Iodine quantification and detectability thresholds among major dual-energy CT platforms

Abstract

Objectives:

To estimate the minimum detectable iodine concentration on multiple dual-energy CT (DECT) platforms.

Methods and materials:

A phantom containing iodine concentrations ranging from 0 to 10 mg ml⁻¹ was scanned with five dual-energy platforms (two rapid kilo volt switching (r-kVs), one dual source (DS), one sequential acquisition and one split-filter). Serial dilutions of 300 mg ml⁻¹ iodinated contrast material were used to generate concentrations below 2 mg ml⁻¹. Iodine density and virtual monoenergetic images were reviewed by three radiologists to determine the minimum visually detectable iodine concentration. Contrast-to-noise ratios (CNRs) were calculated.

Results:

1 mg mL⁻¹ (~0.8 mg mL⁻¹ corrected) was the minimum visually detectable concentration among the platforms and could be seen by all readers on the third-generation r-kVs and DS platforms.

Conclusions:

At low concentrations, CNR for monoenergetic images was highest on the DS platform and lowest in the sequential acquisition and split-filter platforms.

Advances in knowledge:

The results of this study corroborate previous in vivo estimates of iodine detection limits at DECT and provide a comparison for the performance of different DECT platforms at low iodine concentrations in vitro.

Introduction

Dual-energy CT (DECT) has seen numerous technological advancements since first described by Sir Godfrey Hounsfield in 1973 and has reached clinical implementation by the major vendors within the last decade.¹ In brief, DECT exploits differences in the attenuation of photons by body tissues or by contrast material at different X-ray spectra in order to obtain additional imaging information. Each vendor’s implementation is unique, but current commercially available approaches can be generally categorized as: sequential acquisition, rapid peak kilovolt-switching (r-kVs), dual source (DS), split-filter and multilayer detector-based separation.^2–8 The sequential acquisition technique requires two sequential scans to be obtained, each at a different peak kilovoltage (e.g., an 80 kVp scan followed by a 140 kVp scan). The r-kVs technique involves changing the tube potential on the order of milliseconds, obtaining closely spaced projections with minimal temporal offset at two different energy levels as the tube is rotating around the subject.^2,3,8,9 The current implementation of this technique generates two sets of projection data which are then decomposed into water and iodine components in projection space. In the DS technique, two tube-detector arrays operating at two different energy levels are simultaneously rotating around the patient at a 90 degree offset, generating ‘high’ and ‘low’ energy images which are then combined and/or further processed in image space.^2,4 The split-filter technique was more recently commercially released and involves filtering the beam in the z-direction, with half of the beam filtered by gold and half by tin. The gold-filtered portion constitutes the ‘low’ energy data, while the tin-filtered portion constitutes the ‘high’ energy data. Processing is performed in image space as for the DS technique.^5,6 While the techniques discussed so far all rely on varying the energy at the X-ray source, the first detector-based dual energy solution recently became commercially available and utilizes a single conventional X-ray tube but employs a dual-layer detector. The top layer consists of an yttrium-based garnet scintillator which selectively absorbs low-energy photons, while the high-energy photons penetrate this layer and are instead absorbed in the bottom layer which consists of gadolinium-oxysulfide.⁷ In addition to these commercially available platforms, a prototype exists which employs novel photon counting detectors. These convert individual X-ray photons directly into an electric signal, unlike conventional energy integrating detectors, which require the additional step of converting photons to visible light. The system counts the pulses that exceed preset threshold levels and can sort the incoming photons into a number of energy bins (typically two to eight), depending on their energy.^10,11 However, the system is not commercially available at this time. Regardless of approach, the resulting information can be displayed in various ways, and vendor-specific software is typically needed for this purpose. However, a few basic display types are conserved among vendors, such as: virtual unenhanced images, virtual monoenergetic images and iodine density.^2,12,13

