In conventional x-ray imaging, contrast depends on material density. While e.g. soft tissue gives good contrast against bone, materials with similar density are harder to distinguish. This can be circumvented by exploiting the energy dependence of the linear attenuation coefficient μ_E as proposed by [1]. In particular, the element specific absorption edges allow for material decomposition, when energy bins can be selected in the vicinity of the edge. Acquiring images with different x-ray energies will thus allow for separation of materials with similar density based on their specific absorption edges [2, 3].

By applying material decomposition, additional information can be obtained from a sample. It can be used to separate materials that are indistinguishable from another in conventional imaging as shown by [4]. Further, materials and theirs concentrations in a sample can be assessed, providing spectroscopic information. This approach can greatly simplify segmentation of computed tomography (CT) scans and allows to extract material quantities or concentrations in a non-destructive way.

Typically, material decomposition is performed in dual energy CT scanners. A common method with clinical CT scanners is to manipulate the emitted spectrum of the x-ray source by changing the acceleration voltage, commonly known as the dual-kV method [1, 5, 6]. Utilizing the non-linearity of the attenuation coefficients of different elements, materials can be separated. However, this approach requires two consecutive acquisitions. Alternatively, systems with two x-ray sources and detectors can be used as demonstrated by [7] and [8]. A different approach is to use a photon counting detector with multiple energy-discriminating thresholds as shown by [9–11]. Such a detector allows to create energy bins covering a certain energy range. Each individual photon creates a signal proportional to its energy, which is only counted if exceeding the threshold energy [see [12]]. Photon counting detectors are showing great potential for pre-clinical and clinical applications as outlined by [13–16]. An alternative approach is to measure the time-over-threshold (ToT), where the duration of a signal, which is proportional to its energy, is measured and then related to a specific energy as implemented for the Timepix detector developed by [17].

In clinical settings, with x-ray energies well above 15 keV and decimeter sized objects (such as patients), the elements available of K-edge contrast are those with Z ≥ 37. However, most of the elements relevant in organic tissue are in the periods 1–4 in the periodic table (Z ≤ 36), and consequently a contrast agent containing an element with an absorption edge in the high energy range must be added to make use of this effect. Common x-ray contrast agents contain iodine (Z = 53) or gadolinium (Z = 64) with K-shell binding energies above 30 keV (E_B = 33 and 50 keV, respectively), and can be exploited in the clinical scenario [2]. In order to use the K-edge attenuation discontinuities of lighter elements, such as iron (Z = 26) or zinc (Z = 30), naturally occurring in the human body, an imaging setup sensitive to lower energies is required. This is feasible for x-ray microscopy studies of millimeter to centimeter sized samples.

Our setup, working in a low energy range (4–11 keV), allows us to decompose elements with K-edges in this range (Cr, Mn, Fe, Co, Ni, Cu, Zn, Ga). Further, Lanthanide elements (Z = 57–71), e.g., gadolinium, can be used for materials decomposition due to their L-edge discontinuities. However, L-edges are less dominant in the attenuation spectrum and will be affected more by sample thickness. Energy windows have to be selected as narrow as possible in the vicinity of the absorption edges, to best exploit the contrast difference. Potential limitations are beam hardening effects and photon starvation in narrow energy windows.

In this study, micro-CT material decomposition with a dual-threshold photon counting detector is evaluated in the energy range 4–11 keV. The method is verified using a phantom and applied to biomedical samples containing materials with absorption edges in the given energy range. This method is proposed for spectral evaluation of samples. When requiring higher resolution, higher sensitivity and better energy resolution, this method can be used for pre-scanning to guide sample and region-of-interest (ROI) selection prior to experiments, e.g., x-ray fluorescence or high resolution imaging, at a synchrotron or other large scale facilities. With time and access being the limiting factors at large facilities, this approach has the potential to use allocated time more efficiently.

The verification sample demonstrates that the implemented decomposition method works, albeit the result is affected by beam hardening and ring artifacts.

The attenuation plot in Figure 3B does not show the theoretically expected decrease of the attenuation with increasing energy. But the increase of the attenuation in the copper regions after the copper K-edge at 8.99 keV is visible, enabling the identification of copper in the sample.

For the verification sample, the selection of regions for material decomposition attenuation values are straight forward as the materials are well-separated. For the plaque sample, a region representing iron was selected in an area with a high increase of attenuation but not in the outermost edge of the calcification where beam hardening is affecting the measured attenuation. In similarity with the increase in attenuation with energy of aluminum in Figure 3B, almost the whole calcification increased in attenuation in the upper energy window. Therefore, a region representing calcium was selected within the calcification with only a slight increase in attenuation. The selected regions are shown in Supplementary Figure 3.

The highest content of both iron and calcium was measured in the slice in plaque 1. Slices 14 and 257 in plaque 2 are from different calcifications in the plaque. Comparing the decomposition of these slices, it is clear that there are differences in the material content of calcifications within a plaque.

The implemented post-image reconstruction material decomposition method decomposes the sample into three basic materials. With the constraint that the material fractions should sum up to 1, a third material can be obtained from only two images. Hence, the addition of more energy windows should in theory enable the decomposition of more materials.

The method shows potential to be used as a pre-scanning method to identify regions of interest for further investigation, such as measurements at a synchrotron facility and histo-chemical analysis.

While the low energy range limits the sample size, it also allows for exploiting the K-edge attenuation discontinuity of naturally occurring elements. In the application of x-ray micro-tomography, the energy range applied here is compatible with soft tissue samples extracted in clinical routines.

Material decomposition attenuation values are manually selected in the reconstructed images. Therefore, a priori knowledge about the sample is necessary for calibration and the outcome of the decomposition is highly dependent on the operator. In this paper, the three plaque slices have been decomposed based on the same values from plaque 1 for relative comparison between the slices. To be noted is that a material fraction of 1 does not mean that the volume contains only one solid material but that it has the same content as the volume selected to represent 100% of the given material. For future work, a calibration sample with known concentrations of iron, calcium and paraffin could be constructed for quantitative data analysis. Decompositions performed with decomposition attenuation values from the known material concentrations would result in a material fraction corresponding to a known concentration. Hence, a calibration sample would enable quantification of the material concentrations, absolute comparison between samples and eliminate the dependency on the operator.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: www.osf.io/wmvn2.

The studies involving human participants were reviewed and approved by Local Swedish Ethical Review Authority (registration number 472/2005). The patients/participants provided their written informed consent to participate in this study.

The research project was proposed by MB and TD. Experiments were designed by TD and RS, and executed by RS. Data was obtained and analyzed by RS and IG. The manuscript was written by RS and TD, with revisions from MB and IG. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.