Radiation Dose Reduction for Coronary Artery Calcium Scoring Using a Virtual Noniodine Algorithm on Photon-Counting Detector Computed-Tomography Phantom Data

Background: On the basis of the hypothesis that virtual noniodine (VNI)-based coronary artery calcium scoring (CACS) is feasible at reduced radiation doses, this study assesses the impact of radiation dose reduction on the accuracy of this VNI algorithm on a photon-counting detector (PCD)-CT. Methods: In a systematic in vitro setting, a phantom for CACS simulating three chest sizes was scanned on a clinical PCD-CT. The standard radiation dose was chosen at volumetric CT dose indices (CTDIVol) of 1.5, 3.3, 7.0 mGy for small, medium-sized, and large phantoms, and was gradually reduced by adjusting the tube current resulting in 100, 75, 50, and 25%, respectively. VNI images were reconstructed at 55 keV, quantum iterative reconstruction (QIR)1, and at 60 keV/QIR4, and evaluated regarding image quality (image noise (IN), contrast-to-noise ratio (CNR)), and CACS. All VNI results were compared to true noncontrast (TNC)-based CACS at 70 keV and standard radiation dose (reference). Results: INTNC was significantly higher than INVNI, and INVNI at 55 keV/QIR1 higher than at 60 keV/QIR4 (100% dose: 16.7 ± 1.9 vs. 12.8 ± 1.7 vs. 7.7 ± 0.9; p < 0.001 for every radiation dose). CNRTNC was higher than CNRVNI, but it was better to use 60 keV/QIR4 (p < 0.001). CACSVNI showed strong correlation and agreement at every radiation dose (p < 0.001, r > 0.9, intraclass correlation coefficient > 0.9). The coefficients of the variation in root-mean squared error were less than 10% and thus clinically nonrelevant for the CACSVNI of every radiation dose. Conclusion: This phantom study suggests that CACSVNI is feasible on PCD-CT, even at reduced radiation dose while maintaining image quality and CACS accuracy.


Introduction
Coronary artery calcium scoring (CACS) is a strong predictor of future adverse cardiovascular events and outcomes [1][2][3][4][5], and is thereby used for cardiovascular risk stratification [6,7]. The cardiac computed tomography (CT)-based Agatston method (tube voltage, 120 kVp; reconstruction with slice thickness, 3.0 mm) with a threshold of 130 Hounsfield Diagnostics 2023, 13, 1540 2 of 13 units (HU) is considered the standard method for CACS [8]. The 2019 American College of Cardiology/American Heart Association (ACC/AHA) Guideline on the Primary Prevention of Cardiovascular Disease supports the assessment of CACS for further decision making in patients at intermediate risk and selected patients at borderline risk [9]. However, when considering this method as a screening tool to prevent adverse cardiovascular events, the radiation dose remains a concern.
Photon-counting detector (PCD)-CT, a recent and promising technology, enables energy discrimination at high spatial and temporal resolutions by separately counting every incoming photon, weighing it with regard to its energy [10][11][12][13], and allowing for low-dose imaging with better image quality than that of energy-integrating detectors (EIDs) [14][15][16]. Moreover, compared to dual-energy CT systems, improved spectral imaging enables multimaterial differentiation [12], e.g., among soft tissue, calcium, and iodine. By creating virtual noniodine (VNI) images on the basis of coronary CT angiography (CCTA) data, the true noncontrast (TNC) examination normally performed for CACS could potentially be eliminated, possibly reducing both the examination time and total radiation burden to the patient.
In this context, an iodine-removal algorithm specifically developed for PCD-CT outperformed virtual noncontrast (VNC) reconstructions for CACS at a standard radiation dose [17]. Due to the advantages of PCD-CT and accordingly this new VNI algorithm, we hypothesized that VNI reconstructions enable accurate CACS even in low-dose examinations.
Therefore, this study assesses the impact of radiation dose reduction on image quality and CACS accuracy using a VNI algorithm on PCD-CT phantom data. In a systematic analysis of a well-standardized chest phantom with a CACS insert and two extension rings simulating three different chest diameters, VNI-based CACS was examined at four different radiation dose levels regarding the calcium score and image quality.

Phantom Setup
For this study, a commercially available anthropomorphic chest phantom (QRM GmbH, Moehrendorf, Germany) was used with an insert designed for calcium scoring at the heart position. This insert contained nine cylinders with different diameters (1, 3, and 5 mm) and different densities (200, 400, and 800 mg/cm 3 calcium hydroxyapatite), and two calibration rods (one water-equivalent and one with 200 mg/cm 3 calcium hydroxyapatite) as recently described [17]. To simulate the different chest diameters, the phantom was scanned with and without extension rings, resulting in a total of three chest diameters: 300 × 200 mm (small; no extension ring), 350 × 250 mm (medium; extension ring M), and 400 × 300 mm (large; extension ring L).

Data Acquisition
All phantoms were scanned using a first-generation dual-source PCD-CT (NAEOTOM Alpha; Siemens Healthineers, Forchheim, Germany) containing two PCDs (cadmium telluride; collimation of 144 × 0.4 mm). The tube voltage was 120 kVp in QuantumPlus scan mode, gantry rotation time was 0.25 s, and scans were performed with a simulated electrocardiogram (ECG) signal at 60 beats per minute (bpm) and an ECG triggering with a single phase of 75%. As previously described [18], the standard radiation dose (100%) was chosen according to clinically used protocols with volumetric CT dose index (CTDI Vol ) values of 1.5, 3.3, and 7.0 mGy for the small, medium-sized, and large phantoms, respectively [19][20][21], with calculated mean CTDI Vol of 1.4 ± 0.1, 3.1 ± 0.04, and 6.1 ± 0.2 mGy, respectively. Mean CTDI Vol values were calculated with the CT scanner and extracted from patient exposure protocol data. Radiation dose was gradually reduced by adjusting tube current modulation, which resulted in four different dose levels: 100% (standard radiation dose), 75%, 50%, and 25%. Detailed radiation dose parameters are illustrated in Table 1. Since interscan variability may be high in CACS [22], each scan was repeated five times with the slight repositioning of the phantom between the scans (approximately 2 mm translational and 2 degrees rotational) [23].

Image Reconstruction
Reconstructions were performed using a proprietary offline raw data image reconstruction platform (ReconCT version 15.0.57554.0; Siemens Healthineers) with the following parameters: reconstruction kernel Qr36f, slice thickness 3.0 mm, slice increment 1.5 mm, as recommended by the vendor, and field of view 200 mm. In addition to the standard options, such as slice thickness and convolutional kernel, PCD-CT offers further possibilities in image reconstruction. Measuring the energy of every incoming photon separately enables to create virtual monoenergetic images (VMI) instead of conventional polychromatic CT reconstructions [10]. In addition, iterative reconstructions (IR) can be used to substantially lower image noise [24,25], which is particularly advantageous in low-dose imaging [26][27][28][29]. Since conventional IR algorithms cannot simply be transferred to PCD-CT systems [30], the vendor of this first-generation PCD-CT introduced a quantum IR (QIR) algorithm with four strength levels. The images of all four radiation dose levels were postprocessed using a VNI algorithm (PureCalcium, Siemens Healthineers) as previously described [17], and reconstructed at two different VMI and QIR levels: 55 keV/QIR1 and 60 keV/QIR4. These reconstructions showed that VNI-based scores were the closest to the corresponding scores from reference TNC scans [31]. TNC reconstructions without VNI postprocessing at 70 keV and QIR off at the standard radiation dose served as the reference as recommended by the vendor (given by the factory protocol).

Image Analysis
Scans were analyzed by a radiologist with four years of experience in cardiovascular imaging regarding image quality and CACS. Image quality was determined on the basis of image noise (IN) and the contrast-to-noise ratio (CNR). Regions of interest (ROIs) of 1.5 cm 2 were used to evaluate the consistency of the CT numbers (HU, mean, and standard deviation (SD)) in the center of the calibration rod and the background of the calcium scoring insert of the phantom. Each measurement was performed three times at different locations, and the mean values were calculated for further analysis. IN was defined as the SD of the background measurements, and the CNR was calculated according to the following formula [32]: According to the guideline for minimizing radiation exposure during the acquisition of coronary artery calcium scans with the use of multidetector computed tomography [33], the recommended IN targets of 20 HU for small/medium-sized phantoms and 23 HU for large phantoms were used to further compare the IN between different reconstructions. CACS quantification was performed on VNI (CACS VNI ) and TNC (CACS TNC ) datasets using commercially available software (syngo.via Version VB60; Siemens Healthineers) that highlights calcified lesions by discriminating them from the background using a threshold of 130 HU [8], as recently described [17].

Statistical Analysis
Statistical analysis was performed using GraphPad Prism (Version 8.4.2; GraphPad, San Diego, CA, USA) and MedCalc (Version 19.4; MedCalc, Ostend, Belgium). Continuous variables were tested for normality using the Kolmogorov-Smirnov test. Normally distributed values are reported as mean ± standard deviation, and non-normally distributed values as the median with an interquartile range. Categorical values are reported as absolute frequencies and proportions. Depending on the distribution and number of variables, we used the paired-samples t-test or Wilcoxon matched-pairs signed rank test and oneway ANOVA with the Greenhouse correction or Friedman test to evaluate differences among the different reconstructions (VNI postprocessed, 55 keV/QIR1; VNI postprocessed, 60 keV/QIR4; and without VNI postprocessing, 70 keV/QIR off), and between among radiation dose levels (100%, 75%, 50% and 25%). The Bonferroni correction was applied for multiple t tests, and a p value of 0.05/3 = 0.017 was considered significant for different reconstructions and 0.05/6 = 0.008 for different radiation dose levels. The correlation between the CACS VNI of every radiation dose level and the CACS TNC of standard radiation dose was assessed using the Pearson or Spearman correlation coefficient (r) depending on the distribution. The corresponding agreement was tested using Bland-Altman analysis (mean bias, upper and lower limits of agreement (LoA)) and single-rater intraclass correlation (ICC) with 2-way mixed effects and absolute agreement (<0.5: poor; 0.5 to 0.75: moderate; 0.75 to 0.9: good; >0.9: excellent [34]). The percentage deviation of IN, CNR, and CACS between VNI and TNC reconstructions was calculated as follows: The coefficients of variation (CV) of the root-mean squared error (RMSE) were calculated to evaluate the variation between CACS VNI at different radiation dose levels and CACS TNC at the standard radiation dose. Values < 10% were considered acceptable agreement and thus clinically nonrelevant deviation [35].

Image Quality
For every radiation dose level, mean IN VNI was significantly lower than IN TNC (p < 0.001 for all), and higher in VNI reconstructions at 55 keV/QIR1 compared to 60 keV/ QIR4 (p < 0.001 for all) ( Table 2). The mean INVNI and INTNC increased significantly with decreasing radiation dose level from 100 to 25% (p < 0.001 for all) ( Figure 1A). Compared to TNC reconstructions, IN was reduced by 23.5% and 54.0% at the standard radiation dose, and by 17.9% and 50.8% at the 25% radiation dose level using VNI reconstructions at 55 keV/QIR1 and 60 keV/QIR4, respectively. The mean CNRVNI was significantly lower than CNRTNC (p < 0.001 for all), and lower in VNI reconstructions at 55 keV/QIR1 compared to 60 keV/QIR4 (p < 0.001 for all) ( Table  2). Mean CNRVNI and CNRTNC decreased significantly with decreasing radiation dose level from 100 to 25% (p < 0.001 for all) ( Figure 1B). In comparison to TNC reconstructions, CNR was reduced by 43.4% and 19.8% at standard radiation dose, and by 40.8% and 19.8% at radiation dose level 25% VNI reconstructions at 55 keV/QIR1 and 60 keV/QIR4, respectively.

Calcium Scoring
CACSVNI at 55 keV/QIR1 and 60 keV/QIR4 of every radiation dose level from 100 to 25% showed strong correlation and excellent agreement with CACSTNC at standard radiation dose (p < 0.001, r ≥ 0.88, ICC > 0.9 for all) ( Table 3). The mean CNR VNI was significantly lower than CNR TNC (p < 0.001 for all), and lower in VNI reconstructions at 55 keV/QIR1 compared to 60 keV/QIR4 (p < 0.001 for all) ( Table 2). Mean CNR VNI and CNR TNC decreased significantly with decreasing radiation dose level from 100 to 25% (p < 0.001 for all) ( Figure 1B). In comparison to TNC reconstructions, CNR was reduced by 43.4% and 19.8% at standard radiation dose, and by 40.8% and 19.8% at radiation dose level 25% VNI reconstructions at 55 keV/QIR1 and 60 keV/QIR4, respectively.

Calcium Scoring
CACS VNI at 55 keV/QIR1 and 60 keV/QIR4 of every radiation dose level from 100 to 25% showed strong correlation and excellent agreement with CACS TNC at standard radiation dose (p < 0.001, r ≥ 0.88, ICC > 0.9 for all) ( Table 3).
For the VNI reconstructions at 55 keV/QIR1 and at 60 keV/QIR4, there were no significant differences in calcium scores among the different radiation dose levels from 100 to 25% (Figure 2). Averaged over all phantom sizes, total CACSVNI scores deviated from CACSTNC standard radiation dose, by up to 4.5% and 3.7% for 55 keV/QIR1 and 60 keV/QIR4, spectively, each with the highest deviation at the 50% radiation dose level. Depending the radiation dose, CV(RSME) values were 6.9% at 100%, 5.2% at 75%, 6.6% at 50% a 7.8% at the 25% radiation dose level when comparing the total CACSVNI scores at keV/QIR1 with CACSTNC at standard radiation dose. The respective values of CACSVN 60 keV/QIR4 were slightly lower: 5.4%, 4.0%, 6.2% and 7.1% (Table 3).

Discussion
This study investigated the impact of radiation dose reduction on image quality a CACS using a VNI algorithm on PCD data. The main findings of our study are as follow Averaged over all phantom sizes, total CACS VNI scores deviated from CACS TNC at standard radiation dose, by up to 4.5% and 3.7% for 55 keV/QIR1 and 60 keV/QIR4, respectively, each with the highest deviation at the 50% radiation dose level. Depending on the radiation dose, CV(RSME) values were 6.9% at 100%, 5.2% at 75%, 6.6% at 50% and 7.8% at the 25% radiation dose level when comparing the total CACS VNI scores at 55 keV/QIR1 with CACS TNC at standard radiation dose. The respective values of CACS VNI at 60 keV/QIR4 were slightly lower: 5.4%, 4.0%, 6.2% and 7.1% (Table 3).

Discussion
This study investigated the impact of radiation dose reduction on image quality and CACS using a VNI algorithm on PCD data. The main findings of our study are as follows: First, VNI reconstructions at 55 keV/QIR1 and 60 keV/QIR 4 showed comparable or even better image quality in terms of noise than that of TNC reconstructions at 70 keV/QIR off. Second, CACS VNI at every dose level from 100 to 25% showed excellent correlation and agreement with CACS TNC at standard radiation dose with a clinically nonrelevant deviation of scores. Third, CACS VNI at 60 keV/QIR4 seems to slightly outperform 55 keV/QIR1 in terms of image quality and score deviation.
Several studies demonstrated the feasibility of radiation dose reduced CACS without compromising image quality and scoring in TNC scans obtained by previous dualsource [20,33,[36][37][38][39], as well as PCD CT [40][41][42][43]. However, due to its added diagnostic and prognostic value [44,45], CCTA and CACS are often combined in clinical practice; the latter are usually based on a separate TNC scan because of the difficulty in discriminating between iodine and calcium. The in 2022 updated version of the Coronary Artery Disease-Reporting and Data System (CAD-RADS 2.0) also supports the combination of CACS with CCTA: Due to the growing evidence that the overall plaque burden is prognostically relevant, a category "P" was added to the previous CCTA-based stenosis assessment to classify the plaque burden, ideally using the calcium score [46]. In this context, eliminating the need for a TNC scan by generating VNI images had been discussed as a method to reduce radiation dose exposure [47]. However, reducing radiation dose leads to increased image noise and artifacts, which may affect CACS by increasing noise in voxels that exceed the 130 HU threshold, thus are erroneously counted as calcified lesions. This interaction of radiation dose and image quality has to be considered, and, in accordance with the ALARA principle, radiation dose should not be as low as possible but as low as reasonably achievable.
In this study, the investigated VNI reconstructions showed similar or, regarding IN, even better image quality compared to that of TNC reconstructions. Regardless of radiation dose level, IN VNI at 55 kev/QIR1 was reduced by around 20% in comparison to IN TNC , at 60 keV/QIR4 even by over 50%. Moreover, even at radiation dose of 25%, IN VNI with 60 keV/QIR was lower than IN TNC at standard radiation dose of 100%. Overall, CNR VNI was significantly lower than CNR TNC but less reduced at 60 keV/QIR4 than at 55 keV/QIR1. Overall, these results demonstrate improved image noise of VNI reconstructions, but at the same time lower CNR in comparison to TNC reconstructions with a slightly better performance of VNI at 60 keV/QIR4 compared to 55 keV/QIR1. Although lower CNR did not seem to significantly influence CACS values, it limits radiation dose reduction in VNI-based CACS. Furthermore, it underlines the importance of choosing appropriate VMI and QIR settings when reducing radiation dose. Better image noise of VNI reconstructions, even at low dose acquisitions, is likely due to the application of the QIR algorithm, which also explains slightly better performance of 60 keV reconstructions with QIR 4 compared to 55 keV with QIR 1. Iterative reconstructions reduce image noise [24] and allow for low dose calcium scoring [26,28]. At the same time, reducing noise and blooming artifacts using iterative reconstructions may affect CACS by resulting in fewer voxels exceeding the 130 HU threshold and thus decreasing calcium scores [48][49][50][51].
Despite the application of iterative reconstructions, this study demonstrated strong correlation and excellent agreement of CACS VNI with CACS TNC at every radiation dose level from 100 to 25%. This is in line with a previous study that evaluated the performance of the same VNI algorithm at standard radiation dose and demonstrated high accuracy for CACS [17]. CACS VNI differed significantly from CACS TNC at standard radiation dose only at 55 keV/QIR1 and dose level 50% but showed no significant differences at any radiation dose level for VNI reconstructions at 60 keV/QIR4. At the same time, at radiation doses 50 and 25%, CACS VNI was significantly higher at 60 keV/QIR4 than the respective values at 55 keV/QIR1. Eberhard et al. recently evaluated the effects of different VMI and QIR levels on CACS using TNC PCD-CT scans and demonstrated increasing scores with increasing keV as well as decreasing QIR level, without significant differences among various radiation dose levels [52]. According to these results, scores should be higher at 55 keV/QIR1 than that at 60 keV/QIR4. However, while Eberhard et al. evaluated CACS using TNC reconstructions, these effects may differ due to the use of the VNI algorithm in this study. Furthermore, VNI reconstructions at 50 keV/QIR1 seem to be more susceptible to the influence of low radiation dose levels. The exact reason for this effect remains unclear but might be due to the lower QIR level.
However, overall and for both postprocessing settings, CACS VNI values maintained accurate and stable, even when decreasing the radiation dose to 25%, without any significant differences between different radiation dose levels. Regardless of radiation dose level, CACS VNI values differed less than 5% from CACS TNC at standard radiation dose. At the same time, CV(RSME) values of CACS VNI were less than 10% at each investigated dose level, demonstrating a clinically nonrelevant variation compared to CACS TNC at standard radiation dose. Regarding CACS deviation and variation, VNI reconstructions at 60 keV/QIR4 performed slightly better than those at 55 keV/QIR1. Radiation reduction is important since CACS is a strong predictor of cardiovascular morbidity and mortality in symptomatic [53] and, more importantly, asymptomatic individuals [4,54]. In addition to the calcium score, plaque density also determines the individual cardiovascular risk, with lower risk in case of high-density plaques [55,56]. The 2010 ACC/AHA Guideline for Assessment of Cardiovascular Risk in Asymptomatic Adults suggests that CACS is reasonable to consider in asymptomatic patients at intermediate risk and possibly even at low to intermediate risk [57]. However, radiation exposure was one reason why CACS was downgraded from a Class IIA to a Class IIB recommendation for patients for whom risk-based treatment decision remains uncertain after formal risk stratification in the 2013 ACC/AHA Guideline on the Assessment of Cardiovascular Risk [58]. Hence, it is still handled as such in the 2019 ACC/AHA Guideline on the Primary Prevention of Cardiovascular Disease [9]. The benefit of CACS screening has to be weighed against the risk of exposing asymptomatic patients to ionizing radiation [57,59,60] and strategies have to be developed to keep radiation dose as low as reasonably achievable while maintaining image quality and CACS values. To our knowledge, this is the first study to evaluate the feasibility of radiation dose reduced CACS using a VNI reconstruction algorithm on PCD data and indicates its feasible implementation in clinical practice to ensure appropriate individual cardiac risk stratification using CACS. This enables adding the prognostic value Diagnostics 2023, 13, 1540 9 of 13 of CACS to every CCTA examination, thus providing the full spectrum of CT-based cardiac assessment while addressing concerns about radiation exposure.
Our study has the following limitations: First, our results were based on a static phantom that only contains calcium but no iodine inserts, which means that we mainly investigated the performance of this VNI reconstruction algorithm in not erroneously removing calcium. However, this systematic phantom study is crucial as a basis for further in vivo studies to analyze whether a reduction in radiation dose is possible, and to prevent patients from being exposed to radiation for scans that may not be diagnostic. Second, we only investigated four predefined radiation dose levels without investigating the lowest possible level. Third, the radiation dose levels examined in this study were lower than those usually expected in a CCTA scan. Fourth, this study did not investigate the impact of different plaque characteristics, such as diameter and density, on VNI-based CACS. These characteristics determine individual cardiovascular risk [55,56], so this evaluation should be part of future studies. Nevertheless, the results of this study provide the first standardized overview of the accuracy of low-dose CACS using VNI reconstructions. Further in vivo or ex vivo studies should be conducted to validate these results in order to transfer them to clinical practice. These studies should then preferably also investigate the impact of cardiac motion on the VNI algorithm's performance.

Conclusions
In conclusion, the present study indicated the feasibility of low-dose CACS without decisively compromising image quality using a VNI reconstruction algorithm. However, lower CNR values compared to TNC-based CACS have to be taken into account in VNIbased CACS. Although the CACS results of the present study suggest that this does not significantly impact calcium scores, due to the slightly better image quality performance and less score deviation, VNI reconstructions at 60 keV, QIR 4 should be preferred over those at 55 keV, QIR 1 when reducing the radiation dose. This may enable a radiation dose reduction in VNI-based CACS of potentially up to 75%.