Material decomposition consists of modeling the subject as a linear combination of two materials (three or more in advanced models) based on the change in attenuation between the two energies.¹⁴ This has been implemented clinically in several ways; for instance, to separate urate stones from calcium oxalate kidney stones and to detect the presence of urate crystals in the diagnosis and follow-up of gout patients.^15,16 Virtual non-contrast imaging is also possible with this technique and may have multiple applications such as eliminating pre-contrast scans from multiphase pre- and post-contrast protocols as well as retrospectively generating unenhanced images when they were not prospectively obtained, but are found wanting at the time of interpretation (ex. evaluation of adrenal nodules, renal masses, renal calculi, vascular studies, etc.).^17–27 Nonetheless, these remain largely niche applications. Iodine quantification, however, is perhaps the most broadly applicable use of DECT material decomposition for several reasons. First, the widespread use of iodinated contrast material makes it applicable to a large proportion of cases. Second, iodine is uniquely suited for material decomposition due to its k-edge near the lower end of the diagnostic energy range, allowing it to be easily isolated and separated from water, the primary constituent of most tissues (Figure 1).² Prior studies have utilized this property to reduce contrast dose by as much as 50% without an accompanying loss in attenuation or contrast-to-noise ratio (CNR).^28,29 Finally, it is conceivable that quantitative evaluation of iodine enhancement may one day be validated as an imaging biomarker.^30,31

Figure 1. .

Graph of iodine and water attenuation as a function of photon energy

Imaging biomarkers are defined by Smith et al as “any anatomic, physiologic, biochemical, or molecular parameter detectable with one or more imaging methods used to help establish the presence and/or severity of disease.”³² Accurate, validated imaging biomarkers hold great potential to fundamentally alter disease diagnosis and monitoring, due to the relative safety, cost-effectiveness and speed of acquisition of imaging modalities. Recent studies have used iodine concentration as an imaging biomarker to differentiate between benign cysts and malignant lesions in the liver, kidney and pancreas, as well as define tumor extent and response to treatment.^{30,31,33–35} In order for iodine concentration to be useful as an imaging biomarker, it is imperative to know the lower limits of detection with current DECT platforms. Phantom studies have demonstrated the linear response of r-kVs and DS platforms across a wide range of concentrations spanning 2–20 mg ml⁻¹; however, relatively little has been written about concentrations below 2 mg ml⁻¹.^22,36,37 This may be partially due to the lack of a commercially available iodine phantom below this concentration. In vivo studies have indirectly addressed this to some extent; for example, a study in hepatocellular carcinoma patients defined a cutoff of 0.9 mg ml⁻¹ to differentiate bland from tumor thrombus.³⁸ Indirectly, the implication is that iodine concentrations below 0.9 mg ml⁻¹ do not represent true enhancement, suggesting that concentrations below this level are undetectable. Similarly, several studies have established cutoffs ranging from 0.5 to 1.3 mg ml⁻¹ in differentiating benign renal cysts from renal cell carcinoma.^39–42 The aim of our study was to establish the lower limits of accurate iodine quantification across multiple DECT platforms in a custom phantom.

Methods and materials

Phantom preparation

Serial dilution techniques were utilized beginning with a 300 mg ml⁻¹ solution of iodine (OMNIPAQUE^® 300; GE Healthcare, Waukesha, WI) to achieve concentrations of 3.0, 1.5, 1.0, 0.6, 0.3, and 0.15 mg I mL⁻¹. These were loaded in plastic syringes which were capped, sealed and loaded into the inner ring of the 33-cm-diameter and 5-cm-thick water equivalent phantom disc shown in Figure 2 with inner and outer wells arranged in concentric ring patterns (Gammex^® 472; Gammex Inc., Middleton, WI). Syringes were wrapped in several layers of paper or cellophane tape to enable a tight fit within the phantom well. Commercial inserts of 0, 2, 2.5, 5, 7.5, and 10 mg I mL⁻¹ were loaded in the outer ring of the phantom disc. Additionally, two extra inserts were added to fill the empty wells in the outer ring and avoid resulting streak artifacts (one water equivalent and one low concentration calcium equivalent). The relative location of the inner and outer rings did vary between scans, although the concentrations in each ring were kept in sequential order.

