Virtual Non-Contrast versus True Native in Photon-Counting CT: Stability of Density of Upper Abdominal Organs and Vessels

The clinical use of photon-counting CT (PCCT) allows for the generation of virtual non-contrast (VNC) series from contrast-enhanced images. In routine clinical practice, specific issues such as ruling out acute bleeding require non-contrast images. The aim of this study is to evaluate the use of PCCT-derived VNC reconstructions in abdominal imaging. PCCT scans of 17 patients including early arterial, portal venous and native sequences were enrolled. VNC reconstructions have been calculated. In every sequence and VNC reconstruction, 10 ROIs were measured (portal vein, descending aorta, inferior vena cava, liver parenchyma, spleen parenchyma, erector spinae muscle, subcutaneous adipose tissue, first lumbar vertebral body, air, and psoas muscle) and density values were compared. The VNC reconstructions show significant changes in density compared to the contrast-enhanced images. However, there were no significant differences present between the true non-contrast (TNC) and any VNC reconstructions in the observed organs and vessels. Significant differences (p < 0.05) between the measured mean density values in the TNC versus VNC reconstructions were found in fat and bone tissue. The PCCT-derived VNC reconstructions seemed to be comparable to the TNC images, despite some deviations shown in the adipose tissue and bone structures. However, the further benefits in terms of specific clinical issues need to be evaluated.


Introduction
Computed tomography (CT) is an essential imaging modality in routine clinical practice, which is constantly being optimized by technical improvements.One of the latest and most revolutionary innovations is the use of photon-counting detectors, which enable the direct measurement of photons and their energy.In contrast, traditional CT detectors rely on scintillator technology and include a separate step involving the generation of visual light, which is measured in a photodiode [1].Through the direct measurement of photons, PCCT has improved the spatial resolution and the ability to obtain energy information from the measurement of photon energy.This novel detector technology delivers additional information, which results in improved dose efficiency and spectral information [1,2].As energy information from the measurement of photon energy is detected, PCCT enables spectral reconstructions of the acquired data sets [3].Using these, Virtual Monoenergetic Images (VMIs) can be generated with post-processing techniques [4].In scans acquired using contrast media, spectral information can be used to create a virtual non-contrast (VNC) series.Subsequently, iodine-containing contrast media is identified and subtracted from the image series [5].
In clinical routine, there is a demand for non-contrast, i.e., native imaging in several specific clinical questions such as bleeding exclusion or evaluation of vessel grafts [6].If the VNC sequences can substitute true non-contrast (TNC) sequences, the scan could be shortened and radiation exposure for the patient can be reduced.
Previous studies using Dual Energy CT (DECT)-derived VNC reconstructions have shown promising results.A high level of consistency between VNC and TNC was shown in many cases, such as, for example, in the imaging of brain structures [7], abdominal organs [8][9][10][11] and vessels [12].Similarly, studies on DECT-derived data sets have repeatedly described the direct clinical benefits of VNC.For example, intramural aortic haematomas can be diagnosed using VNC reconstructions [13].Also, VNC reconstructions from DECT and PCCT have been shown to be useful for coronary calc screening in cardiac CT [14,15].However, DECT-derived VNC reconstructions also seem the be imitated in coronary calc screening, as the calc volumes measured in VNC tend to be smaller, like those measured in TNC (DOI: 10.3348/kjr.2016.17.3.321).Additionally, several studies have shown that DECT-derived VNC reconstructions cannot totally substitute TNC images.In the imaging of the neck, a significant variation in the density of the thyroid tissue has been demonstrated between VNC reconstructions and TNC [16].In the detection of kidney stones, the accuracy of DECT-derived VNC reconstructions was also limited by the surrounding iodine-enhanced contrast [17].Nevertheless, DECT-derived VNC reconstructions seem to be good additions to conventional image data even if they should also be evaluated critically.
Looking at PCCT, there is a need for studies evaluating the potential of VNC reconstructions.Further studies by Schoenbeck et al. [18] and Niehoff et al. [19] have already described the potential of VNC to substitute TNC in abdominal imaging using PCCT.However, PCCT is a novel technology and the number of patients which can be included in comparative studies is limited.In this context, there is still a need for additional studies dealing with the comparability of VNC reconstructions and TNC images in studies using standardised scan parameters.Also, the usability of VNC sequences for specific clinical issues in PCCT has to be further evaluated.
The aim of the presented study is to further evaluate and validate the application of VNC reconstructions in abdominal imaging, which is a promising approach in abdominal PCCT imaging.Additionally, this work will investigate if there is any difference between VNC reconstructions derived from the arterial or venous contrast phases.

Patient Cohort
This retrospective single-centre study was approved by the local institutional review board and was conducted in accordance with the Helsinki declaration [20].The following inclusion criteria were applied: (I.) age ≥18 years; (II.) multiphasic CT of the upper abdomen (native, early arterial, and portal venous contrast phase); (III.)PCCT performed; (IV.) patients gave written and informed consent.Exclusion criteria were (I.) age < 18 years; (II.) study incomplete or aborted; (III.)previous administration of contrast media on the same day.According to inclusion and exclusion criteria, a total of 17 patients were retrospectively enrolled in this study from December 2021 to July 2023.

CT Scan and Contrast Protocol
A clinical, CE-marked photon-counting CT scanner (NAEOTOM ALPHA, Siemens Healthineers AG, Forchheim, Germany) was used for all examinations.All scans were performed at 120 kVP tube voltage with adaptive tube current (CARE keV IQ).Rotation time was 0.25 s and pitch was set to 1.5.After native imaging of the abdomen, 60 mL iodine-based contrast agent (Iomeprol 350 mg/mL, Bracco Imaging Deutschland GmbH, Konstanz, Germany) was injected at a rate of 4.0 mL/second followed by 30 mL of saline chaser.Image acquisition was started using bolus tracking in the descending aorta at a threshold of 100 HU.After acquisition of early arterial contrast phase images, additional series were acquired in the portal venous contrast phase (60 s delay).
Reconstruction was performed with a dedicated algorithm and a fixed kernel using medium denoising strength (Quantitative Iterative Reconstruction QR40 and Q3).All images were reconstructed on a 512 × 512 matrix with a slice thickness and section increment of 1.0 mm and 0.5 mm, respectively.Additionally, corresponding VNC reconstructions of native (VNCn), early arterial (VNCa) and portal venous (VNCv) contrast phase were created in the vendor console (Syngo.Via, version VB60A, Siemens Healthineers AG, Forchheim, Germany).Figure 1 illustrates post-processing of the acquired images.Konstanz, Germany) was injected at a rate of 4.0 mL/second followed by 30 mL of saline chaser.Image acquisition was started using bolus tracking in the descending aorta at a threshold of 100 HU.After acquisition of early arterial contrast phase images, additional series were acquired in the portal venous contrast phase (60 s delay).
Reconstruction was performed with a dedicated algorithm and a fixed kernel using medium denoising strength (Quantitative Iterative Reconstruction QR40 and Q3).All images were reconstructed on a 512 × 512 matrix with a slice thickness and section increment of 1.0 mm and 0.5 mm, respectively.Additionally, corresponding VNC reconstructions of native (VNCn), early arterial (VNCa) and portal venous (VNCv) contrast phase were created in the vendor console (Syngo.Via, version VB60A, Siemens Healthineers AG, Forchheim, Germany).Figure 1 illustrates post-processing of the acquired images.

Structured Measurements
A standardized and structured measurement workflow was performed on polyenergetic (T3D reconstructions) images of the abdomen in native, early arterial and portal venous contrast phase and also in the corresponding VNC reconstructions (VNCn, VNCa and VNCv).According to the standardized measurement workflow, circular 2D regions of interest (ROI) with 1cm diameter were placed in the portal vein (PV), in the descending aorta (DA), in the inferior vena cava (IVC), in the liver parenchyma, in the spleen parenchyma, in the erector spinae muscle (ESM), in the subcutaneous adipose tissue (SAT), in the first lumbar vertebral body (L1), in the surrounding air ventral of the abdomen and in the psoas muscle (PM).In case the target structure was smaller than 1 cm, the size of the circular ROI was adjusted by the reader accordingly.A detailed overview of the measured structures is given in Figure 2. All measurements were performed by the same observer.

Structured Measurements
A standardized and structured measurement workflow was performed on polyenergetic (T3D reconstructions) images of the abdomen in native, early arterial and portal venous contrast phase and also in the corresponding VNC reconstructions (VNCn, VNCa and VNCv).According to the standardized measurement workflow, circular 2D regions of interest (ROI) with 1cm diameter were placed in the portal vein (PV), in the descending aorta (DA), in the inferior vena cava (IVC), in the liver parenchyma, in the spleen parenchyma, in the erector spinae muscle (ESM), in the subcutaneous adipose tissue (SAT), in the first lumbar vertebral body (L1), in the surrounding air ventral of the abdomen and in the psoas muscle (PM).In case the target structure was smaller than 1 cm, the size of the circular ROI was adjusted by the reader accordingly.A detailed overview of the measured structures is given in Figure 2. All measurements were performed by the same observer.4), in the spleen parenchyma (5), in the psoas muscle (6), in the subcutaneous adipose tissue (7), in the surrounding air (8), in the erector spinae muscle (9) and in the first lumbar vertebral body (10).

Statistical Analysis
Statistical analysis was performed with R statistics (R Statistics, Version 4.1.0,R Core Team, Vienna, Austria)) [21].Mean values and standard deviation (SD) were calculated for quantitative variables; percentages (%) were calculated for categorical variables.Comparison tables were created with the tableone package (version 0.13.2).Boxplots were cre-  1), in the descending aorta (2), in the inferior vena cava (3), in the liver parenchyma (4), in the spleen parenchyma (5), in the psoas muscle (6), in the subcutaneous adipose tissue (7), in the surrounding air (8), in the erector spinae muscle (9) and in the first lumbar vertebral body (10).

Statistical Analysis
Statistical analysis was performed with R statistics (R Statistics, Version 4.1.0,R Core Team, Vienna, Austria)) [21].Mean values and standard deviation (SD) were calculated for quantitative variables; percentages (%) were calculated for categorical variables.Comparison tables were created with the tableone package (version 0.13.2).Boxplots were created with ggplot2 (version 3.3.5).Bland-Altman plots were created with the BlandAltmanLeh package (version 0.3.1).An additional comparison of the mean values was performed by Kruskal-Wallis test using SPSS (Statistical Package for Social Science, version 29.0.0.0,IBM, Armonk, NY, USA).

Patient Cohort
The presented retrospective single-centre study included 17 patients (6 female), in accordance with the mentioned inclusion criteria.The mean age of the patients included was 75.65 years.The examinations were performed to examine several clinical questions (staging of oncological disease: 9, vessel evaluation: 7 and bleeding examination: (1).An overview of the patient cohort is provided in Table 1.

Structured Measurement
Based on the used measurement scheme, the mean values and SD of CT attenuation in HU were measured in all the patients using polyenergetic images (true non-contrast, early arterial and portal venous) and in the corresponding VNC reconstructions (VNCn, VNCa and VNCv), resulting in a total of 1020 ROI.The measured mean values and standard deviations are given in Table 2. Compared to the mean values measured using the polyenergetic TNC images, these were significantly higher than the mean values measured using the polyenergetic portal venous contrast phase in the PV, VCI, liver parenchyma, spleen parenchyma, ESM and PM.In the corresponding VNC reconstructions, there were no significant differences observed compared to the polyenergetic TNC.The mean values of the DA and spleen parenchyma measured in the polyenergetic early arterial contrast phase were significantly higher than the mean values measured in the polyenergetic TNC.In the VNCa reconstructions, the measured mean values of the DA and spleen parenchyma showed no significant differences compared to the TNC reconstructions.The mean values of the SAT and L1 were significantly lower in all the observed VNC reconstructions (VNCn, VNCa and VNCv) compared to the polyenergetic TNC.Accordingly, there were no significant differences observed between the TNC and contrast-enhanced image series (early arterial and portal venous).The mean values of the measured density in air did not show any significant differences between the TNC images, the other observed contrast phases nor the VNC images.An overview of the calculated significance levels is given in Table 3.The boxplots given in Figure 3 illustrate that the mean value measurements of the early arterial and portal venous contrast phase differ especially in the vessels (DA, PV, VCI), muscles (PM, ESM) and parenchymal organs (liver and spleen) from the TNC reconstructions.In these structures, the VNC reconstructions of the different contrast phases also graphically approximate the TNC reconstructions.When looking at the SAT and L1, the different contrast phases differ less from each other.However, there is a considerable difference between the original sequences and the calculated VNC sequences.There were no significant differences between the VNCa and VNCv images in any of the observed structures (all p > 0.05).The mean values of the measured SD are given in Table 4 and illustrated in Figure 4.The boxplots in Figure 4 also show differences from the TNC to the early arterial and portal venous contrast phase reconstructions.Nevertheless, the differences between the measurements are also particularly noticeable when looking at the structures that absorb contrast media (DA, PV, VCI, PM, ESM, liver and spleen).Looking at the SAT and L1, the boxplots also illustrate the significant difference between the TNC and the VNC sequences.For better visualisation, the mean values were illustrated as Bland-Altman plots (Figures 5 and A1-A4), whereby the difference between the attenuation of the TNC and VNC (VNCn, VNCa and VNCv) reconstructions was plotted on the y-axis and the mean of the attenuation values was plotted on the x-axis.As an example, for the abdominal organs, the mean values of the density measurements of the liver are illustrated in Figure A1.This shows that the VNCn, VNCa and VNCv images are comparable to the TNC images regarding the liver attenuation.The mean values of the vessels are exemplified in Figure 5 by a Bland-Altman plot of the measurements of the portal vein.Figure 5 shows that the mean values measured in the portal vein are comparable between the TNC and VNC images.Considering the psoas muscle, the Bland-Altman plot of the measured mean values also indicate that the VNCn, VNCa and VNCv images are comparable to the TNC images (Figure A2).The plots of the SAT (Figure A3) and L1 (Figure A4) show, in contrast to the other plots, a wide scatter, which suggests that the VNC images are not comparable to the TNC images in these cases.In comparison, the illustrations of the Bland-Altman plots emphasise the results described and illustrated above.

Discussion
The presented study demonstrates the potential of PCCT-derived VNC reconstructions to substitute TNC sequences and add clinical value to abdominal imaging.
In routine clinical practice, there are several use cases for TNC sequences, e.g., bleeding evaluation, detection of urinary stones or exclusion of intraluminal haematoma [22][23][24].In some cases, however, these specific questions arise during further clinical procedures, when a CT with contrast medium has already been performed without a non-contrast phase.In these specific cases, it could be helpful to use the spectral reconstructions of PCCT and calculate a VNC sequence.However, there is a need for comparative studies between VNC and TNC to evaluate if they are comparable in these specific clinical issues.
In DECT, the potential of VNC in comparison to TNC is well examined, indicating that VNC can substitute TNC in several specific cases.However, in DECT, the use of VNC to detect fatty liver tissue [25,26] to differentiate adrenal adenomas from adrenal metastases [27], to detect urolithiasis [28,29] and to discriminate intraperitoneal haematoma [30] has been described previously.Despite these promising findings, pitfalls of VNC in DECT have been described.Lethi et al. were able to show a higher attenuation of vessels and higher image noise levels in VNC reconstructions derived from arterial sequences compared to TNC [31].Also, other authors have described differences in density values measured in vessels between VNC and TNC [12,32].Furthermore, Durieux et al. illustrated that VNC is not capable of replacing TNC images in abdominal imaging.Their work showed substantial differences in the measured attenuation for fluid, fat and renal tissues [33].In summary, VNC demonstrated potential in DECT for substituting TNC in specific and selected cases.Nevertheless, these reconstructions have limitations and must be used carefully.There are only a few studies that describe the use of VNC in PCCT.Due to its higher spatial resolution and image quality, the VNC images derived from PCCT may have an improved quality and the potential to improve CT diagnostics in routine clinical practice [1].In line with the findings in DECT, further work was able to demonstrate the good clinical applicability of VNC images in the evaluation of liver and spleen tissue [18], well in line with our findings.Other preparatory work has described differences between TNC and VNC reconstructions in terms of the measured attenuation values [19].This work also showed differences between the TNC and VNC images, but these were only significant in L1 and SAT.This study observed significantly lower density values of L1 and SAT measured in the VNCa, VNCn and VNCv reconstructions compared to the TNC images.These findings may be due to the algorithm on which VNC is based.However, these findings indicate that VNC reconstructions have to be used carefully in adipose and calcified structures.Previous work based on DECT-derived image data was able to show the same deviation between the TNC and VNC images of adipose tissue [27].Keeping this in mind, the evaluation of adipose structures as hepatosteatosis should be carried out carefully.Nevertheless, Niehoff et al. were able to show that PCCT-derived VNC sequences can be used for the assessment of hepatic steatosis [34].This indicates that the described differences in adipose tissue might be present but clinical use might still be possible.Looking at calcified structures, there is preliminary work available, especially in the field of cardiac imaging.Several studies have evaluated the usability of VNC reconstructions for calcium score evaluation in cardiac imaging.Using PCCT-derived image data, Sharma et al. demonstrated that calcium scores calculated from VNC images were underestimated [35].Also, in DECT, the calcium scores calculated from VNC reconstructions seemed to be lower than these calculated from the TNC reconstructions [36,37].This is well in line with our findings, as the current study also showed significantly lower mean values measured in L1 in the VNC reconstructions compared to the TNC reconstructions.However, it has also been reported that VNC images can be used for calcium score calculations [37,38] even if there is a need for optimisation.
Looking at the structures with contrast media uptake, the presented study was able to show an approximation of the mean density values.The significant differences in the mean values between the TNC and early arterial contrast phase reconstructions measured in the DA and spleen parenchyma could be brought to a non-significant level by creating VNC sequences.Likewise, the significant differences between the mean values of the TNC and portal venous contrast phase reconstructions measured in the PV, ICV, liver parenchyma, spleen parenchyma, PM and ESM were brought to a non-significant level by creating VNC reconstructions.These findings illustrate that VNC reconstructions may allow for the subtraction of the given contrast media and allow for an evaluation of parenchymal organs, muscles and vessels.In line with these results, other studies were able to show that VNC reconstructions can be used to evaluate abdominal structures equally as well as TNC, e.g., for the classification of adrenal masses [27], evaluation of liver tissue [25,39], classification of pancreatic lesions [11] or classification of renal masses [40,41].Furthermore, it was shown that the VNC reconstructions derived from venous contrast phase might be more accurate than the VNC reconstructions derived from the arterial contrast phase [42,43].These could neither be refuted nor confirmed by the current study.However, no significant differences between the VNCn, VNCa and VNCv reconstructions could be demonstrated.
The present study now allows for the basic statement that both VNCa and VNCv reconstructions represent an approach to measuring the density values that is similar to TNC reconstructions and equalise the significant differences between them.This supports the external findings listed above, describing the clinical benefits and potential applications of VNC reconstructions in the assessment of parenchymal organs.Nevertheless, according to the demonstrated deviations in the SAT and L1 and the previously described limitations [12,[31][32][33], VNC reconstructions should be used carefully.Additionally, the results of the current study show that significant differences in muscle tissue between the TNC and contrast-enhanced scans in the portal venous phase can be diminished by creating VNC reconstructions.This finding suggests that VNCv is a promising addition to the evaluation of muscular tissue in PCCT.In addition, the use of this finding, for example, for the evaluation of muscular haemorrhages, could be investigated in subsequent studies.On the other hand, our results illustrate that VNC cannot completely substitute TNC, which is highlighted by the findings for the SAT and L1.However, there were no significant differences between VNCn, VNCa and VNCv observed.
The current study was limited by its rather small group of enrolled patients, which is caused by the fact that PCCT is relatively new and so far, only a few patients have been examined with this specific protocol.Nevertheless, our study provides a proof of concept and supports existing data that show that VNC can be used carefully in specific clinical settings.However, subsequent studies should enrol a bigger, more heterogenous number of patients.For further studies, it would also be interesting to compare the PureCalcium VNC reconstructions in addition to the conventional VNC reconstructions [44].Additionally, following studies with a larger patient cohorts can also address the described results in the context of clinical issues such as bleeding examination, the evaluation of vasosclerosis or the detection of kidney stones.

