Image quality evaluation of dual-layer spectral CT in comparison to single-layer CT in a reduced-dose setting

Objectives To quantitatively and qualitatively evaluate image quality in dual-layer CT (DLCT) compared to single-layer CT (SLCT) in the thorax, abdomen, and pelvis in a reduced-dose setting. Methods Intraindividual, retrospective comparisons were performed in 25 patients who received at least one acquisition of all three acquisition protocols SLCTlow (100 kVp), DLCThigh (120 kVp), and DLCTlow (120 kVp), all covering the venous-phase thorax, abdomen, and pelvis with matched CTDIvol between SLCTlow and DLCTlow. Reconstruction parameters were identical between all scans. Image quality was assessed quantitatively at 10 measurement locations in the thorax, abdomen, and pelvis by two independent observers, and subjectively with an intraindividual forced choice test between the three acquisitions. Dose-length product (DLP) and CTDIvol were extracted for dose comparison. Results Despite matched CTDIvol in acquisition protocols, CTDIvol and DLP were lower for SLCTlow compared to DLCTlow and DLCThigh (DLP 408.58, 444.68, 647.08 mGy·cm, respectively; p < 0.0004), as automated tube current modulation for DLCTlow reached the lower limit in the thorax (mean 66.1 mAs vs limit 65 mAs). Noise and CNR were comparable between SLCTlow and DLCTlow (p values, 0.29–0.51 and 0.05–0.20), but CT numbers were significantly higher for organs and vessels in the upper abdomen for SLCTlow compared to DLCTlow. DLCThigh had significantly better image quality (Noise and CNR). Subjective image quality was superior for DLCThigh, but no difference was found between SLCTlow and DLCTlow. Conclusions DLCTlow showed comparable image quality to SLCTlow, with the additional possibility of spectral post-processing. Further dose reduction seems possible by decreasing the lower limit of the tube current for the thorax. Key Points • Clinical use of reduced-dose DLCT is feasible despite the required higher tube potential. • DLCT with reduced dose shows comparable objective and subjective image quality to reduced-dose SLCT. • Further dose reduction in the thorax might be possible by adjusting mAs thresholds.


Introduction
The introduction of dual-layer detector technology in computed tomography (CT) enabled the acquisition of spectral data for all performed scans without the need of an additional CT x-ray tube or additional acquisitions. Dual-layer spectral CT (DLCT) acquisitions allow material decomposition (virtual non-contrast, iodine-only imaging, and effective atomic numbers) as well as the calculation of virtual monoenergetic images. Several clinical studies have already been performed showing the advantages of DLCT for head CT for imaging intracerebral lesions and hemorrhage, for thoracic CT, for vertebral CT for differentiating bone lesions, and for abdominal CT angiographies for improved delineation of visceral arteries [1][2][3][4][5]. However, for the image acquisition of such data, a tube potential of either 140 kVp or 120 kVp is necessary to allow for spectral decomposition under the exploitation of the energy-specific x-ray absorption of different materials. In contrast to changes in tube current, changes in tube potential have a non-linear effect on radiation dose: in comparison to 80 kVp, the x-ray tube output (i.e., air kerma or exposure) is 1.5 times higher for 100 kVp, 2.5 times higher for 120 kVp, and 3.4 times higher for 140 kVp [6]. While patient radiation exposure has long been a topic of interest in CT, as CT accounts for 49-66% of overall patient radiation exposure, this interest has recently led to new regulations in the European Union via the EURATOM directive, with the deadline for implementation into federal law in 2018 [7,8]. Based on surveys, diagnostic reference levels (DRLs) at the 75th percentile of dose distribution have been defined as the limits of appropriate practice for CT acquisitions. In our institution, acquisition protocols for the thorax, abdomen, and pelvis have been previously optimized to reduce patient radiation exposure beyond these legal requirements. This was achieved, in part, by reducing tube potential from 120 to 100 kVp for single-layer CT (SLCT), resulting in an average patient dose 60% below the federal DRL for the thorax, abdomen, and pelvis, which is a DLP of 1000 mGy·cm and a CTDI vol of 13 mGy.
With the installation of a novel DLCT, vendor's recommended settings were applied including a tube potential of 120 kVp (acquisition protocol DLCT high ), which led to an increase in radiation exposure in comparison to previous reduced-dose SLCT protocols (acquisition protocol SLCT low ), albeit still remaining below the applicable diagnostic reference levels. After initial experience with DLCT high , the tube current was reduced (acquisition protocol DLCT low ) to match the CTDI vol of the reduced-dose protocol previously performed on the SLCT scanner at 100 kVp (SLCT low ).
The aim of the study was to quantitatively and qualitatively compare the image quality and radiation exposure of the SLCT acquisition protocol with reduced dose, SLCT low (41% of the national DRL), and the DLCT acquisition protocol with reduced dose, DLCT low (44% of the DRL). Thus, by comparing DLCT low and SLCT low , we want to aid in the deployment of protocols with reduced dose in DLCT with image quality comparable to previous low-tube potential SLCT low , while allowing to benefit from the spectral information in a clinical setting.