Figure 2. .

Sample phantom arrangement and sample ROI placement

Image acquisition

The phantom disc was then scanned on the various platforms. Two r-kVs platforms were tested, including a third-generation 256-detector model (Revolution CT; GE Healthcare, Waukesha, WI) and a second-generation 64-detector model (Revolution HD; GE Healthcare, Waukesha, WI). One third-generation DS platform was tested (SOMATOM Force; Siemens Healthineers, Forchheim, Germany) in two different kilo voltage settings (80/150Sn and 100/150Sn). Finally, sequential acquisition and split-filter technique were both tested on a fourth scanner (Definition Edge; Siemens Healthineers, Forchheim, Germany). We did not have access to a dual-layer detector-based dual-energy scanner. Scan settings for each platform are summarized in Table 1. Due to the specific constraints of each system, it was impossible to replicate the exact settings between all scanners (i.e., pitch, mA, rotation time, etc.). Therefore, the scan parameters were adjusted to maintain similar photon flux across platforms by adjusting scan settings to achieve a target CTDI_vol of approximately 25 mGy, to the extent possible. This was selected as it is the ACR diagnostic reference level for abdominal CT in a 32-cm-diameter phantom (our phantom is very similar, measuring 33 cm) and we wanted to evaluate scanner performance in an ideal setting, although we acknowledge that lower doses are typically used in routine clinical practice.⁴³ Scans were performed with the phantom supported upright in the gantry. The split-filter platform requires a minimum cranial-caudal thickness of 10 cm, therefore, for this platform only, a second phantom disc without inserts was piggy-backed to the iodine phantom described above to achieve 10 cm thickness.⁴⁴

Table 1. .

Scan Settings. 150Sn denotes 150 kVp with Sn filtration. 120Au and 120Sn denote 120 kVp with Au and Sn filtration, respectively

DECT method	Scanner model	Low kVp	High kVp	Rotation time	mAs	Pitch	CTDIvol	Reconstruction Kernel
Rapid kilovolt Switching	Revolution HD	80	140	1s	600	1.375	26 mGy	Standard
	Revolution CT	80	140	1s	475	0.992	23 mGy	Standard
Sequential Acquisition	Definition Edge	80	140	1s	210/100	0.35/0.7	26 mGy	D30s
Dual Source	SOMATOM Force	80	150Sn	0.5 s	600/300	0.65	25 mGy	QR40d
	SOMATOM Force	100	150Sn	0.5 s	375/188	0.65	25 mGy	QR40d
Split-Filter	Definition Edge	120Au	120Sn	1s	1200	0.25	25 mGy	D30s

Image analysis

Iodine quantification

Image analysis was performed using vendor-specific software, as is typically required. Rapid-kVs scans were processed using the GSI viewer application in AW Server 3.2 (GE Healthcare, Waukesha, WI). DS, sequential acquisition and split-filter images were processed using the Liver VNC application in Syngo.via VB20A (Siemens Healthcare, Erlangen, Germany). Axial slices were reconstructed at 5 mm thickness, without employing iterative reconstruction techniques. This was carried out to remove any influences or bias due to the varying iterative reconstruction implementations, although we do acknowledge that iterative reconstruction techniques are commonly employed in clinical practice. Regions of interest (ROIs) were hand drawn to maximally cover the contents of the syringe, without including the syringe wall itself or extending beyond. Sample ROI placement is shown in Figure 2. Depending on the functionality of the image analysis software, this ROI was either copied onto adjacent background or an ROI of the same size (assessed visually) was hand-drawn in the adjacent background phantom disc for comparison. Average ROI values as well as SD were recorded. Additionally, CNR was calculated according to the following formula:

$$\frac{({ROI}^{syringe}-{ROI}^{adjacentbackground})}{{SD}^{adjacentbackground}}$$

