## Abstract

Background/Aim: The effective atomic number (Z_{eff}) and electron density relative to water (ρ_{e} or Rho) of elements can be derived in dual-energy computed tomography (DECT). The aim of this phantom study was to investigate the effect of different photon energies, radiation doses, and reconstruction kernels on Z_{eff} and Rho measured in DECT. Materials and Methods: An anthropomorphic head phantom including five probes of known composition was scanned under three tube-voltage combinations in DECT: Sn140/100 kV, 140/80 kV and Sn140/80 kV with incremented radiation doses. Raw data were reconstructed with four reconstruction kernels (I30, I40, I50, and I70). Rho and Z_{eff} were measured for each probe for all possible combinations of scan and reconstruction parameters. Results: DECT-based Rho and Z_{eff} closely approached the reference values with a mean and maximum error of 1.7% and 6.8%, respectively. Rho was lower for 140/80 kV compared with Sn140/100 kV and Sn140/80 kV with differences being 0.009. Z_{eff} differed among all tube voltages with the most prominent difference being 0.28 between 140/80 kV and Sn140/100 kV. Z_{eff} was lower in I70 compared with those of I30 and I40 with a difference of 0.07. Varying radiation dose yielded a variation of 0.0002 in Rho and 0.03 in Z, both considered negligible in practice. Conclusion: DECT comprises a feasible method for the extraction of material-specific information. Slight variations should be taken into account when different radiation doses, photon energies, and kernels are applied; however, they are considered small and in practice not crucial for an effective tissue differentiation.

Traditional single-energy computed tomography (SECT) lacks high sensitivity in soft tissue differentiation because the attenuation numbers of many organs and tissues show overlaps in Hounsfield unit (HU) measurements. These overlaps arise because the linear coefficients of attenuation of the tissues are similar and mainly depend on two physical effects: photoelectric absorption and Compton scattering. The photoelectric absorption refers to the energy of X-ray photons that interact with the tightly bound electrons of the inner-shell, specifically, the K-shell. These interactions lead to the absorption of X-ray photons while electrons are ejected (1-3). The atomic number Z corresponds to the number of protons in the nucleus. As the K-shell binding energy is proportional to Z of an element, the photoelectric effect depends strongly on Z of the scanned material (approximately proportional to Z^{3}) (3). Photoelectric absorption depends on the incident photon energy (beam energy). The more the initial X-ray photon energy approaches the K-shell binding energy or so called “K-edge”, the more probable the photoelectric effect is to occur (3). The photoelectric effect dominates at lower energies (4). Compton scattering refers to the ejection of the weakly bound outer shell electrons at different angles by the interaction with the X-ray photons. It is independent of photon energy at energies >30 keV, but depends on the mass density of the scanned material (4) and dominates at low Z (1-3). Prominent Compton scattering leads to loss of image contrast (3, 5, 6).

Dual-energy computed tomography (DECT) was first described in 1973 (2, 7). The objective, when used in clinical practice, is to isolate and quantify different elements for tissue differentiation by simultaneously or consecutively scanning materials (depending on the computed-tomography (CT) scanner used, *i.e.*, single or dual-source) in two different X-ray energy spectra (8). This concept allows the extraction of material-specific information based on the trends observed in the linear coefficients of attenuation within the different energy levels of the material. Differences that are not obvious by simple HU measurements on SECT may be revealed in the significant Z differences between different materials reflected in their coefficients (9). As the range between the applied radiation dose levels increases, the variations of coefficients of attenuation, and subsequently of their matrices, become more prominent and allow for a DECT-based tissue differentiation (9).

DECT has increasingly been applied during the past years for the elemental discrimination of urinary stones (10, 11), renal cell carcinoma (12), the differentiation of adrenal gland lesions (13), and the detection of uric acid crystals in patients with gout (14), among other medical conditions. Given the difference in HU (ΔHU) for the same element at different energy levels, both the effective atomic number (Z_{eff}) and electron density relative to water (Rho or ρ_{e}) can be derived with small errors through DECT Rho/Z imaging data (2, 15-17).