Figure 1 .
Figure 1.Principles of virtual non-contrast (VNC) reconstructions.Patients with triphasic photoncounting CT (PCCT) scans of the abdomen (true non-contrast, early arterial and portal venous) have been enrolled.VNC reconstructions have been performed from true non-contrast, early arterial and portal venous contrast phase and compared to true non-contrast.

Figure 1 .
Figure 1.Principles of virtual non-contrast (VNC) reconstructions.Patients with triphasic photoncounting CT (PCCT) scans of the abdomen (true non-contrast, early arterial and portal venous) have been enrolled.VNC reconstructions have been performed from true non-contrast, early arterial and portal venous contrast phase and compared to true non-contrast.

Figure 2 .
Figure 2. Overview of the measured structures.Circular regions of interest (ROI) were placed in the portal vein (1), in the descending aorta (2), in the inferior vena cava (3), in the liver parenchyma (4), in the spleen parenchyma(5), in the psoas muscle(6), in the subcutaneous adipose tissue(7), in the surrounding air(8), in the erector spinae muscle(9) and in the first lumbar vertebral body(10).

Figure 5 .
Figure 5. Bland-Altman plots of difference in attenuation values in the portal vein.True non-contrast (TNC) is plotted against virtual non-contrast reconstructions (VNC) derived from TNC (VNCn), early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).
early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).

Figure A2 .
Figure A2.Bland-Altman plots of difference in attenuation values in the psoas muscle.True noncontrast (TNC) is plotted against virtual non-contrast reconstructions (VNC) derived from TNC (VNCn), early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).

Figure A2 . 18 Figure A3 .
Figure A2.Bland-Altman plots of difference in attenuation values in the psoas muscle.True noncontrast (TNC) is plotted against virtual non-contrast reconstructions (VNC) derived from TNC (VNCn), early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).Diagnostics 2024, 14, x FOR PEER REVIEW 15 of 18

Figure A3 .
Figure A3.Bland-Altman plots of difference in attenuation values in the subcutaneous adipose tissue.True non-contrast (TNC) is plotted against virtual non-contrast reconstructions (VNC) derived from TNC (VNCn), early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).

Figure A3 .
Figure A3.Bland-Altman plots of difference in attenuation values in the subcutaneous adipose tissue.True non-contrast (TNC) is plotted against virtual non-contrast reconstructions (VNC) derived from TNC (VNCn), early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).

Figure A4 .
Figure A4.Bland-Altman plots of difference in attenuation values in the first lumbar vertebral body.True non-contrast (TNC) is plotted against virtual non-contrast reconstructions (VNC) derived from TNC (VNCn), early arterial contrast phase (VNCa) and portal venous contrast phase (VNCv).The differences in attenuation values are plotted against mean attenuation values of both measurements (TNC vs. VNC).

Table 2 .
Measured density mean values with significance levels after comparison to TNC.The density values are given in Hounsfield units.

Table 3 .
Listing of p-values in comparison to true non-contrast.Abbreviations: native-derived virtual non-contrast (VNCn), early arterial (ea), ea-derived VNC (VNCa), portal venous (pv) and pv-derived VNC (VNCv).a: multiple comparisons were not carried out because the overall test did not show any significant differences between samples.b : after Bonferroni correction.