Virtual monoenergetic analysis

CNR was also calculated according to the above formula for virtual monoenergetic images with keV values ranging from 40 to 80 keV. 3-mm-axial slice thickness was used for this purpose, to mimic performance in thinner slice abdominal applications. Rapid-kVs scans were again processed using the GSI viewer application in AW Server 3.2 (GE Healthcare, Waukesha, WI). DS, sequential acquisition and split-filter images were processed using the Monoenergetic Plus (Mono+) application in Syngo.via VB20A (Siemens Healthcare, Erlangen, Germany). The Monoenergetic Plus application was selected over the standard Monoenergetic (Mono_E) algorithm as it is a noise optimized technique which employs a more advanced spatial frequency blending algorithm to enhance CNR and image quality, and is what we use in clinical practice.⁴⁵ ROIs were placed using similar technique as described above for iodine quantification.

Inter-reader agreement

Three independent board-certified radiologists (N.C.Y., D.P.K. and N.G., with 8, 10 and 4 years’ experience, respectively, and all with subspecialty fellowship training in abdominal imaging) reviewed the iodine density images for each platform. The lowest concentration which was confidently visible above the surrounding background phantom disc was recorded. Fleiss’ multireader κ analysis was performed to estimate inter-reader agreement.^46,47 Statistical analysis was performed utilizing Stata Statistical Software: Release 15 (StataCorp LLC; College Station, TX). The κ statistic was interpreted according to the scale suggested by Landis and Koch:<0.0= poor, 0.0–0.20 = slight, 0.21–0.40 = fair, 0.41–0.60 = moderate, 0.61–0.80 = substantial, and 0.81–1.0 = almost perfect.⁴⁸

Results

Measured iodine concentrations for commercial phantoms and in-house prepared dilutions on the various CT systems are shown in in Figure 3. Both third-generation scanners performed similarly on the commercial standards (Figure 3a), with the DS scanner at 100/150Sn settings closest to unity in this range, followed closely by the same scanner at 80/150Sn and the third-generation r-kVs scanner. Second-generation r-kVs, split-filter and sequential techniques tended to underestimate the concentration, with the split-filter technique reporting the lowest values. However, all systems responded in a nearly linear manner for concentrations of 2 mg ml⁻¹ or higher. A similar pattern was seen at lower concentrations, where the third-generation DS scanner at 100/150Sn remained the most linear on the in-house dilutions as well. The split-filter scanner had significant artifacts in this area (see discussion) and results should not be generalized. On the low concentration in-house dilutions, all of the scanners tended to underestimate the concentrations (Figure 3b).

Figure 3. .

Measured iodine concentrations for both commercial inserts and in-house dilutions

Given this systematic bias across all scanners, we felt this was likely due to errors propagated during the creation of the dilutions (see discussion below). Therefore, we estimated ‘corrected’ values for these as follows. A line of best fit was fitted to the third-generation DS 100/150Sn commercial phantom results since these were most linear and nearest unity in the range from 2 to 10 mg ml⁻¹. The 3 mg ml⁻¹ in house dilution was then fit to this line, yielding a corrected value of 2.4 mg ml⁻¹. Since the remaining dilutions (1.5, 1.0, 0.6, 0.3, 0.15 mg ml⁻¹) were made from the 3 mg ml⁻¹ solution, they were corrected to 1.2, 0.8, 0.5, 0.24 and 0.12 mg ml⁻¹, respectively. Iodine density images for each tested platform are shown in Figure 4. These images were reviewed by three independent radiologists, and the minimum visually detectable iodine concentrations are recorded in Table 2. Fleiss κ was calculated to be 0.625, indicating substantial agreement among the readers. 1 mg ml⁻¹ (0.8 mg ml⁻¹ corrected) was the lowest visible concentration and was visible to all three readers on the third-generation DS platform (at both 80/150Sn and 100/150Sn settings) as well as on the third-generation r-kVs platform. 1.5 mg ml⁻¹ (1.2 mg ml⁻¹ corrected) was the median lowest visible concentration on the second-generation r-kVs and sequential acquisition platforms. The 2.0 mg ml⁻¹ commercial standard was the median lowest visible concentration on the split-filter platform.