Computationally intensive algorithms known as reconstruction kernels are used to modify the frequency content of the image data prior to back projection during image reconstruction in a CT scanner. Kernels adjust the spatial resolution and therefore affect the image quality by sharpening or softening the image. Different kernels exist for the evaluation of different anatomical structures, *i.e.*, soft tissue and bone kernels. Evidence based on SECT have shown that the application of different reconstruction kernels affects attenuation measurements, especially of tissues with extreme low or high attenuation values taking as reference for the central HU regions, *i.e.*, that of water with 0 HU (18-21). As the spatial frequency of a reconstruction kernel increases, *i.e.*, as the kernel sharpens, edge-ringing artifacts appear and image noise increases leading to a broadened distribution of the attenuation values derived from the CT-image (18). Thus, softer kernels have generally been recommended for accurate attenuation measurements and for subsequent comparisons, for example, regarding the assessment of pulmonary emphysema (22). Hünemohr, *et al.* (23) observed a significant effect of image noise on Z_{eff} and ρ_{e} values measured in DECT with a standard deviation (SD) of 10% in the mass fraction predictions of carbon and oxygen (23).

The aim of this phantom study was to investigate the effect of different combinations of photon energy (tube voltage), radiation dose levels (tube current), and reconstruction kernels on Z_{eff} and Rho, the values of which are obtained in DECT Rho/Z imaging for tissue type differentiation.

## Materials and Methods

*Phantom scanning*. An anthropometric head CT calibration phantom (CIRS, model 711HN modified) of 16-cm diameter filled with tap water and included five cylindrical probes of known composition was used (Figure 1). These probes were substitutes for sinus cavities, soft tissue, brain, spinal cord, and trabecular bone (Table I) (24). DECT of the phantom was performed in a dual-source CT SOMATOM Definition Flash scanner (Siemens Healthineers, Erlangen, Germany). The following scanning parameter settings involving three different voltage combinations were applied: 1) Sn140/100 kV, ref.mAs 20/40/60/80/100/120/140/160/180/200, rotation time 0.5 s, pitch 1.0, collimation 128×0.6 mm; 2) 140/80 kV, ref.mAs 20/40/60/80/100, rotation time 1 s, pitch 1.0, collimation 128×0.6 mm; 3) Sn140/80 kV, ref.mAs 40/60/80/100/120/140/160/180/200, rotation time 0.5 s, pitch 1.0, collimation 128×0.6 mm (Sn denotes the use of a tin filter).

*Post-processing – application of different reconstruction algorithms and Rho/Z measurements*. Raw data obtained by all scans of the various voltage and current combinations of the tube were reconstructed in four different reconstruction kernels: I30 (medium smooth sharpness), I40 (medium sharpness), I50 (medium sharp), and I70 (very sharp). All Images were evaluated using Syngo.via^{®} software for multimodality reading (Syngo.via Dual Energy, Siemens Healthcare GmbH 2009-2018, Version 05.01.000.0030, Erlangen, Germany).

One reader with four years of experience in cross-sectional imaging performed the attenuation measurements in the Rho/Z application profile in Syngo.via^{®} (Siemens Healthineers, Forchheim, Germany). The electron density relative to water (HURho) and the effective atomic number (Z_{eff}) were measured in this profile for each probe situated in the middle of the cylinder after axis correction (Figure 2). The reader was free to adjust the windowing for the most optimal margin differentiation for each of the probes. The size of the regions of interest used were adjusted as large as possible in the axial slices through each probe but excluded the rims.

The relative density to water is encoded in HU_{Rho} in the Rho/Z maps of Syngo.via, therefore a linear relation, was used to extract the electron density values as proposed by Saito *et al.* (16). For simplification, is written Rho in the following text.

*Statistical analysis*. Microsoft Excel 2016 (Microsoft Corp., Redmond, WA, USA) was used for recording attenuation measurements and creating the graphs for this study. R statistics package (25) was used in the statistical evaluation of the results. Rho and Z_{eff} values for each photon energy and reconstruction kernel combination were averaged over the different radiation doses (tube currents); moreover, DECT-based measurements were subtracted from the known reference values for each probe for the estimation of the measurement error. With photon energy and reconstruction kernel as fixed effects and the phantom probe and radiation dose as random effects, a repeated-measures linear-mixed-effect model was applied to explore the effect of photon energy and reconstruction kernel on Rho and Z_{eff} values, respectively. Pairwise comparisons with Tukey’s Honest-Significant-Difference method were performed to compare the means among the distinct photon energy and reconstruction kernel groups. The level of significance was 0.05.

## Results

Rho and Z_{eff} values in the different photon-energy and reconstruction-kernel combinations were averaged over the different tube currents. Graphics of the Rho and Z_{eff} measurements of each probe are shown in Figure 3 and Figure 4, respectively.

*Rho*. Rho values measured in DECT after averaging all measurements over the different photon energies, radiation doses, and the reconstruction kernels that were applied are presented in Table I for each phantom probe. DECT-based Rho measurements closely approached the reference Rho values of each probe with the maximum measurement error being 0.014 (6.8%) for the LAA347 probe.