Ethics approval and consent
This retrospective study was approved by the institutional review board. The need for written informed consent was waived.

Patient selection
Ninety-six patients who underwent low-dose DLCT (DLCT low ) and previous low-dose SLCT low with clinical indication for oncological staging, since October 2018, were screened for inclusion into the study (Fig. 1). Intensive care patients and obese patients were excluded from the study, as low-dose protocols are not routinely used in these circumstances and obese patients require adapted acquisition protocols to guarantee adequate image quality. Arterial-phase acquisitions or venous-phase examinations with additional arterial angiography phase were not considered for the study to ensure comparability of the evaluated acquisitions, as those are to prone to effects of contrast agent volume, injection speed of the contrast agent, and circulation and cardiac output. In consequence, only patients with a routine CT acquisition of the thorax, abdomen, and pelvis in the venous phase were included into the study. All patients received a similar total amount of contrast agent in all their examinations with a maximum intraindividual difference of 5 ml. Patients with diffuse liver parenchyma diseases (liver cirrhosis and liver metastases), liver perfusion deficit, or intracorporal metal implants (e.g., hip prosthesis or spinal hardware) were excluded from the study. Furthermore, non-standard CT examinations as defined by differences in the acquisition protocol were excluded from the study. In summary, 31 patients were included into the study, who previously underwent 51 SLCT low examinations, 39 DLCT high examinations, and 34 DLCT low examinations with adjusted parameters (Fig. 1). The mean interval between the acquisitions with the different acquisition protocols was 224.6 ± 89.8 days (median, 190 days) between SLCT low and DLCT high and 182.2 ± 70.5 days (median, 183.5 days) between DLCT high and DLCT low . A subgroup of 25 patients had been examined with all three CT acquisition protocols and was used for qualitative analysis.

CT acquisition parameters
CT acquisitions were performed with a standard acquisition protocol for oncological staging covering the thorax, abdomen, and pelvis by a single continuous acquisition (all three scan protocols). Acquisition parameters of SLCT low (iCT, Philips Healthcare), DLCT low , and DLCT high (IQon Spectral CT, Philips Healthcare) are shown in Table 1. SLCT low was performed with a tube potential of 100 kVp, while all DLCTs were performed with a tube potential 120 kVp. For DLCT high , standard acquisition settings were used as per the manufacturer's presets. For DLCT low , the automated tube current modulation (DoseRight 3D-DOM, Philips Healthcare) was adjusted by reducing dose right index (DRI) from 17 to 13 to match the SLCT low protocol, achieving a similar predicted CTDI vol , as displayed when editing the acquisition protocol in the CT scanner interface (DLCT low 7.5 mGy, SLCT low 7.8 mGy). Accordingly, predicted average mAs was lower for DLCTl ow (74 mAs) compared to DLCT high (116 mAs).
Collimation (single collimation width 0.625 mm and total collimation width 40 mm) and pitch (0.798) was identical for all scans. All examinations were conducted in craniocaudal direction and supine position, with automatic exposure control as described above. For all acquisitions, iohexol contrast agent (AccupaqueTM 350, GE Healthcare) was used. Contrast agent application was performed using a power injector with an injection rate of 3 ml/s. A routine biphasic contrast-injection scan protocol was used covering the thorax, abdomen, and pelvis in the venous phase. The first contrast agent injection and saline solution chaser bolus was followed by a 30-s break and a second contrast agent injection with a saline solution chaser bolus. Then the CT acquisition was performed 60 s after the second contrast agent injection. The combined acquisition of thorax and abdominal region in the venous phase achieves better delineation of the lymph nodes, thoracic wall, and mediastinal soft tissue with less artifacts than a thoracic arterial phase for oncological staging [9]. As previously stated, only patients who received a venous-phase CT acquisition of the thorax, abdomen, and pelvis were included in the study.

Image reconstruction and post-processing
For all examinations, axial series with a slice thickness of 3 mm and increment of 1.5 mm were reconstructed. Images were reconstructed with a vendor-specific iterative reconstruction algorithm, IMR at level 1 (Iterative Model Reconstruction, Philips Healthcare), soft-tissue setting, and a standard abdomen setting (window center/width 40/400 HU).

