction imaging, the term has a further extension: the first X-ray pulse captures the object’s initial state and a time-delayed second pulse records its subsequent evolution, with both diffraction images landing on the same detector frame and separated afterward on the basis of photon energy (Hecht et al., 27 Aug 2025). This suggests that two-color X-ray imaging is best understood as a family of energy-resolved imaging strategies rather than a single detector class.

2. Large-area radiographic implementations

The most direct large-area radiographic route is photon counting with direct conversion. The PIXIRAD detector couples a thin pixellated CdTe crystal of 650 µm thickness to a large-area CMOS pixel ASIC with a 512 × 476 matrix on an active area of 30.7 × 24.8 mm². The CdTe sensor operates in 1–100 keV, with reported detection efficiency of 100% at 10 keV and 98% at 50 keV. The ASIC distributes the analog chain across the detector surface, supports 2-color reading mode with two thresholds and two counters per pixel, and provides very high spatial resolution, reported as 11 line pairs/mm at 50% MTF and also as more than 10 l.p./mm at MTF50. Modular assemblies of 1, 2, 4, and 8 tile units were built, with the 8 tiles unit reaching 25 cm × 2.5 cm sensitive area (Bellazzini et al., 2012).

A different large-area strategy replaces photon counting with scintillator stacking and optical color separation. In the multilayer FPXI architecture, the proof-of-concept dual-energy system used blue-emitting $\mathrm{Cs_3Cu_2I_5}$ and red-emitting $\mathrm{C_4H_{12}NMnCl_4}$, with radioluminescence peaks at 452 nm and 643 nm, respectively, and no overlap. Both were prepared as films in a PMMA matrix using scalable solution processes, and the equivalent thickness was defined as

$T = \frac{M}{\rho \times S},$

where $M$ is the mass of scintillator, $\rho$ its density, and $S$ the film area. For dual-energy imaging, the best compromise was found with 84 μm $\mathrm{C_4H_{12}NMnCl_4}$ on top and 113 μm $R_1 \sim R_4$0 below. The stacked detector yielded 10.8 lp mm$R_1 \sim R_4$1 in the blue channel and 6.8 lp mm$R_1 \sim R_4$2 in the red channel, and the concept was extended to a four-layer prototype comprising FAPbI$R_1 \sim R_4$3, $R_1 \sim R_4$4, $R_1 \sim R_4$5, and $R_1 \sim R_4$6, with channels centered at 5, 15, 30, and 50 keV (Ran et al., 2022).

These two architectures address different bottlenecks. Photon-counting detectors provide explicit event-by-event energy sorting, but large-area flat-panel deployment is difficult because of pile-up at high flux, expensive and hard-to-grow CZT crystals, and complex readout circuits. The stacked scintillator approach avoids photon counting and instead uses ordinary color or multispectral visible-light cameras, making it inherently compatible with conventional flat-panel X-ray imaging workflows (Ran et al., 2022). By contrast, the CdTe-CMOS route preserves direct energy thresholding at the pixel level and real-time operation at radiographic imaging speed (Bellazzini et al., 2012).

3. Spectral-spatial cameras and energy-coded imaging

Full-field color X-ray cameras combine single-photon spectroscopy with spatial imaging. In the SLcam(R)-type system, a pnCCD with 48 μm pixel size is coupled to polycapillary optics, and each absorbed photon produces an electron cloud whose total charge determines energy according to

$R_1 \sim R_4$7

Because the charge pattern across neighboring pixels also encodes hit position, the same detector is simultaneously energy- and position-sensitive. The sub-pixel localization method proposed in this context uses all photon events, including events from pixel centers, rather than only corner events; for a $R_1 \sim R_4$8 pnCCD, a division up to about $R_1 \sim R_4$9 sub-pixels was concluded to be a realistic upper limit. In imaging demonstrations with 8:1 conical magnifying optics, the method enabled distinction of $E_1 \sim E_4$0 and even $E_1 \sim E_4$1 bars in some profiles (Nowak et al., 2015).

Polycapillary optimization pushes this approach toward micron resolution. The optic-controlled point-spread function depends on capillary diameter, acceptance angle, sample-optic distance, optic-detector gap, magnification, and photon energy. With the 8:1/2 magnifying optic, measured Cu K-line resolution reached about 6.5 μm with 1×1 pixels and improved to 4.6–3.9 μm with subpixel divisions; for focusing optics with $E_1 \sim E_4$2, a theoretical limit of about 2.6 μm was reported, and hexagonal lines of width comparable to 2 μm could be resolved. The same framework was used for false-color multi-energy imaging of a snail radula combining Fe Kα, Cu Kα, and scattered X-rays (Nowak et al., 2017).