The results from random effects showed that the estimated SD of the Rho values in relation to radiation dose (tube current) was 0.0002 (95%CI=0-0.0007).

From the results for the fixed effects, the photon energy (tube voltage) showed a significant effect on Rho values (repeated ANOVA: *p*<0.001, Table II). Rho values were significantly lower for 140/80 kV compared with those for both Sn140/100 kV and Sn140/80 kV, the Rho difference being 0.009 between 140/80 kV and Sn140/100 kV and 0.008 between 140/80 and Sn140/80 kV (pairwise comparison with Tukey’s method: both *p*<0.001, Table II). No significant differences were noted between Sn140/100 kV and Sn140/80 kV energies with the difference in Rho being notably lower (0.002) (pairwise comparison with Tukey’s method: *p*=0.3).

There was no significant effect of the different reconstruction kernels on Rho values (repeated ANOVA: *p*=0.99, Table II).

*Z _{eff}*. DECT-based Z

_{eff}measurements closely approached the reference Z values for each probe with the maximum measurement error being 0.13 (1.3%) for the DTB109 probe (Table I).

The results of the random effects showed that the estimated SD of the Z values in relation to radiation dose (tube current) was 0.03 (95%CI=0.02-0.06).

Similarly, from the results on the fixed effects, both photon energy (tube voltage) and reconstruction kernel were found to have a significant effect on Z_{eff} values (repeated ANOVA for both radiation dose and reconstruction kernel: *p*<0.001, Table II):

The Z_{eff} values were significantly different between all three different photon energies (pairwise comparisons with Tukey’s method for all three pairs: *p*<0.001), whereas the most prominent differences were found between 140/80 kV and Sn140/100 kV (0.28) and 140/80 kV and Sn140/80 kV (0.2). The Z_{eff} difference between Sn140/100 kV and Sn140/80 kV was less prominent (0.09) but also statistically significant (Table II).

Z_{eff} values were significantly lower in I70 compared with those of the I30 and I40 kernels (Tukey’s pairwise comparison: both pairs *p*<0.001 with both Z_{eff} differences being approximately 0.07). No further significant differences in Z_{eff} values were found between the other kernels (Tukey’s pairwise comparisons: *p*>0.05, Table II).

## Discussion

To the best of our knowledge, this analysis is the first attempt to assess the possible effects of different radiation doses, photon energies, and reconstruction algorithms in the DE Rho/Z imaging measurements for material differentiation. This phantom study showed that the application of different radiation doses, photon energies, and reconstruction kernels might cause slight variations in Rho and Z_{eff} measurements on DECT. Some differences in Rho and Z_{eff} values among the different groups, although statistically significant, were small (maximum difference for Rho was 0.009, for Z_{eff} 0.28, both between Sn140/100 and 140/80 kV tube voltages) and their practical relevance should be considered.

The Rho and Z_{eff} values obtained by the DECT-measurements closely reflected the reference Rho and Z_{eff} values provided by the manufacturer for each phantom probe (Table I). The measurement errors varied between 0.1%-6.8% for Rho and 0.4%-1.3% for Z_{eff}. These percentages agree with previous studies in the literature presenting low measurement errors of Rho/Z imaging (2, 16, 17, 26) indicating that DECT offers a feasible method for the extraction of material-specific information.

The photon energy as determined from the tube voltage was shown to have an effect on Rho/Z imaging. The 140/80 kV combination primarily showed significantly lower Rho values compared with both Sn140/100 kV and Sn140/80 kV. Spectral shaping for increased energy spectra separation is applied as an alternative to the standard low-dose protocols to reduce the radiation dose in constantly kept high tube voltages to achieve an adequate image quality with reduced image noise. This is attained using a tin filter placed between the standard aluminum filter and the patient to absorb the low-energy photons that are irrelevant in high-contrast imaging, for instance in that of bone or calcium urinary stones (27, 28). With the tin filter, a decrease in radiation-dose overlap of the two applied energy spectra occurs allowing for better tissue discrimination (29, 30). The tin filter is known to lower HU attenuation measurements and subsequently lower the contrast-to-noise ratio of an image, which is proportional to the HU attenuation of the structure under examination (29, 31). Rho is proportional to ΔHU between the two radiation spectra, (16). Taking into account that the radiation dose decreases when the tin filter is applied, the HU attenuation measurements (30) and, therefore, the ΔHU are expected to decrease as well, leading to lower Rho values as observed in this study.

