Photon-Counting Computed Tomography-Based Hepatic iron Quantification Using a Tungsten-Based Contrast Agent

Abstract

Objectives:

This study explores the potential of quantifying hepatic iron in computed tomography (CT) scans in the presence of iodine (I)- or tungsten (W)-based contrast media (CM).

Materials and Methods:

Experimental work was performed on a commercial photon-counting CT system able to simultaneously acquire up to 4 spectral data sets in a single scan. We examined 2 anthropomorphic abdominal phantoms with material samples of liquid liver tissue surrogate, fat, iron, and I- or W-based CM to mimic different liver compositions in an enhanced CT scan. Iron was quantified by material decomposition of reconstructed spectral CT images.

Results:

Two-material decomposition based on 2 spectral data sets provided material images of iron and liver with an accuracy of 1.4 mg/mL in the iron image of CM-free samples. The presence of W affected the iron quantification: For 2 and 4 mgW/mL in the material samples, the iron concentration was overestimated (P < 0.05) with accuracies of 2.7 and 4.7 mg/mL, respectively. Three-material decomposition based on 4 spectral data sets provided material images of iron, liver, and W, with an accuracy of 1.4 mg/mL in the images without W and 1.5 (nonsignificant difference, P > 0.07) and 1.6 mg/mL (overestimation, P > 0.03) in the iron image at 2 and 4 mgW/mL, respectively. The presence of I affected the iron quantification more than W in both 2- and 3-material decomposition: For 2 and 4 mgI/mL in the material samples, the measured iron concentration was even higher (P < 0.05), with accuracies >18 and >37 mg/mL, respectively.

Conclusions:

The accuracy of iron quantification from a 3-material decomposition suggested clinically feasible detection and quantification of critical hepatic iron levels in enhanced CT scans with suitable CM. In a 2-material decomposition, severe pathology is required to detect an iron liver. W-based CM was superior to I-based CM.

Iron overload in the liver can result from genetic, acquired, or other causes. The most common genetic cause is hereditary hemochromatosis, often linked to high-iron (HFE) gene mutations, which increase intestinal iron absorption. Non-HFE mutations can also lead to hereditary forms. Acquired causes include excessive iron supplementation; chronic blood transfusions, such as in thalassemia or sickle cell disease; and chronic liver diseases, such as nonalcoholic fatty liver disease or hepatitis C. Other causes include the dysmetabolic iron overload syndrome, juvenile hemochromatosis, and neonatal hemochromatosis. Iron overload can lead to oxidative stress and hepatic fibrosis and increases the risk of cirrhosis and hepatocellular carcinoma.^1,2

Early detection and quantification of hepatic iron overload can thus prevent irreversible organ damage.^1,3 Measuring the hepatic iron concentration (IC) is particularly relevant in view of the current clinical routine that assesses the plasma ferritin concentration. However, plasma ferritin has several limitations because it is not only a marker of iron storage but also a marker of acute reactions such as infections.^4,5 Magnetic resonance imaging (MRI) or spectroscopy and dual-energy computed tomography (DECT)-based methods are noninvasive alternatives to determine the hepatic IC^6,7 and are being increasingly used because of their accuracy, safety, and reproducibility.

For example, MRI-based methods such as R2 relaxometry allow for measuring the hepatic IC via tissue signal decay rates, which correlate with iron content.⁸ Standardized methods using FerriScan (Resonance Health, Burswood, WA, Australia) have been validated against biopsy results and are used for clinical purposes. All of these techniques can also monitor changes in hepatic IC over time. Additionally, MRI has the potential for whole-organ assessment, capturing heterogeneous iron distribution, which may be missed by biopsy. However, MRI has limitations. MRI is susceptible to motion artifacts, liver comorbidities, and intercenter variability, making standardization a challenge. In addition, MRI is contraindicated for patients with certain implanted devices or severe claustrophobia and requires expert interpretation.

Computed tomography (CT) has also been explored as a potential tool for detecting hepatic iron overload.^9,10 The main advantages of CT include its widespread availability, speed, ability to provide structural information on the liver and other abdominal organs, and ability to derive 3D volumetric data.¹¹ DECT has been shown to accurately assess clinically significant hepatic IC on unenhanced abdominal CT scans.^4,12,13 A 34-patient DECT study indicated diagnostic performance of virtual iron content (VIC) imaging to be comparable to R2* MRI at clinically significant hepatic IC levels for initiating or monitoring iron chelation therapy. At a hepatic IC level of 7.0-mg/g dry tissue, both VIC sensitivity and specificity were 100%, making VIC imaging a potential tool for guiding therapy. However, VIC sensitivity was less than R2* at lower hepatic IC levels, limiting its utility for detecting mild iron accumulation.¹⁴

In DECT, however, scans without contrast medium (CM) injection can usually be replaced by virtual noncontrast images of an enhanced scan. This replacement ensures an optimal spatial fit of unenhanced and enhanced images, saves radiation dose, and relieves patient compliance.^15–17 Accurate iron quantification on enhanced liver images using CT requires 2 things: first, a CT system able to separate 3 materials, such as the first commercial PCCT system, which allows a prior knowledge-free multimaterial decomposition (MD) in a single scan and thus for separating 3 materials.¹⁸ Second, a CM which is distinguishable from iron and soft tissue. To date, only CM based on iodine (I) and barium have been approved for clinical use.^19,20 Elements with higher atomic numbers can be more readily distinguished from surrounding highly attenuating body tissues, implanted materials such as stents, and other elements such as iron. A promising approach for multimaterial imaging of the human body is a CM incorporating tungsten (W) with atomic number 74.^20,21

This study uses the novel CT system to investigate ICs in a liver-like environment based on CT scans enhanced by W or I. For this, highly standardized liquid liver tissue surrogate (LTS) solutions were used, which mimic the x-ray attenuation of human liver tissue.²² These LTS solutions were mixed with CM at various concentrations to simulate contrast-enhanced liver tissue.²³

MATERIALS AND METHODS

Materials

LTS solutions were prepared based on the method described previously.^22,23 LTS consists of glycerin (28% mass fraction), methanol (9%), sodium chloride (0.6%), potassium nitrate (1.2%), and water (61.2%) and has been validated against hepatic tissue.²² Hepatic fat was simulated by a mixture of butanol (59.5%) and glycerin (40.5%). Iron in the form of Fe-DTPA (Sigma Aldrich, St. Louis) was added to obtain samples with an IC of 1, 2, 3, 5, 8, and 11 mg/mL and an iron-free control sample. These ICs, analytically quantified by inductively coupled plasma mass spectrometry, represent wet weight concentrations while the hepatic IC for therapy recommendations is often related to dry tissue weight. With a conversation factor of 3, the IC range covers 3–33-mg/g dry tissue and covers the range from normal to highly increased IC.²⁴ An increased risk for complications such as hepatic fibrosis or even more severe cardiac iron-induced cardiac diseases was reported for IC > 7 and >15-mg/g dry tissue.⁵

For each IC, 5 samples with different CM content were prepared: 2 mgI/mL and 4 mgI/mL (Ultravist; Bayer, Leverkusen, Germany), 2 mgW/mL and 4 mgW/mL,²¹ and a CM-free control sample. In sum, we prepared 35 material samples for 100% LTS and 35 samples with 80% LTS and 20% fat. W has a k-edge at 69.5 keV, which is substantially higher than that of I (33.2 keV), cf. Figure 1. This results in higher x-ray attenuations at higher photon energy levels and leads to a better basis for discrimination against iron (k-edge at 7.11 keV), which has a much lower x-ray attenuation in this energy range. The nonzero CM concentrations represent typical CM concentrations of healthy patients during portal-venous phase CT acquisitions. Contrast-enhanced CT requires a parenchymal enhancement of at least 50 HU during portal-venous phase.²⁶ For conventional CT, this requires a local I concentration of 2 to 4 mg/mL, depending on tube voltage and patient size.^27,28 The attenuation of W is at least comparable or higher than that of I.²⁰ Thus, the concentrations of 2 and 4 mgW/mL used are clinically realistic.

FIGURE 1.

Attenuation of various metals relative to water.²⁵ The peak of the tungsten curve amid the relevant energy range promises a better separation between tungsten and iron compared with iodine and iron.

All samples were mixed and shaken using a vortex mixer immediately prior to the CT scan to evenly distribute the iron. The samples were then placed in an abdominal phantom as shown in Figure 2.²⁹ This phantom, which is 300 mm wide, 200 mm high, and 103 mm deep, is made of tissue-, liver-, spleen-, and bone-equivalent materials. It can be enlarged with a fat-equivalent extension ring, 400 mm wide and 300 mm high, to mimic different patient habitus.

FIGURE 2.

Abdominal phantom with width of 300 mm, height of 200 mm, and depth of 103 mm, consisting of body-equivalent materials, with material samples. The phantom can be extended with a fat-equivalent extension ring with a width of 400 mm and a height of 300 mm.

Image Acquisition of Material Samples

To acquire input images for MD, CT scans were performed on a first-generation clinical PCCT system (NAEOTOM Alpha; Siemens Healthineers AG, Forchheim, Germany; software version VB10). A dedicated spiral scan protocol, based on the factory scan protocol “Body Research” for research purposes, with a tube voltage of 140 kVp, rotation time of 0.25 s, pitch 0.6, and axial collimation of 96 × 0.4 mm generates 4 threshold images with threshold energies 20, 55, 72, and 90 keV, denoted T1, T2, T3, and T4. This combination of energies is best suited to W among the available energy quadruples.³⁰ Scans with and without the obese extension ring, and with 2 different exposures in terms of tube current were performed to represent different patient sizes and dose levels and to test the method's reliability. The phantom was placed on the patient table such that the center of the enclosed water cylinder coincided with the isocenter of the gantry. Image reconstruction used the quantitative kernel Qr40 and a field of view of 128 mm to maximize the resolution of the regions of interest during image analysis. The reconstruction produced slices with a thickness of 2.0 mm and an increment of 1.0 mm, as in previously published abdominal examinations.¹⁷ Table 1 summarizes the scan and reconstruction parameters.

TABLE 1.

Scan and Reconstruction Parameters Used for the Samples With Liver Tissue Surrogate. All Samples Were Scanned Once in a Small and Once in a Large Abdominal Phantom, the Former With Width and Height of 200 and 300 mm and the Latter With 300 and 400 mm, Each at 2 Dose Levels

Phantom	Small	Large
Exposure, eff. mAs	200, 75	300, 100
Tube voltage, kVp	140
Rotation time, s	0.25
Spiral pitch factor	0.6
Axial collimation, mm	96 × 0.4
Threshold energies, keV	20, 55, 72, 90
Iterative reconstruction level, QIR	2
Convolution kernel	Qr40
Field of view, mm	128
Slice thickness, mm	2.0
Slice increment, mm	1.0

Material Calibration

MD requires prior material calibration, that is, the knowledge of the image contrast per unit of material concentration, which typically depends on several variables, such as the tube voltage, the selected energy thresholds and the water-equivalent object diameter. We applied a combined approach of material calibration and MD previously published by Neumann et al.³⁰ The workflow used in this study is identical except for the differences explained in the following 2 paragraphs.

The calibration process involves modeling material CT numbers and the image noise covariance matrix (INCM) for efficient image-based MD. To calibrate those, we performed the calibration procedure described by Neumann et al using aqueous solutions of each metal and scan setups shown in Table 2. To predict LTS and fat CT numbers, models were calibrated using the measurements of the sample containing 100% LTS and a sample containing 100% fat: Averaged CT numbers from cylindrical volumes with a diameter of 10 mm over the central 20 reconstructed image slices were used to calibrate an affine-linear model in the water-equivalent object diameter and the x-ray tube current, 1 model for LTS and 1 for fat.

TABLE 2.

Altered CT Scan Setup for the Calibration Measurements Compared With a Previously Published Setup

Parameter Scan Mode	Value Sequence
Phantom diameter, mm	100	200	300	400
X-ray tube current, mA	25, 50, 75, 100	100, 200, 300, 400	350, 500, 650, 800	500, 600, 700, 800
Object off-centering, mm	0, 25, 50

Material Decomposition

The MD is an image voxel-wise process on each CT number N-tuple, which is constructed by aligned voxels in N reconstructed threshold images. This generates M ≤ N material images. This study performed MDs for 2 different cases: First, we performed a 2-material decomposition (M = 2), hereafter called 2MD, into LTS and iron based on N = 2 threshold images. To optimize decomposition results, the 2MD's input were selected as T1, which is the noise-minimal reconstructed series, and T4, which provides the best spectral separation between the input series. Second, we performed a 3-material decomposition (M = 3), hereafter called 3MD, into LTS, iron, and W based on all (N = 4) threshold images for better separating iron and CM. In both cases, the LTS prevalence in the material samples required a modification of the MD procedure compared with the prior art as the section Supplemental Material (http://links.lww.com/RLI/A975). The MD resulted in voxel-wise material coefficients $\overrightarrow{\rho }\in {ℝ}^M$ with the amount of LTS ρ ₁ and the iron and W concentration ρ ₂ and ρ ₃. The value ρ ₁ = 1 corresponds to 100% LTS and ρ ₁ = 0 to 100% fat.

Quantitative Analysis

After generating material images with 2MD and 3MD, the IC measured in the material samples was compared with the ground truth. For this purpose, the IC was averaged in a circular region of 10 mm diameter for each material sample in the central 20 iron image slices of each decomposed reconstructed image series. The standard deviation (SD) of the image noise was measured in the same circular regions. For ease of comparison in the samples where measurements and ground truth differed by only a few mg/mL, the differences were visualized using Bland-Altman plots.³¹ Confidence intervals for the mean difference were provided.³¹ The comparison was expected to reveal a discrepancy between measurements and ground truth, as the section Appendix (http://links.lww.com/RLI/B34) describes. A Wilcoxon signed rank test was performed to compare the measured IC for different sample CM contents in the material samples, with P < 0.05 considered to be statistically significant. Analysis was performed using MATLAB (The MathWorks Inc., Natick, MA; software version R2024a).

RESULTS

Figures 3–5 show exemplary input and output images of the MD. The LTS images (Figs. 4A, 5A) display approximately 1 in the liver-equivalent phantom part and a lower coefficient in the soft tissue- and water-equivalent parts. The iron images (Figs. 4B, 5B) display approximately 0 mg/mL in the liver-equivalent phantom part, a few mg/mL in the soft tissue- and water-equivalent parts, and more than 50 mg/mL in the air gap between the syringes and the phantom. The W image (Fig. 5C) displays nonzero concentrations right in the W-containing samples on the top and the right. The LTS coefficients the I-containing samples on the left were lower and the IC higher than expected. The 2MD overestimated the LTS content in the W-containing samples.

FIGURE 3.

Exemplary input images (A) T1, (B) T2, (C) T3, and (D) T4 of the decomposition algorithm from 1 high-dose scan of the small phantom with abdominal window 40/300. The phantom was fed with the samples of iron/tungsten/iodine concentration of 5/2/0 (top), 5/4/0 (right), 8/0/0 (bottom), and 8/0/2 (left) mg/mL.

FIGURE 5.

Exemplary output images of a 3-material decomposition into (A) liver tissue surrogate, (B) iron, and (C) tungsten based on 4 threshold images from the scan described in Figure 3

FIGURE 4.

Exemplary output images of a 2-material decomposition into (A) liver tissue surrogate and (B) iron based on the threshold images T1 and T4 from the scan described in Figure 3.

Iron Quantification

Figures 6 and 7 show absolute and relative ICs evaluated from the 2MD. Figures 8 and 9 show absolute and relative ICs evaluated from the 3MD. Tables 3 and 4 summarize the IC accuracy of the 2MD and 3MD in terms of the difference between the measured and ground truth concentration for each combination of iron and CM concentration in the sample. The tables also show confidence intervals for the mean difference for each CM concentration.

FIGURE 6.

Iron measurement from a two-material decomposition plotted versus the ground truth in samples with 2 mgW/mL (◊), 4 mgW/mL (×), 2 mgI/mL (∇), 4 mgI/mL (∆), and no contrast agent (○) for 100% (A–D) and 80% (E–H) liver tissue surrogate, the small (A, B, E, and F) and the large (C, D, G, and H) phantom and high (A, C, E, and G) and low (B, D, F, and H) exposure. The legend indicates the R-squared value of a tangent fit for each graph (P < 0.01).

FIGURE 7.

Difference of the ironmeasurement froma 2-material decomposition and the ground truth plotted against their mean in samples (A) without contrast medium, (B) with 2 mgW/mL, and (C) with 4 mgW/mL. MEAN and SD are the arithmetic mean and standard deviation of all differences.

FIGURE 8.

Iron measurement from a 3-material decomposition plotted versus the ground truth in samples with 2 mgW/mL (◊) and 4 mgW/mL (×), 2 mgI/mL (∇), 4 mgI/mL (∆), and no contrast agent (○) for 100% (A–D) and 80% (E–H) liver tissue surrogate, the small (A, B, E, and F) and the large (C, D, G, and H) phantom and high (A, C, E, and G) and low (B, D, F, and H) exposure. The legend indicates the R-squared value of a tangent fit for each graph (P < 0.01).

FIGURE 9.

Difference of the iron measurement from a 3-material decomposition and the ground truth plotted against their mean in samples (A) without contrast medium, (B) with 2 mgW/mL, and (C) with 4 mgW/mL. MEAN and SD are the arithmetic mean and standard deviation of all differences.

TABLE 3.

Iron Measurements for the Different Combinations of Sample iron and Contrast Medium Content in the 2-Material Decomposition. All Values Are in Units of mg/mL

Contrast Medium	Iron	0	1	2	3	5	8	11	Mean RMS error	CIB
w/o W, I	Mean	−0.7 ± 0.7	−0.1 ± 1.1	2.8 ± 1.1	2.9 ± 0.9	5.3 ± 1.3	8.8 ± 1.3	12.4 ± 1.3
	RMS error	1	1.5	1.3	0.8	1.3	1.4	1.8	1.4	[−0.2, 0.6]
2 mgW/mL	Mean	2.5 ± 1.4	2.2 ± 1.0	3.9 ± 1.1	6.4 ± 1.1	8.1 ± 0.5	9.9 ± 1.3	13.8 ± 1.1
	RMS error	2.9	1.6	2.1	3.5	3.1	2.2	3	2.7	[2.1, 2.7]
4 mgW/mL	Mean	3.2 ± 1.5	4.9 ± 1.2	6.9 ± 1.0	7.1 ± 0.8	9.2 ± 1.2	13.0 ± 1.9	17.0 ± 1.0
	RMS error	3.5	4	5	4.1	4.3	5.3	6	4.7	[4.1, 4.8]
2 mgI/mL	Mean	19.0 ± 1.9	20.5 ± 1.3	19.9 ± 1.0	20.8 ± 1.7	24.7 ± 1.6	27.8 ± 0.9	30.2 ± 1.4
	RMS error	19.1	19.6	18	17.9	19.8	19.8	19.2	19.1	[18.6, 19.4]
4 mgI/mL	Mean	37.3 ± 2.2	37.8 ± 1.9	38.9 ± 3.1	40.3 ± 2.3	43.9 ± 2.1	46.0 ± 2.5	49.9 ± 3.9
	RMS error	37.3	36.9	37	37.3	39	38.1	39.1	37.8	[37.0, 38.4]

Mean, average of the different combinations of phantom, exposure, and liver tissue surrogate content in the samples using the arithmetic mean. The uncertainty is the standard deviation of the averaged values.

CIB, 95% confidence interval for the bias; I, iodine; W, tungsten.

TABLE 4.

Iron Measurements for the Different Combinations of Sample iron and Contrast Medium Content in the 3-Material Decomposition. All Values Are in Units of mg/mL

Contrast Medium	Iron	0	1	2	3	5	8	11	Mean RMS error	CIB
w/o W, I	Mean	−0.6 ± 0.7	0.1 ± 0.9	2.9 ± 1.0	2.9 ± 0.8	5.5 ± 1.3	9.1 ± 1.1	13.0 ± 1.1
	RMS error	0.9	1.3	1.3	0.8	1.4	1.5	2.2	1.4	[0.0, 0.8]
2 mgW/mL	Mean	0.8 ± 1.2	0.6 ± 0.8	2.1 ± 1.1	4.9 ± 1.2	6.5 ± 0.6	8.4 ± 1.1	12.4 ± 1.2
	RMS error	1.4	0.9	1	2.2	1.7	1.1	1.8	1.5	[0.5, 1.2]
4 mgW/mL	Mean	−0.2 ± 0.8	1.5 ± 1.3	3.5 ± 1.3	3.5 ± 0.8	6.0 ± 0.6	9.5 ± 1.3	13.3 ± 0.6
	RMS error	0.7	1.3	1.9	0.9	1.2	1.9	2.4	1.6	[0.7, 1.4]
2 mgI/mL	Mean	19.3 +/− 1.7	20.7 +/− 1.2	20.3 +/− 1.0	21.3 +/− 1.7	25.1 +/− 1.5	28.2 +/− 0.9	30.8 +/− 1.2
	RMS error	19.4	19.8	18.3	18.3	20.1	20.2	19.8	19.4	[19.0, 19.8]
4 mgI/mL	Mean	37.9 +/− 2.0	38.2 +/− 1.8	39.4 +/− 2.9	40.7 +/− 2.2	44.5 +/− 1.9	46.3 +/− 2.3	50.2 +/− 3.5
	RMS error	37.9	37.2	37.5	37.7	39.5	38.4	39.3	38.2	[37.5, 38.8]

Mean, arithmetic mean over the different combinations of phantom, exposure, and liver tissue surrogate content in the samples. The uncertainty is the standard deviation of the averaged values.

CIB, 95% confidence interval for the bias; I, iodine; W, tungsten.

In the 2MD iron images, the SD values were 7.9, 10.6, 15.9, and 24.5 mg/mL for the phantom exposure combinations small-high, small-low, large-high, and large-low. The corresponding SD values of the 3MD were 7.2, 9.3, 14.8, and 21.5 mg/mL. The propagated nonlinearity of the sample CT numbers, described in the section Appendix (http://links.lww.com/RLI/B34), was 1.3 and 1.4 mg/mL on average for the 2MD and the 3MD, respectively.

In the 2MD, for each sample IC, the mean measured IC in the CA-free samples, shown in Table 3, agreed with the sample IC to within 0.1–1.4 mg/mL. The average accuracy in terms of RMS error was 1.4 mg/mL, slightly above the propagated nonlinearity. For samples at 2 and 4 mgW/mL, the IC was significantly (P < 0.05) overestimated, apparent from Figure 7, with average accuracies of 2.7 and 4.7 mg/mL.

In the 3MD, for each sample IC, the mean measured IC in the CA-free samples, shown in Table 4, agreed with the sample IC to within 0.1–2.0 mg/mL. The average accuracy was 1.4 mg/mL, consistent with the propagated nonlinearity. The measured values tended to be overestimated, but the trend was not significant (P > 0.4). For samples at 2 and 4 mgW/mL, the IC tended to be overestimated with average accuracies of 1.5 and 1.6 mg/mL, which is below most of the sample ICs. The overestimation was not significant (P > 0.07) for 2 mgW/mL but significant (P > 0.03) for 4 mgW/mL.

For the 2MD and 3MD, the ICs in samples at 2 and 4 mgI/mL were significantly (P < 0.05) overestimated with mean accuracies of >19 and >37 mg/mL, respectively. The overestimation was also significantly (P < 0.05) higher than for W-containing samples. The RMS errors are multiples compared with the W-containing samples at the same concentration and, in particular, well above the sample ICs.

Viewing the row of CA-free samples in Table 3 and Table 4, the tendency to underestimate or overestimate ICs was consistent between the 2MD and the 3MD. For instance, the 2MD and the 3MD measured −0.7 and −0.6 mg/mL, respectively, in the metal-free samples on average, both below the sample IC of 0 mg/mL.

For the 2MD and 3MD, the sample IC was lower than the SD in the iron image slices for most scan setups. The variation of the measured IC over the different combinations of LTS content, phantom, and exposure was smaller than the RMS error for most samples and much smaller than the SD for all samples in all scan setups.

DISCUSSION

Accurate quantification of hepatic iron is essential for diagnosing, monitoring, and managing conditions associated with iron overload. In CT, hepatic iron is typically measured on unenhanced scans. A DECT study, however, examined hepatic IC in 11 patients who underwent multiple scans, among which were enhanced scans, without histologic verification of the results.³² In contrast, the presented CT-based phantom study examined IC in the presence of CM in material samples of known constitution. LTS samples mixed with different ICs were decomposed into multiple materials to quantify iron in the samples. This frugal model allows for standardized and reproducible conditions needed to develop and validate PCCT-based IC measurement techniques.

Several methods for MRI-based iron quantification exist, but the R2* relaxometry is the most widely used approach. Systematic studies of its accuracy are not available and are complicated because the biopsy as a reference standard represents only a local concentration. The diagnostic accuracy of the R2 method compared with DECT obtained in 34 patients showed a significant correlation of the hepatic IC and a similar diagnostic performance.¹⁴ For the presented iron quantification based on PCCT and MD, no comparative data to MRI is available. The costs for both, MRI and PCCT methods, are significantly higher than for the standard laboratory quantification in plasma. Once the method is more established, the fast scan time and the quantitative nature of CT may result in lower costs. To date, however, the availability of PCCT is limited to research centers.

We applied a workflow previously published by Neumann et al.³⁰ Building on their work, this study investigated material samples with a wider range of metal concentrations and used samples which, due to the LTS dominance, more closely resemble human tissue than aqueous metal solutions. The decomposition algorithm imposes no boundary conditions on the material coefficients, that is, it allows and expects an LTS coefficient ρ ₁ < 0 and >1 and an IC ρ ₂ < 0 mg/mL depending on the image noise and the combination of base materials and material content of the decomposed volume.

Liver-equivalent phantom parts and the LTS material sample decomposed uniquely into LTS, as expected. Other regions without a k-edge material, thus the rest of the phantom, air, and W-free samples, induced contributions only to the LTS and iron image. This is because materials uniquely interacting by photoelectric effect and Compton effects decompose into any set of 2 spectrally independent non–k-edge materials.

In 2MD, CM were not in the set of base materials and thus distorted the measurements of LTS and IC. Owing to the MD's linearity, the distortion increased correspondingly with increasing CM concentration, in line with the increased RMS error of 4 mgW/mL compared with 2 mgW/mL and likewise for I. Errors ensuing from W were in the same order of magnitude as the IC to be determined. In 3MD, the mean RMS error on the IC measurements was less than 2 mg/mL in I-free samples. In 2MD and 3MD, at equal CM concentration, W-containing samples were quantified much more accurately than I-containing samples as expected because of the similar attenuation behavior of I and iron.³² Moreover, the IC overestimation in samples containing I based on 2 spectral data sets is consistent with the results of Basso et al.³²

In 2MD of CM-free samples and 3MD of I-free samples, the effect of sample CT numbers nonlinear in the metal concentrations was in the order of magnitude of the IC measurement errors and can thus be considered the dominant source of systematic errors. Although all sample compounds were base materials, the IC measurements were prone to overestimation, which was significant for samples with 4 mgW/mL in the 3MD. The positive bias can be attributed to the beam hardening effect of nonwater-equivalent phantom materials and highly absorbing material samples, both of which cause weaker attenuation of hard radiation than predicted for the same water-equivalent object diameter. Due to the higher attenuation in the sample of 4 mgW/mL, the effect is slightly stronger here compared with 2 mgW/mL, bringing the overestimation above the significance threshold.

The consistent tendency to underestimate or overestimate IC in the CA-free samples between the 2MD and the 3MD reflects the absence of W in the material samples, which makes the third base material W obsolete in the 3MD. Differences in the results between the 2MD and the 3MD in these samples were thus due only to image noise.

The SD was lower in the images resulting from the 3MD than from the 2MD due to the higher number of spectral images used for the 3MD. In both 2MD and 3MD, the SD of the iron images exceeded the sample IC in most scan setups. However, the standard error on the IC mean can be expected to be smaller than most of the sample ICs, given the small variation of the measured IC over the different combinations of LTS content, phantom, and exposure. In human studies, with the evaluation of larger volumes in the liver compared with the material samples investigated in this study, the random error on the IC measurements owing to noise will be smaller, leading to higher accuracy of the measurements.

Considering the risk of complications from excessive hepatic iron content above 7-mg/g dry tissue and corresponding 2.33 mg/mL of wet weight IC, the iron quantification in a 3MD was of clinical accuracy. A 2MD of W-containing samples required severe pathology to detect an iron liver. A 2MD of I-containing samples was too inaccurate for clinical use.

The W-based CM used in this study was investigational and not approved for human use. The requirements for CM development are high from a regulatory and toxicological perspective.¹⁹ This study used the optimal threshold energy combination with respect to W among the available ones. Using another combination, the results might deteriorate. Likewise, in a different PCCT system, the results might change depending on the spectral configuration.

The validity of the results is limited by the low fat content of 20% or less in the material samples analyzed. Higher fat contents were not used because the mixtures become unstable. Higher fat contents in human imaging might mitigate the beam hardening assigned as responsible for IC overestimation but might overcompensate in the case of even higher contents. The effect of higher fat contents both in phantom and human studies requires thus further investigations. Moreover, the calibration of LTS and fat was performed with the same phantoms and exposures as those used to measure the IC and should therefore be tested with different setups for a more reliable validation.

The aim of our study was to demonstrate feasibility in a phantom. The approach is therefore only a first step toward the development of an iron quantification method on enhanced CT scans. The LTS-based samples used mimic liver tissue in terms of x-ray attenuation but cannot replicate physiological liver characteristics. Thus, the method thus needs further validation for clinical use. We suggest that ex vivo phantom data that better reflect in vivo conditions are needed to better evaluate patients with normal or mild iron deposition. A follow-up study is planned to examine liver samples in enhanced CT scans, followed by analytic IC measurements to obtain histologic reference data. As a next step, IC shall be measured in patients using the presented method and compared with an alternative technique such as MRI-based methods. Further improvement of the quantification can be achieved by machine learning approaches, which have shown promising material quantification results in spectral x-ray and CT imaging.^33,34

CONCLUSIONS

The presented study applied a previously published method for multimaterial decomposition to quantify hepatic IC despite prevalent CM. The evaluation focused on the accuracy of iron quantification in the presence of tungsten and compared the suitability of tungsten and iodine for enhancement in the clinically common dual-energy CT. Iron was quantified more accurately with prevalent tungsten than with iodine. In a prior knowledge-free 3-material decomposition with prevalent tungsten, the quantification was found to be clinically accurate and, in particular, more accurate than in a 2-material decomposition. In a 2-material decomposition, severe pathology is required to detect an iron liver.