For hot dense plasmas, EPiC extends the spectral-spatial idea to single-shot diagnostics over a broad energy range. The Energy-encoded Pinhole Camera uses an array of many pinholes and a large-area CCD operating in the single-photon-counting regime, so that each event carries both position and energy information. The available X-ray spectral domain extends from a few keV up to a few tens of keV, and monochromatic images can be reconstructed at selected energies provided that sufficient photons are collected. In one Vulcan PW experiment, the reconstructed spectrum showed a strong Ti He$E_1 \sim E_4$3 line at about $E_1 \sim E_4$4, while images at $E_1 \sim E_4$5 isolated a smaller continuum-emitting region, demonstrating that line-emission and continuum images can be separated from the same shot (Labate et al., 2012).

Energy-coded fluorescence imaging can also be implemented with gas detectors. A triple-GEM system with resistive charge division reconstructs $E_1 \sim E_4$6 from only four strip-end channels plus a fifth trigger/energy channel, and after temporal and spatial gain corrections it achieved an energy resolution of 6.8% ($E_1 \sim E_4$7). The final image maps energy into RGB color and intensity into brightness. Spatial performance was reported as approximately 1.8 mm at the 10% MTF point, with the best position resolution of 1.2 mm in the 8–9 keV range. Demonstrations with cadmium yellow, chrome yellow, cerulean blue, and cobalt blue showed qualitative pigment separation by fluorescence energy (Souza et al., 2019).

4. Two-color X-ray sources for ultrafast imaging

In free-electron-laser science, two-color X-ray imaging depends as much on source engineering as on detector design. One route is the sextupole-based SwissFEL method, where a sextupole magnet in a dispersive bunch compressor, together with standard orbit correction, creates a quadratic transverse tilt of the electron bunch. This deliberately suppresses lasing from the bunch core while keeping the head and tail slices lasing at different photon energies. The physical basis is the FEL resonance condition,

$E_1 \sim E_4$8

so slices with different $E_1 \sim E_4$9 emit different photon energies. Experimentally, two-color operation with relative central photon-energy separations of approximately 1.2% to 2.2% and temporal separations of approximately 26 to 48 fs was demonstrated, with pulse energies up to 100 μJ in the two-color mode (Dijkstal et al., 2019).

A second route is attosecond two-color generation by a dual chirp-taper FEL with bunching inheritance. In this architecture, a 1.5-cycle, 800 nm laser modulates the beam in a two-period wiggler, a long tapered main undulator generates the first attosecond pulse from a positively chirped slice, and a chicane plus a short oppositely tapered afterburner generates the second color from a negatively chirped slice using inherited microbunching. Simulations used a 2.53 GeV beam with 2 kA peak current; the main undulator pulse reached 1.5 GW peak power with 317 as FWHM duration, while the afterburner pulse reached 1.2 GW with 198 as FWHM duration. The temporal separation was controllable from hundreds of attoseconds to several femtoseconds, with an example chicane delay of 1.5 fs, and the photon-energy difference was about 30 eV in simulation. Stable two-color generation was maintained within ±0.3π CEP phase jitter, and the afterburner radiation showed better shot-to-shot stability than the main undulator radiation (Sun et al., 2024).

For imaging, these source developments are important because they provide two synchronized X-ray colors from a single electron beam or a single bunch, reducing temporal jitter relative to independent sources. They are therefore directly relevant to static two-color imaging, time-resolved coherent diffraction imaging, and pump-probe experiments in which excitation and probing at different photon energies must be tightly synchronized (Dijkstal et al., 2019, Sun et al., 2024).

5. Separation of overlapping two-color data

When both colors are recorded simultaneously on one detector, the central computational problem is disentanglement. This was demonstrated directly for two-color coherent diffraction imaging of individual helium nanodroplets at the European XFEL SASE3 beamline. Two co-propagating FEL pulses were generated at mean photon energies of about 992 eV and 1192 eV, with delays varied from 50 fs up to 750 fs, and the superposed diffraction patterns were recorded on a large-area pnCCD detector. Because the pnCCD is energy sensitive, the deposited charge in each pixel differs statistically for the two photon energies, enabling software-based separation rather than spatial beam separation (Hecht et al., 27 Aug 2025).