The photon energy showed the exact opposite effect on Z_{eff} values compared with Rho, where the 140/80 kV combination was associated with higher Z_{eff} values compared with both Sn140/100 kV and Sn140/80 kV. This may be explained by the fact that Z_{eff} is inversely proportional to the HU attenuation and the Rho values according to the proposed algorithms for Z_{eff} calculation based on Rho on DECT (16, 32, 33). These algorithms are commercially employed in the Rho/Z Maps application profile of the Syngo.via^{®} software that was used for the measurements performed in this study. The Z_{eff} values could not be estimated in DECT for the LAA347 probe, the substitute for a sinus cavity mainly containing air, because of a known instability in the algorithm appearing in the ρ_{e}/ρ_{e,w} component required for the calculation of Z_{eff} (32).

The radiation dose as determined by the tube current was statistically handled as a random effect in the linear models used because of the small sample size within the distinct tube-current groups. This small size impedes a reliable statistical analysis for the extraction of a *p*-value regarding the effect of radiation dose on the Rho and Z_{eff}. Therefore, a descriptive statistical approach was implemented using the calculation of the SD of Rho and Z_{eff} within each tube voltage group to reflect the variation incurred with the different tube currents applied. The SDs of both Rho (0.0002) and Z (0.03) are very small and therefore considered negligible from a practical point of view indicating that varying the radiation dose does not have a significant effect on Rho and Z_{eff} values critical in tissue and material differentiation based on DECT techniques.

The different reconstruction kernels did not significantly affect the Rho measurements in DECT. Regarding Z_{eff}, the sharpest I70 was the only kernel found to be related with significantly lower values compared with the smoother I30 and I40 kernels. Sharper kernels provide a higher spatial resolution and image contrast at the expense of higher image noise and subsequently lower image quality (34), which may generate more prominent variations in the HU measurements on CT. Nevertheless, despite higher radiation doses being associated with better image quality, the most prominent variation in Z_{eff} in regard to I70 was unexplainably observed for 200 mAs (Figure 4), even after repeated measurements were performed to exclude possible measurement errors. However, all differences reported to be statistically significant in this study were small (maximum differences: for Rho 0.009, for Z_{eff} 0.28). In such instances, we can consider whether these small differences, although statistically significant, are effectively relevant and crucial in practice for material- and tissue-differentiation processes based on DECT. The highest significant difference for Rho was found to be 0.009±0.007 and for Z_{eff} 0.28±0.09, both between Sn140/100 kV and 140/80 kV photon energies. The presence of such small differences should always be considered in Rho and Z_{eff} evaluations when applying different tube voltages or reconstruction kernels; however, it is believed that for an effective tissue differentiation they may not be essential or misleading in practice.

The main limitation of this study is that it concerns a phantom study and current results may differ for other tissue types and human organs; this awaits further study. In addition, the X-ray tube of the CT-scanner used did not allow for a stepwise variation in the radiation dose between the range 100 to 200 mAs in the 140/80 kV combination (unlike in Sn140/100 kV and Sn140/80 kV) because of dangers in overheating; therefore, a 140/80 kV radiation dose could reach a maximum value of 100 mAs.

## Conclusion

Varying radiation dose, photon energy, and reconstruction kernel may have an effect on Rho and Z_{eff} measurements in Rho/Z imaging applications performed in DECT for material and tissue differentiation. The application of a tin filter leads to a variation of approximately 0.009 in Rho and 0.28 in Z_{eff}. The presence of these possible slight variations should be taken into account when evaluating Rho and Z_{eff} under different combinations of scan parameters and reconstruction algorithms that take place; however, the expected differences are considered small and in practice not critical for an effective DECT-based tissue differentiation.

## Acknowledgements

V.C. has received a scientific grant from Guerbet, Zürich, Schweiz. The Authors thank Richard Haase, PhD, from Edanz Group (www.edanzediting.com/ac) for editing a draft of this manuscript.

## Footnotes

**Authors’ Contributions**Study design and methodology (VC, AB, RK, BS, AL, DW, MT, TN), design and construction of phantom (VC, AB, AL, TN), image acquisition (VC, TN), Image analysis (VC, RK, TN), Data analysis (VC, RK, MT, TN), Writing the manuscript/statistics (VC, AB, RK, BS, AL, DW, MT, TN). All Authors have read and agreed to the published version of the manuscript.

**Conflicts of Interest**Siemens Healthineers provided technical support for this study. One author is an employee of Siemens Healthineers (B.S.). He had no involvement in the study design; collection, analysis, and interpretation of data; writing of the report; or the decision to submit the article for publication. There is no further conflict of interest for any of the Authors.

- Received January 13, 2022.
- Revision received January 28, 2022.
- Accepted January 31, 2022.

- Copyright© 2022, International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved