Photon-counting detector coronary CT angiography: impact of virtual monoenergetic imaging and iterative reconstruction on image quality

Abstract

Objectives:

To assess the impact of low kilo-electronvolt (keV) virtual monoenergetic image (VMI) energies and iterative reconstruction on image quality of clinical photon-counting detector coronary CT angiography (CCTA).

Methods:

CCTA with PCD-CT (prospective ECG-triggering, 120 kVp, automatic tube current modulation) was performed in a high-end cardiovascular phantom with dynamic flow, pulsatile heart motion, and including different calcified plaques with various stenosis grades and in 10 consecutive patients. VMI at 40,50,60 and 70 keV were reconstructed without (QIR-off) and with all quantum iterative reconstruction (QIR) levels (QIR-1 to 4). In the phantom, noise power spectrum, vessel attenuation, contrast-to-noise-ratio (CNR), and vessel sharpness were measured. Two readers graded stenoses in the phantom and graded overall image quality, subjective noise, vessel sharpness, vascular contrast, and coronary artery plaque delineation on 5-point Likert scales in patients.

Results:

In the phantom, noise texture was only slightly affected by keV and QIR while noise increased by 69% from 70 keV QIR-4 to 40 keV QIR-off. Reconstructions at 40 keV QIR-4 exhibited the highest CNR (46.1 ± 1.8), vessel sharpness (425 ± 42 ∆HU/mm), and vessel attenuation (1098 ± 14 HU). Stenosis measurements were not affected by keV or QIR level (p > 0.12) with an average error of 3%/6% for reader 1/reader 2, respectively. In patients, across all subjective categories and both readers, 40 keV QIR-3 and QIR-4 images received the best scores (p < 0.001).

Conclusion:

Forty keV VMI with QIR-4 significantly improved image quality of CCTA with PCD-CT.

Advances in knowledge:

PCD-CT at 40 keV and QIR-4 improves image quality of CCTA.

Computed Tomography Angiography; Coronary Vessels; Photons; Phantoms, Imaging

Introduction

In patients with a low-to-intermediate cardiovascular risk, coronary computed tomography angiography (CCTA) has emerged as the preferred non-invasive imaging modality for the initial diagnostic workup of coronary artery disease (CAD).¹ CCTA image data are usually acquired on an energy-integrating detector CT (EID-CT) system in the single-energy (SE) mode. More recently, dual-energy (DE) cardiac CT enabling the reconstruction of virtual monoenergetic images (VMI) has been introduced.^2–4 Reconstruction of VMI using high kilo-electronvolt (keV) levels may help reduce blooming artifacts from calcified coronary plaques.⁵ On the other hand, reconstructions at low keV levels are of interest in CT angiography because of the energy dependent attenuation of iodine. Specifically, attenuation and contrast-to-noise-ratio in the arteries are increased due to the proximity of low monoenergetic energy levels to the k-edge of iodine.^2,6–8 Thus, DE-based cardiac CT with VMI reconstructions can be used to improve image quality and/or to reduce the amount of required contrast media.^3,9

Recently, a whole-body full field-of-view dual-source photon-counting detector CT (PCD-CT) system has been approved for clinical use.¹⁰ In contrast to previous EID-CT systems, PCD-CT enables a direct conversion of photons to electrical signal. PCD-CT has shown potential to improve spatial resolution, contrast-to-noise ratio (CNR), and to lower image noise as compared to EID-CT systems.^10–15 Furthermore, PCD-CT has intrinsic spectral capabilities as PCD-generated signals include the energy information of every individual photon. With this scanner, VMI reconstructions represent the new reference standard for image evaluation, thereby substituting conventional polychromatic EID-CT images.^11,16 In regard to cardiac CT imaging, the inherent spectral capabilities of the PCD-CT system enable DE imaging at the unrestricted highest temporal resolution of the scanner.¹⁷

Along with the introduction of the new PCD-CT system, a novel iterative reconstruction (IR) algorithm called quantum iterative reconstruction (QIR) has been introduced. With this novel PCD-CT system, images can be reconstructed either without QIR (QIR-off) or with QIR at four different strength levels (QIR 1–4).^10,18

To the best of our knowledge, no study so far has systematically assessed the impact of keV levels and QIR on CCTA on this novel clinical PCD-CT system. Previous studies on EID-CT systems suggested a trade-off between contrast enhancement and noise levels on low keV VMI reconstructions generated with various DECT technologies,^19–21 and between noise reduction and changes in noise texture at higher strength levels of various IR algorithms.^22–25

In a dedicated phantom and in patients, we sought to systematically analyze the impact of low keV virtual monoenergetic images and quantum iterative reconstruction levels on objective and subjective image quality on CCTA on a clinical PCD-CT.

Methods and materials

CT imaging protocol

For all phantom and patient scans the same CT protocol was used. All scans were acquired on a first-generation, whole-body, dual-source PCD-CT system (NAEOTOM α; Siemens Healthineers, Forchheim, Germany, Version VA40) equipped with two photon-counting detectors (cadmium telluride), each with a 144 × 0.4 mm collimation. ECG-triggered sequential acquisitions were performed with ECG-pulsing depending on the heart rate. Importantly image acquisition was performed in the QuantumPlus (i.e., spectral) mode (and not in the ultra-high resolution (UHR) mode as shown elsewhere^26,27). Tube voltage was set at 120kVp and the image quality (IQ) level was 44, representing the default CCTA protocol for clinical routine. The IQ level represents the effective mAs applied for the protocol-specific reference water-equivalent diameter with a CT geometry correction hereby representing a system- and reconstruction-independent image quality definition. The CTDI_vol was 12.8 mGy (water-filled container) and 9.4 mGy in the phantom experiments, and 14.5 ± 10.2 mGy in patients, respectively. Gantry rotation time was 0.25 s. In patients, CCTA was performed after the intravenous injection of 40–70 ml contrast media (Iopromide, Ultravist, Bayer Healthcare, Berlin, Germany) at an iodine delivery rate of 1.85–2.22 gI/s depending on the body mass index (Table 1 for details) of the individual patient, followed by a second isovolumetric phase consisting of 20% contrast agent and 80% saline solution (NaCl 0.9%), followed by a saline flush.

Table 1.

Patient demographics

Age	59 ± 13 years
Sex	five female, five male
Body mass index	27.8 ± 4.6 kg/m2
Body height	168 ± 10 cm
Body weight	79 ± 14 kg
Indications for coronary CT angiography	rule-out coronary artery disease at a low-to-intermediate pretest probability (8 patients, 80%). rule-out coronary artery disease prior to aortic valve surgery (2 patients, 20%)
Cardiovascular risk factors	Positive family history	7 (70%)
	Smoking	3 (30%)
	Diabetes	2 (20%)
	Arterial hypertension	6 (60%)
	Dyslipidemia	8 (80%)
	Obesity	1 (10%)

CT image reconstruction

For both phantom and patient data, VMI at 40-70keV were reconstructed in 10keV increments with QIR-off and with all strength levels of QIR (QIR 1–4). A medium soft convolution kernel (Siemens Bv40), a section thickness of 0.6 mm, an increment of 0.3 mm, a field of view (FOV) of 200 mm, and a matrix size of 512 × 512 pixels were used.

For PCD-CT, a pure FBP-type reconstruction algorithm for spectral results is not available However, “QIR-off” is offered in which minimally possible statistical optimization is achieved compared to standard-weighted FBP. QIR strength levels 1–4 trigger an additional statistical optimization in terms of a globally reduced target noise level. Specifically, higher strength levels lead to stronger optimization, i.e., greater noise reduction.²⁸

Phantom study

Phantom experiments were performed using a physiological, high-end generation III cardiovascular phantom (Figure 1) as described elsewhere in detail.²⁹ The phantom includes interconnected cerebral, thoracic, abdominal, and peripheral vasculature with hemodynamically accurate flows however, the focal region of this evaluation was the cardiac frame. The phantom includes a highly physiological heart model with four chambers and embedded tricuspid, pulmonary, mitral, and aortic valves. Modular connection points at the base of the aortic root lead to left and right coronary branching models overlaid on the exterior of the heart model. The coronary models are capable of being exchanged between healthy and diseased versions, including with soft (i.e., non-calcified) or densely calcified plaques of varying occlusion percentages. In this investigation, the left coronary model (Figure 2) included three plaques composed of deposited calcium carbonate with occlusion percentages of 50% in diagonal branch of the left anterior descending artery (D/LAD), 70% in the main branch of the LAD, and 90% in the circumflex artery (CX) as defined by the vendor. Specifically, the plaques with 50%, 70%, and 90% occlusion percentages had measurements of 1 mm lumen from 2-mm vessel inner diameter, 1.2-mm lumen from 4-mm vessel inner diameter, and 0.50 mm lumen from 5 mm inner diameter, respectively. All plaques were blow-deposited into the vessels. The length of the calcified sections was between 8 and 12 mm in length. The attenuation of the plaque material, in this case calcium carbonate crystals, measured between 200 and 600HU (120kV) across the three plaques depending on ROI and location along the plaque. The lower values than expected in some areas were explained by the manufacturer as likely micro-pores in the calcified plaque that absorbed water when the vessel was filled. Additionally, an electromechanical pump (SuperPump, ViVitro Labs, Victoria, BC, Canada) was connected to the phantom allowing for control over heart rate, stroke volume, and blood pressure, which together facilitate cardiac motion of the model.

Figure 1.

Overview of the cardiovascular phantom. The phantom setup on the CT scanner table is shown on the left. A close-up picture of the heart is shown in the middle together with a volume rendered CT image in the lower left part. A representative CT image of the heart is shown on the top right, whereby calcified plaques in the coronary artery wall are marked with small orange arrows. Close-up short- and long-axis images of a coronary plaque are shown in the lower right part of the figure.

Figure 2.

Overview of the left coronary model. The model included three plaques composed of deposited calcium carbonate with occlusion percentages of 50% in diagonal branch of the left anterior descending artery (D/LAD), 70% in the main branch of the LAD, and 90% in the circumflex artery (CX) as defined by the vendor.

For image acquisition the phantom was positioned in an acrylic container. The heart was mounted on top of two plastic 500-ml saline bottles to simulate attenuation of surrounding anatomical structures as encountered in vivo. Subsequently, the coronaries were manually filled with iodinated contrast media (Iopromide) titrated with water to a concentration of 15 mgI ml⁻¹ as recommended elsewhere.³⁰ Although connected to the phantom, the coronaries were isolated from the dynamic flow through the rest of the vasculature to ensure precise concentration of contrast media in the target vessels throughout the various trials and to protect the geometry of the coronary plaques. The dynamic flow in the remaining vasculature and the heart chambers, resulting in a pulsatile movement of the heart and coronaries, was achieved by circulating room temperature water at a heart rate of 60 beats per minute (BPM) and a stroke volume of 80 ml. Internal blood pressure of the phantom was maintained at 120/80 mm Hg.

Objective image analysis - Noise Power Spectrum

A water-filled cylindrical container mimicking an intermediate-sized patient (diameter 30 cm) was scanned with the same CT protocol as used for the vascular phantom and patient scans. Noise power spectrum (NPS) analysis was performed with an open-source software (ImQuest Version 7, Duke University) in an effort to quantify image noise texture. Specifically, quadratic ROIs with an area of 15 cm² were placed in the center of the phantom across 200 consecutive slices. 1D-NPS profiles depicting the radial average of the 2D-NPS profile were generated. The average (f_av) and peak spatial frequencies (f_peak) and the area under the curve representing the noise level of the NPS curves were extracted for further analyses.

Objective image analysis – Further Metrics

First, one reader measured coronary attenuation and the CNR by manually placing regions of interests (ROI) into the coronaries and into the water filled left ventricle. CNR was calculated as follows:

$$\mathrm{C}\mathrm{N}\mathrm{R}=\frac{{\mathrm{H}\mathrm{U}}_{\mathrm{V}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{l}}-{\mathrm{H}\mathrm{U}}_{\mathrm{W}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}\mathrm{}}{{\mathrm{S}\mathrm{D}}_{\mathrm{W}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}}$$

Furthermore, the mean HU value (i.e., HU_Vessel) within the ROI was taken as the coronary attenuation. The copy-paste function of our institutions image viewer (DeepUnity Viewer, Dedalus HealthCare) was used to copy ROIs from image to image. ROI placement was performed three times and the average values were taken for further analyses.

Second, vessel sharpness defined as the change in HU values per mm (∆HU/mm) was measured. A line profile perpendicular to a vessel (in our case the CX artery) was generated using ImageJ’s “Line Profile” function as shown elsewhere in detail³¹ (ImageJ 1.53a, National Institutes of Health, USA). Then, the maximum slopes of the regression lines for the anterior and posterior vessel border were calculated and averaged by computing the derivative of the line profiles function in R using the package “pspline”.

Third, two readers (board-certified radiologists with 5 and 10 years of experience in cardiovascular imaging, respectively) independently and blindly measured and recorded the maximum luminal diameter of all stenoses and calculated the percentage stenosis on all reconstructions. Specifically, the readers performed diameter stenosis measurements as routinely done in coronary CT angiography examinations using the electronic caliper tools of our institutions image viewer (DeepUnity Viewer, Dedalus HealthCare).

Patient study

Patients undergoing clinically indicated CCTA on the PCD-CT system between November and December 2021 were retrospectively screened. Consecutive patients for whom raw data were available for image reconstruction were reviewed (n = 14). Exclusion criteria were degraded image quality due to foreign materials (n = 2), previous coronary stent implantation (n = 1) and foregoing coronary artery bypass graft surgery (n = 1). Thus, a total of 10 patients were included in this study (Table 1).

Two board-certified radiologists (with 5 and 10 years of experience in cardiovascular imaging, respectively) performed subjective image analyses independently in a blinded and randomized manner. This study part was approved by our local ethics committee. All patients provided the written general informed consent statement of our hospital.

Readers used 5-point Likert scales (5: excellent, 4: good, 3: moderate, 2: poor, 1: non-diagnostic) to grade images qualitatively for the following categories: overall image quality, subjective image noise, vessel sharpness and, if applicable, coronary plaque visualization.

Statistical analysis

Quantitative data from the phantom scans were presented visually and by means of descriptive statistics. Generalized linear models (GLM) were fitted with keV and QIR as predictors and image quality metrics (i.e., coronary attenuation, CNR, vessel sharpness, stenosis measurements) as response variables. Bland-Altman analysis was performed on the data from stenosis measurements in the phantom using the verified lumen and occlusion dimensions provided by the vendor as the reference standard.

To check for differences in qualitative scores between the various image reconstructions of patient scans, Friedman tests with post-hoc sign tests were used. Additionally, interreader agreement of qualitative scores was quantified with Krippendorff’s α coefficients (0.0–0.20 = poor agreement, 0.21–0.40 = fair agreement, 0.41–0.60 = moderate agreement, 0.61–0.80 = substantial agreement, and 0.81–1.00 = almost perfect agreement).¹¹ Two-tailed p-values < 0.05 were considered to indicate statistical significance. All analyses were performed using R statistical software (version 4.1.1; R Core Team, R Foundation for Statistical Computing).

Results

Phantom study

Noise Power Spectrum

Average (f_av) and peak (f_peak) spatial frequencies were similar for almost all reconstructions indicating similar image texture (Table 2, Figure 3). Specifically, for all reconstructions except for 40 keV QIR-4, f_peak ranged from 0.28 to 0.3 and f_av ranged from 0.28 to 0.32. For the 40keV QIR-4 reconstruction, f_peak was 0.05 and f_av was 0.26. Thus, for the 40keV QIR-4 images, low frequency noise indicated a slightly blotchier image texture. Noise magnitude decreased both from QIR-off to QIR-4 and from 40-70keV with a maximum reduction of 68.6% (35HU vs 11HU) between 40keV QIR-off and 70keV QIR-4.

Table 2.

Overview of quantitative image quality metrics as assessed in the phantom. keV: kilo electron volt; QIR: quantum iterative reconstruction

keV level	QIR	Contrast-to-Noise Ratio	Coronary Attenuation [HU]	Noise Magnitude [HU]	Peak Spatial Frequency (fpeak)	Average Spatial Frequency (fav)	Vessel Sharpness (∆HU/mm)
40	QIR-off	26 ± 1.8	1107 ± 26	35	0.28	0.3	409 ± 98
50	QIR-off	21.8 ± 1.5	747 ± 17	29	0.28	0.31	269 ± 78
60	QIR-off	19.9 ± 1.1	528 ± 12	24	0.28	0.32	184 ± 62
70	QIR-off	18.5 ± 0.7	395 ± 8	21	0.30	0.32	153 ± 37
40	QIR-1	29.3 ± 1.9	1107 ± 26	31	0.28	0.29	415 ± 91
50	QIR-1	24.6 ± 1.7	747 ± 17	26	0.28	0.31	273 ± 74
60	QIR-1	22.4 ± 1.2	528 ± 12	21	0.28	0.32	187 ± 60
70	QIR-1	20.8 ± 0.7	395 ± 8	18	0.30	0.32	159 ± 36
40	QIR-2	33.5 ± 2.0	1107 ± 26	27	0.28	0.29	419 ± 87
50	QIR-2	28.2 ± 1.9	747 ± 17	22	0.28	0.31	278 ± 70
60	QIR-2	25.6 ± 1.3	528 ± 11	19	0.28	0.32	194 ± 56
70	QIR-2	23.7 ± 0.7	395 ± 7	16	0.30	0.32	165 ± 34
40	QIR-3	39.1 ± 2.2	1109 ± 29	22	0.28	0.28	424 ± 82
50	QIR-3	32.9 ± 2.1	749 ± 19	19	0.28	0.3	283 ± 66
60	QIR-3	29.8 ± 1.5	529 ± 12	16	0.28	0.31	195 ± 55
70	QIR-3	27.7 ± 0.8	395 ± 7	13	0.30	0.31	175 ± 33
40	QIR-4	46.1 ± 1.8	1098 ± 14	18	0.05	0.26	425 ± 42
50	QIR-4	39.3 ± 2.2	746 ± 19	15	0.28	0.29	283 ± 31
60	QIR-4	35.4 ± 1.8	527 ± 12	13	0.28	0.31	196 ± 22
70	QIR-4	32.9 ± 1.3	393 ± 7	11	0.30	0.31	158 ± 11

Figure 3.

Overview of objective image quality analysis in the phantom. (A) shows CNR and coronary attenuation (i.e., mean HU values from ROI measurements) at different keV and QIR levels. (B) indicates data from noise power spectrum analysis. CNR increases at lower keV levels and at higher QIR strength levels. Coronary attenuation increases at lower keV levels but remains virtually unaffected by the QIR strength level. NPS analysis shows comparable image texture among all reconstructions yet decreasing noise levels for higher keV levels and QIR strength levels (graph in the bottom right corner).

Coronary artery attenuation

The lowest coronary artery attenuation was found on 70keV QIR-4 images (397 ± 8 HU) and the highest attenuation on 40keV QIR-3 images (1109 ± 29 HU), corresponding to a 183% difference (Table 2 and Figure 3). keV level (p < 0.001) but not QIR strength level (p = 0.97) was a significant predictor of coronary attenuation. The effect of QIR on attenuation was negligible while decreasing the keV level considerably increased the vessel attenuation.

Contrast-to-Noise ratio

The smallest CNR was found on 70keV QIR-off images (18.5 ± 0.7) and the highest CNR was found on 40keV QIR-4 images (46.1 ± 1.8), corresponding to a 149% difference (Table 2, Figure 3). Both keV level and QIR strength level (both, p < 0.001) were significant predictors of CNR. Specifically, CNR increased at decreasing keV levels and at increasing QIR strength levels.

Vessel sharpness

The lowest vessel sharpness was measured on 70keV QIR-off images (153 ± 37 ∆HU/mm), while the highest vessel sharpness was found on 40keV QIR-4 images (425 ± 42 ∆HU/mm), corresponding to a 177% difference (Table 2, Figure 4). keV level (p < 0.001) but not QIR strength level (p = 0.8) was a significant predictor of vessel sharpness. Thus, the effect of QIR on vessel sharpness was negligible while decreasing the keV level increased the vessel sharpness.

Figure 4.

Overview of the vessel sharpness analysis and stenosis grading in the phantom. (A) shows the line profile and corresponding vessel sharpness as defined by the maximum slopes of the line profile at the vessel edges. (B) shows the stenosis measurements by both readers for all plaques stratified for keV and QIR strength levels. The black-dotted line indicates the true percentage stenosis as provided by the phantom vendor. Vessel sharpness increased at lower keV levels but remained virtually unchanged across the various QIR strength levels. Stenosis measurements were not affected by keV or QIR strength level.

Stenosis measurements

For readers 1 and 2, the average absolute percentage difference in stenosis quantification was 4%/3%/3 and 7%/4%/8% for vessel 1 (D/LAD, 50% stenosis)/vessel 2 (LAD, 70% stenosis) / vessel 3 (CX, 90% stenosis), respectively. Overall, Bland-Altman analysis revealed average errors of 1%/2%, an upper limit of agreement of 6%/12% and a lower limit of agreement of −3%/0% for stenosis grading by readers 1 and 2, respectively. For both readers, there was no impact of keV level (p = 0.23–0.89) and QIR strength level (p = 0.12–0.98) on quantification accuracy (Figure 4).

Patient study

Image examples are shown in Figure 5 and Figure 6. Interreader agreement of scores from subjective image analyses was moderate (α = 0.592). For all categories and for both readers, there were significant differences between reconstructions (all, p < 0.001).

Figure 5.

Representative images of a 70-year-old female patient with an Agatston score of 475 and a calcified plaque in the left anterior descending (LAD) artery. 40 to 70 keV VMI reconstructions at QIR-4 are shown (upper row: standard field-of-view reconstruction, lower row: magnified view of the LAD). Note the high vascular contrast on 40 keV images along with the better delineation of the vessel lumen and plaque.

Figure 6.

Representative images of a 46-year-old male patient with an Agatston score of 61 and a mixed plaque in the proximal left anterior descending (LAD) artery. 40 keV VMI reconstructions at QIR-off to QIR-4 are shown (upper row: standard field-of-view reconstruction, lower row: magnified view of the proximal LAD). Note the lower noise levels at higher levels of QIR.

Subjective image analysis revealed highest scores for 40 keV with QIR-3 and QIR-4 and for 50 keV with QIR-3 and QIR-4. For overall image quality, reader one found higher scores for 40 keV QIR-3 and QIR-4 as well as 50 keV QIR-4 relative to 50 keV QIR-3 (all, p = 0.006), while there were no differences between these four reconstructions for reader 2 (p = 0.4). For image noise, vessel sharpness, and plaque delineation, there were no significant differences between the four reconstructions both for reader 1 and 2 (p = 0.08–1). For vascular contrast, reader 1 deemed 40 keV images better than 50 keV images (p = 0.008), while reader two found no significant differences among reconstructions (p > 0.99). Thus, when considering all categories and both readers, 40 keV QIR-3 and QIR-4 reconstructions (overall image quality: 5^5,5;) received the best subjective scores.

Discussion

This study summarizes an initial experience with CCTA on a novel clinical PCD-CT system as performed with a standard imaging and image reconstruction protocol (i.e., 120 kV tube voltage and medium soft convolution kernel) including a systematic analysis of the impact of various low keV VMI reconstructions and QIR levels on subjective and objective image quality in a phantom and in patients. The phantom data indicate that low keV VMI reconstructions and high strengths levels of QIR may improve objective image quality with only minor changes in image noise texture and with no effect on quantification accuracy of coronary stenosis grading of calcified coronary plaques. Our subjective analysis of patient data suggests that 40keV QIR-3 and QIR-4 reconstructions may significantly improve image quality. Thus, when considering all data, 40keV QIR-4 reconstructions for CCTA with PCD-CT appear to be promising reconstructions that offer optimal objective and subjective image quality.

In a phantom study, Rotzinger et al investigated the performance of a prototype PCD-CT for CCTA relative to an EID-CT system.³² These authors found that PCD-CT outperformed EID-CT for the detection of coronary atherosclerosis while providing images with a lower noise magnitude, higher spatial resolution, and superior lipid core detectability.³² Boccalini et al investigated the performance of a prototype PCD-CT for imaging of coronary stents in eight patients and found improved objective and subjective image quality at a lower radiation dose than EID-CT.¹⁵ Importantly, both these authors used a different PCD-CT system and no VMI reconstructions were used.

Rajendran et al provided a representative image example of a CCTA examination as performed on the same clinical PCD-CT system used in our study, but did not perform further CCTA-specific analyses.¹⁰ Here, we provide evidence that PCD-CT enables high quality CCTA imaging and that image quality can be further optimized by choosing appropriate keV levels and QIR strength levels.

Some previous studies assessed the impact of keV level on objective and subjective image quality of dual-energy EID-CT based CCTA. Arendt et al acquired DE CCTA datasets on a third-generation dual-source EID-CT system at 70/Sn150 kV tube voltage setting and automated tube current modulation in 51 patients.² They compared 40 to 100 keV and standard linear-blended CCTA images reconstructed with an IR algorithm at strength level three and found that 40keV images were superior to the other reconstructions in terms of image quality.²

Our study also indicated that 40keV images enabled significant improvements in terms of image quality, showing an improved vessel attenuation, a higher CNR, and a higher quantitative vessel sharpness. It is important to note that the improvements in vessel sharpness at low keV levels may be also partially related to improvements in vascular contrast at low energies. The differences in attenuation between the adjacent tissue and the iodinated contrast media in the vessel lumen increases at low keV levels thus leading to a steeper increase in the attenuation curve across the vessel wall.

Interestingly, a previous study assessing the impact of IR on vessel sharpness of CCTA on an EID-CT system found a higher vessel sharpness at higher strength levels of IR.³³ This is in contrast to the results of our study using a different CT scanner system, where the QIR strength level did not significantly impact on vessel sharpness. While it remains to be determined whether our findings can be reproduced across different kernels, it should be noted that one of the main aims of QIR (as intended by the vendor) was to reduce image noise without affecting noise texture and sharpness of structures.

We found no effect of keV and QIR level on stenosis grading of calcified coronary plaques. However, it should be considered that only three stenoses were graded and further studies are needed to validate our initial results. Such studies should include invasive coronary angiography as the reference standard to validate the accuracy of various reconstruction settings for the grading of coronary stenoses. Furthermore, the 90% stenosis was underestimated by both readers. After discussion with the vendor, it is expected that this was caused by breaking away of the innermost points of the calcium carbonate deposition for the 90% occlusion during the vessel filling process, resulting in a true diameter slightly less than the manufacturer’s specification.

On the majority of previous EID-CT systems, VMI reconstructions are only available when dual-energy CT protocols are selected prior to the examination. This, however, precludes making use of the highest possible temporal resolution of the scanner.^10,17 With the intrinsic spectral capabilities of the novel PCD-CT, VMI and other dual-/multi-energy-based reconstructions (such as virtual non-contrast images or iodine maps) are available from every scan at full temporal resolution of 66 ms, and VMI were established as the new standard for routine image interpretation.^11,16

We found that vessel contrast in VMI at 40 keV was very high, with attenuation values exceeding 1’000 HU. Interestingly, a previous study using a dual-energy CCTA protocol with EID-CT showed also attenuation values in coronary arteries > 1’000 HU in 40 keV VMI.²⁷ It is obvious that such high vessel attenuation is too high, potentially negatively affecting image interpretation, for which attenuation of around 300 HU is considered sufficient.^34,35 Specifically, very high coronary vessel attenuation may obscure the difference between normal arterial wall and calcium depositions in terms of CT values, which may compromise the detection of calcifications and the grading of stenoses. Furthermore, the detection of non-calcified plaques and stenoses may also be negatively impacted due to high-density artifacts.³⁶ Thus, this very high vascular contrast at low keV images in PCD-CT can be leveraged for reductions in the contrast media dose and/or for lowering of the radiation dose. While this issue was not examined in the current study, it should definitely be addressed in future studies.

Another important aspect of image reconstruction that defines objective and subjective image quality is the reconstruction algorithm. For PCD-CT a novel IR algorithm named QIR has been introduced. In contrast to previous EID-CT systems, the PCD-CT does not offer a standard FBP algorithm because of the complexity of the multi-energy data used for VMI reconstruction. Specifically, the retrieval of monoenergetic information from the threshold data directly without proper noise reduction techniques may result in the amplification of image noise – compromising the diagnosis quality. Our results indicate that QIR, a new iterative reconstruction algorithm specifically designed for PCD-CT, improves quality substantially at all VMI-energies by reducing noise. Therein, important metrics such as vascular enhancement, vessel sharpness or image noise texture remain almost unaffected. The latter may represent an advantage over previous IR for EID-CT, which were shown to alter the noise frequency distribution thus leading to a suboptimal image impression particularly at higher strength levels.^22,24,25,37

The ultimate question remains, which reconstructions should be used for clinical interpretation of CCTA with PCD-CT. Using the current imaging parameters and reconstruction settings (i.e., 120 kV tube voltage and a medium soft convolution kernel) the 40 keV QIR-4 reconstructions enable significant improvements in image quality without compromising the accuracy of stenosis grading of calcified plaques or image noise texture. Thus, our initial results suggest that these reconstructions can be recommended for clinical image interpretation. Nevertheless, particular attention should be given to high keV reconstructions, since these have shown potential to reduce blooming artifacts.⁵

Our study has the following limitations. First, we only included a limited number of patients who were scanned on one CT scanner type at a single-center with a single imaging protocol. This prohibited us from performing subgroup analyses of patient data. Still, we performed a relatively extensive phantom study to obtain solid objective and subjective image quality metrics for CCTA with PCD-CT. Importantly, we used the most advanced vascular, anthropomorphic phantom currently available exceeding the capabilities of previous phantoms.³⁸ Nonetheless we have to acknowledge that the results of our phantom experiments may not be fully transferable to an in vivo clinical setting in patients. In this regard it should be noted that our study was further limited by including only calcified plaques into the analyses. Second, we evaluated the performance of CCTA using only one (i.e., the only currently clinically available) PCD-CT system. Finally, we limited our analysis to certain low keV levels, and other settings might have also impacted image quality.

In conclusion, this study provides initial experience of CCTA on a clinical PCD-CT system as performed with a standard imaging and image reconstruction protocol (i.e., 120 kV tube voltage and medium soft convolution kernel), suggesting that 40keV VMI reconstructions at QIR-4 may enable significant improvements in terms of objective and subjective image quality.