The first separation strategy was pixel-count and pattern-recognition analysis. Detector preprocessing included background subtraction, defective-pixel masking, noise gating, common-mode correction, gain correction, and charge-transfer efficiency correction. Bright pixels were grouped into clusters, standard charge-sharing patterns up to four neighboring pixels were identified, and each cluster total in ADU was compared with Gaussian models for noise, one red photon, one blue photon, two-photon combinations, and mixed-color combinations. A 50% certainty threshold was used in the main analysis. This method worked best in the outer, sparsely illuminated region of the diffraction pattern, roughly for $\mathrm{Cs_3Cu_2I_5}$0 or where the local density was below about 0.1 detected photons per pixel (Hecht et al., 27 Aug 2025).

The second strategy used prior structure knowledge. Because helium nanodroplets are nearly spherical, the radial profile could be modeled as the sum of two Mie-scattering contributions,

$\mathrm{Cs_3Cu_2I_5}$1

with one Mie solution for each photon energy. The approximately 20% energy difference produced a corresponding difference in fringe spacing and a characteristic beating in the combined radial profile. The paper reported excellent agreement between the pixel-based separation and the Mie-based decomposition, especially in the outer parts of the diffraction pattern where fine structural information is encoded. The main discrepancy occurred in the bright central region, where cluster overlap makes pixel-wise separation ambiguous, and where the Mie approach remains usable only because it is model-dependent (Hecht et al., 27 Aug 2025).

The experiment also clarified the role of the pump-probe delay. The helium nanodroplets remained essentially static up to 750 fs, with only about 0.01% of He atoms ionized at the experimental fluence, so the work functioned as a benchmark of the separation method rather than as a demonstration of strong target evolution (Hecht et al., 27 Aug 2025).

6. Applications, limits, and conceptual boundaries

Two-color X-ray imaging is used whenever a single grayscale projection is insufficient. In flat-panel radiography, a stacked scintillator detector imaged a “bone–muscle” phantom made from aluminum and cardboard and showed that one channel highlighted the bone defect while the other highlighted the muscle defect; in the stacked detector, both defects became visible in separate channels (Ran et al., 2022). In chromatic photon-counting radiography, a single-shot image of a man’s watch and leather bracelet produced two simultaneous energy views, with one image emphasizing low absorbing materials such as plastic and leather and the other better visualizing metal parts (Bellazzini et al., 2012). In plasma diagnostics, EPiC separated Ti K-shell line emission from higher-energy continuum in single-shot laser-fusion experiments (Labate et al., 2012). In XRF imaging, spectral channels have been used for elemental localization in pigments and other heterogeneous materials (Souza et al., 2019, Nowak et al., 2017).

The principal limitations depend on architecture. Large-area photon counting remains difficult because of pile-up at high flux, hard-to-grow CZT crystals, and complex readout circuits (Ran et al., 2022). In the PIXIRAD ASIC, each pixel still has a finite dead time of about 300 ns, even though the system supports a dead-time-free single-threshold mode by alternating counters (Bellazzini et al., 2012). In pnCCD-based two-color diffraction imaging, color separation degrades in the bright central region because photon overlap and charge spill-over become severe (Hecht et al., 27 Aug 2025). EPiC is constrained by photon statistics, the single-photon condition, CCD quantum efficiency, and overlap between neighboring pinhole images (Labate et al., 2012). Polycapillary cameras remain bounded by capillary spacing, point-spread blur, sample-optic distance, and the Nyquist-Shannon condition $\mathrm{Cs_3Cu_2I_5}$2, even when subpixel reconstruction mitigates pixel-size limitations (Nowak et al., 2017).

A common misconception is that “two-color” necessarily means visible RGB imaging. In practice, the two channels may be two threshold-defined count maps, two scintillation emission bands, two fluorescence-line images, or two FEL photon energies separated only in the detector signal or in post-processing [(Bellazzini et al., 2012); (Ran et al., 2022); (Hecht et al., 27 Aug 2025)]. A second misconception is that any method using two X-ray images is therefore two-color imaging. Stereo X-ray tomography instead uses two projection images from different directions and explicitly does not reconstruct arbitrary volumetric attenuation fields from only two views; it recovers the 3D locations of sparse point-like features and line endpoints by feature detection, matching, and triangulation (Shang et al., 2023).

The boundary between two-color and multispectral imaging is also fluid rather than categorical. The stacked multilayer scintillator work extended the dual-energy concept to a four-channel prototype with intended-band fractions of 55.2%, 53.1%, 50.2%, and 78.4% for the four layers, thereby demonstrating feasibility of multispectral or even hyperspectral large-area FPXI (Ran et al., 2022). This suggests that two-color X-ray imaging is often the entry point to a broader program of energy-resolved X-ray imaging in which the decisive technical questions are channel separation, photon statistics, spatial fidelity, and compatibility with the required source brightness and acquisition geometry.