Figure 4. .

Iodine density images for each platform. Arrow indicates 1 mg/mL dilution, arrowhead indicates median lowest visible concentration.

Table 2. .

Minimum visually detectable iodine concentration for each platform by each reader. Concentrations are in mg/mL

Minimum Detectable Iodine Concentration
Dual-Energy CT Method	Reader 1	Reader 2	Reader 3
Second Generation Rapid kVs	1.0	1.5	1.5
Third Generation Rapid kVs	1.0	1.0	1.0
Sequential Acquisition	2.0	1.5	1.5
Third Generation Dual Source 80/150 kVp	1.0	1.0	1.0
Third Generation Dual Source 100/150 kVp	1.0	1.0	1.0
Split-Filter	2.0	2.0	2.0
	Fleiss κ		0.625

The mean ± SD for the in-house prepared dilutions are plotted against background mean ± SD in Figure 5. The lowest concentration at which the ROI is significantly greater than background noise (intersection of the shaded bars; indicating two SD’s above background) also occurs near 1.0 mg ml⁻¹ (0.8 mg ml⁻¹ corrected), although this time this was achieved with 80/150Sn kVp settings on the third-generation DS platform and on the third-generation r-kVs platform. While the second-generation r-kVs scanner also achieved this at 1 mg ml⁻¹ (0.8 mg ml⁻¹ corrected), we interpret this result with caution, noting the heterogeneity of the background on this platform as demonstrated by the drop in background ROI value adjacent to the 1 mg ml⁻¹ insert. Additionally, only one of the three readers’ felt that the 1.0 mg ml⁻¹ dilution was visible above background on the second-generation r-kVs platform. Most of the platforms cross-this threshold between ~1 and 1.5 mg ml⁻¹, with the exception of the split-filter platform which demonstrated imaging artifacts (see discussion) resulting in negative ROI values.

Figure 5. .

Noise floor for each of the platforms on low concentration in-house created dilutions

CNR values for each concentration and platform on virtual monoenergetic images are displayed in Figure 6. As the reconstructed energy level approaches the k-edge of iodine (~33 keV), CNR tends to generally increase; however, this pattern does not necessarily hold true when evaluating concentrations below the detection limits for each platform described above. The DS platform tended to have the highest CNRs on virtual monoenergetic images, and at 1 mg ml⁻¹, the 80/150 kVp settings performed best (although there was no clear difference between 100/150Sn and 80/150Sn overall). The r-kVs platforms demonstrate slightly lower CNR than the DS platform on monoenergetic images, although this is likely influenced by the noise optimized Mono +algorithm used for the DS platform in addition to the hardware differences. The r-kVs platforms performed similarly, as well as the sequential acquisition technique. As would be expected, below the 1 mg ml⁻¹ dilution, there is no discernible pattern among any of the platforms. In fact, negative CNR values were often obtained at the lower keV values for concentrations of 0.3 and less, related to increasing background noise heterogeneity.

Figure 6. .

CNR as a function of virtual monoenergetic keV

Discussion

Our results provide in vitro corroboration of earlier in-vivo studies involving HCC and RCC; namely that the lower limits of iodine detection with the third-generation DS and r-kVs platforms seem to be approximately 0.8 mg ml⁻¹ after correcting for errors in our dilution process.^22,38,40,41 We demonstrated this in multiple ways. Subjectively, the 1 mg ml⁻¹ phantom (0.8 mg ml⁻¹ corrected) was the lowest visually detectable on these platforms by all readers. Objectively, at ≤1 mg ml⁻¹ (0.8 mg ml⁻¹ corrected), the error bars between the ROI values and adjacent background begin to overlap on the third-generation platforms, indicating that the ‘noise floor’ is being approached (Figure 5). As a corollary to this, on most of the platforms (with the exception of the split-filter), a CNR of 2 (indicating a mean ROI value two SD’s above background) was reached at iodine concentrations of 1–1.5 mg ml⁻¹ (0.8–1.2 mg ml⁻¹ corrected) on iodine density images (Figure 5). Notably, there were important differences between the studied scanners. For example, with the second-generation r-kVs and sequential acquisition platform, the median lowest visually detectable concentration was slightly higher, 1.5 mg ml⁻¹ (1.2 mg ml⁻¹ corrected), and 2.0 mg ml⁻¹ for the split-filter platform.

On virtual monoenergetic images, the DS platform tended to have the highest CNR, particularly at the 80/150Sn setting. There may be several reasons for the improved CNR of the DS system compared with the other platforms. Perhaps most importantly, it has the best spectral separation, allowing values as low as 70 kVp paired with as high as 150 kVp, which is further bolstered with Sn filtering of the 150 kVp beam. In our study, we studied 80/150Sn and 100/150Sn pairings, while the r-kVs platforms and the sequential acquisition platform were both evaluated at 80/140 kVp. Poorest spectral separation would be expected with the split-filter setup, where a single 120 kVp beam is filtered into high and low components. Additionally, it is possible that the use of a second tube allows more low energy photon flux than can be achieved with rapid voltage switching or split beam, although this remains speculative. Along these lines, it may also be hypothesized that using 100 kVp as the lower energy level on the DS platform might improve photon flux through a larger structure at the expense of spectral separation, although notably we did not see a significant difference in monoenergetic CNR performance between 80/150Sn and 100/150Sn pairings in this study. This would suggest that neither of these effects is significant or that these effects mitigate each other in our 33 cm phantom. It remains possible that one or the other setting may be preferable in larger or smaller patients. Finally, differences in reconstruction and post-processing may play a role, particularly since the DS, sequential and split-filter platforms used an image-based dual energy reconstruction, while the r-kVs platforms use a projection-based approach. Additionally, the image-based platforms benefited from a noise optimized algorithm utilizing spatial frequency blending as described earlier (Mono+). Perhaps development of similar algorithms for the r-kVs platforms would enhance their virtual monoenergetic performance as well.

Our results should be interpreted with caution. These results were obtained in a uniform water phantom of 33 cm diameter. It is possible that slightly different limits of detection may be observed due to differences in beam hardening and scatter related to subject thickness/diameter, as well as subject composition. However, a recent publication by Jacobsen et al reported the “limit of quantification” (LOQ; defined as the analyte concentration at which repeated measurements demonstrated a coefficient of variation of ≤20%) to be 0.5 mg ml⁻¹ for the third-generation dual-source system and 1.0 mg ml⁻¹ for the second-generation r-kVs system in a large phantom of 30 × 40 cm. While these results are close to our corrected reported limits of visual detection of 0.8 mg ml⁻¹ and 1.2 mg ml⁻¹ on the same systems, respectively; they reported a LOQ of 0.21 mg ml⁻¹ for the third-generation r-kVs system which is far below the 0.8 mg ml⁻¹ that we felt was the visually detectable limit.³⁷ These differences are likely related to differing methodologies; while their study focused on accuracy and precision of repeated measurements, our study assessed visibility on a single scan. Another caveat of our study is that our thresholds were obtained with iodine/water material separation. It is possible that different thresholds would be observed to differentiate iodine enhancement relative to proteinaceous or hemorrhagic fluid, as would be the case in evaluating a hyper-dense renal cyst vs an enhancing mass, since calcium or ferrous salts may be falsely represented as iodine in a two material water/iodine decomposition.⁴⁹ This might therefore require a higher threshold to discriminate from true enhancement.

Our study had several limitations. For one, a systematic bias (as described in the Results section) was observed on the in-house dilutions whereby all scanners tended to underestimate the iodine concentration. Our dilutions were created in house, and it is possible there were errors which were propagated while making the dilutions. Additionally, there may be inaccuracies in the actual iodine concentration of our base solution, which was assumed to be 300 mg ml⁻¹, although we did not assay the base solution for confirmation. Another possibility is that there may have been evaporative losses over time, as the scans were not all performed on the same day, although syringes were capped and sealed once the dilutions were created. However, with the exception of evaporative losses, most of these issues should be addressed by the internal correction against the commercial iodine standards which we performed. Future studies employing these techniques should ideally assay the base solution and/or the dilutions using mass spectroscopy (as was done by Jacobsen et al) in order to confirm accuracy.³⁷ Second, each scanner was evaluated only once. It is possible that small differences could arise based on changes in daily scanner calibration, etc. that were not evaluated. This may be the subject of future research. Third, the dual-layer detector platform was not available to us for study and employs a novel detector-based energy separation scheme. This may also be evaluated in future studies. Fourth, the 10 cm combined thickness of the two phantom discs in the split-filter platform only just met the minimum requirements of cranial-caudal thickness on this platform. Therefore, its performance may have been adversely affected and likely does not reflect real world use. Further studies using a larger phantom are planned. Finally, the 5 cm phantom thickness does not adequately model the internal scatter that would be expected in an actual patient, and therefore in vivo accuracy may differ.

While photon flux was kept roughly the same between platforms by keeping target CTDI at approximately 25 mGy, it is possible that detection limits may vary with radiation dose, which is usually lower than this level clinically. This level was selected in order to observe the behavior of the various systems in an optimal environment, and because it serves as the ACR reference level for abdominal CT in a 32 cm phantom. Also, since we were most interested in the performance of the hardware across the various platforms, we did not utilize iterative reconstruction techniques, which are also widely used clinically but which vary among manufacturers. Future studies may aim to evaluate the effects of various iterative reconstruction techniques and potentially deep learning-based reconstruction techniques on the observable iodine concentration.

As CT technology continues to evolve, we may see these limits of detection improve. As a comparator, estimates of the minimum detectable gadolinium concentration at MRI vary in the literature from 1 to 100 µM (0.16–16 µg Gd/mL), and are dependent upon technical factors and specific contrast agent relaxivity.^50–52 This is far superior to the 1 mg ml⁻¹ (0.8 mg ml⁻¹ corrected) demonstrated for iodine detectability with DECT in this study. Even after accounting for the fact that larger amounts of iodinated contrast can be administered relative to their gadolinium-based counterparts, DECT would need to detect ~0.006–0.6 mg I/mL to be able to compete with MRI in contrast sensitivity. While the upper bounds of this range are certainly within reach, the lower limits of this range may or may not be achievable in the foreseeable future. Nonetheless, this exercise provides a benchmark for comparison and motivation for further technical development.

Conclusion

The lowest visible iodine standard was 1 mg ml⁻¹ (0.8 mg ml⁻¹ corrected), achieved on both third-generation systems (r-kVs and DS) and for all three readers. A CNR of 2 was achieved on iodine images at 1–1.5 mg ml⁻¹ (0.8–1.2 mg ml⁻¹ corrected) on all tested platforms except for the split-filter platform, although results for the split-filter platform may have been unreliable due to the thinness of the phantom, and should not be generalized. The third-generation DS platform demonstrated the best CNR on monoenergetic images, which may be the result of the hardware differences as well as the use of a noise-optimized monoenergetic algorithm.