Dose and image analysis
Quantitative and qualitative analysis were performed by two radiologists with 7 and 6 years of experience in abdominal radiology. For quantitative analysis, all acquisitions were included. For qualitative analysis, the most recent examination of each acquisition protocol (SLCT low , DLCT high , and DLCT low ) was included and only patients that were scanned with all three acquisition protocols were included.
Patients' anterior-posterior and lateral diameters were measured separately for the thorax at the level of tracheal bifurcation, the upper abdomen at the level of the portal vein, and the lower abdomen at the level of lumbar spine L4 (Fig. 2). Total scan lengths were calculated from the images.
Quantitative image analysis was performed by placement of regions of interest (ROIs) in axial slices in the following positions ( For all three axial slices, the slice-specific tube currentexposure time product, lateral, and anterior-posterior torso diameter were determined. An in-house developed software was used for semiautomatic quantitative image quality analysis to maintain constant ROI area size of 50 mm 2 in order to enhance reproducibility and time-efficient evaluation [10]. Noise was defined as the standard deviation (SD) of CT numbers of the subcutaneous fat on the right side of the ventral thorax or abdomen for improved comparability. Mean and SD of CT numbers were recorded and CNR was calculated as follows: The following ROIs were used to calculate CNR in comparison to the subcutaneous fat on their respective axial slice: lung parenchyma (thorax), liver parenchyma (upper abdomen), and psoas muscle (lower abdomen).
For qualitative image evaluation comparing SLCT low , DLCT high , and DLCT low (Fig. 3), a forced choice method was used for a blinded and randomized side-by-side review. For randomization, both the order of patients and evaluated series were randomized by the statistician, who was not involved in qualitative image evaluation, using the random function of EXCEL (Microsoft). Two observers independently performed the qualitative analysis on conventional CT images reconstructed from all three acquisition protocols. First, images from all three acquisition protocols were separately ranked for three body regions covered by the acquisition protocols taking into account the following criteria [11]: & Thorax: noise in the subcutaneous fat of the right thorax, delineation of trachea, bronchi, and spinal cord & Upper abdomen: noise in the right-sided subcutaneous fat, delineation of portal vein, aorta, and spinal cord & Lower abdomen: noise in the right-sided subcutaneous fat, delineation of gluteal muscles, and spinal cord Based on the rankings for the three body regions, unique ranks from 1 to 3 were allocated to the evaluated image series (i.e., observes were forced to choose which image has the best image quality compared to the others) as a total score, whereby a rank of 1 corresponded to the best image quality. Fig. 3 Image examples from the same patient from conventional reduced-dose SLCT low (column 1), DLCT low (column 2), and DLCT high (column 3) from thorax (a), upper abdomen (b), and pelvis (c). DLCT high (column 2) showed best quantitative and qualitative analysis whereas SLCT low (column 1) and DLCT low (column 3) showed comparable results DLPs, CTDI vol , and overall tube current-time product were extracted from the patient's dose report images sent to the picture archiving system. Size-specific dose estimates (SSDEs) for each patient and each body region were calculated according to the recommendations of the American Association of Physicsts in Medicine Task Group 204 [12]. CTDI vol and DLP are frequently used parameters for estimation of a patient's radiation exposure. However, SSDE is a more precise method for the patient's absorbed dose as it takes the patient's size into account [13]. To calculate SSDE for the individual anatomical region, a region-specific CTDI vol,region was estimated by dividing the CTDI vol for the complete acquisition by the average tube current-time product and multiplying it with the tube current-time product of the representative axial slice of each region (Fig. 2 The resulting estimated region-specific CTDI vol,region was multiplied with the recommended conversion coefficient depending on the lateral and AP diameter of the selected anatomical region [13].

Statistical analysis
Statistical analysis was performed using SAS Version 9.4 (SAS Institute Inc.) and SPSS Version 19.0 (IBM). Descriptive statistics were calculated, determining means and standard deviations for normal distributed data. CT numbers, image noise approximated by the standard deviation of measured CT numbers, and CNR were analyzed for differences depending on the CT scanner and the acquisition protocol using a mixed model for unbalanced analysis of variances for repeated measures.
Qualitative ratings of image quality were analyzed for differences depending on the CT scanner and the acquisition protocol using non-parametric tests (Friedman test stratified for patients and Wilcoxon signed-rank test in a multistage design).
For the evaluation of the inter-rater agreement of the quantitative analysis, the intraclass correlation coefficient (ICC) was used and for the qualitative analysis Cohen's kappa was calculated and classified according to Landis and Koch [14]. The significance level for statistical testing was set at p < 0.05.

Quantitative analysis
Although the average scan length of DLCT low was significantly shorter than that for SLCT low (p > 0.0001) with an average difference of 13.8 mm, DLP for DLCT low was significantly higher due to higher CTDI vol of DLCT low in comparison to SLCT low (444.68 mGy·cm and 408.58 mGy·cm, respectively; p = 0.0004) ( Table 2). CTDI vol of DLCT low (6.57 mGy) was 30% lower than that of DLCT high (9.34 mGy; p < 0.0001). Despite the efforts to match radiation dose, average CTDI vol was 14.4% higher for DLCT low compared to SLCT low (p < 0.0001). Similar to CTDI vol and DLP, the SSDE was lowest for SLCT low while SSDE of DLCT low was significantly less than DLCT high for all body regions (p < 0.0001). DLCT high had significantly higher CTDI vol (p < 0.0001) and higher DLP (p < 0.0001) resulting from higher tube potential in comparison to SLCT low , despite comparable scan length to SLCT low (p = 0.56).
A decrease of the mean lateral diameter of patients over time was observed, with a maximum difference/change of 10.3 mm (Table 3). There was also a statistically significant difference of both the lateral thoracic and lower abdominal diameter between the first (and earliest) examination with SLCT low in comparison to the acquisitions with DLCT high (p < 0.0006 and 0.0029, respectively), but the mean difference was only 5.6 mm (thorax) and 9.2 mm (lower abdomen). No significant differences in anterior-posterior (AP) diameter were observed for all three abdominal regions.
In general, the lowest tube current-time product was found for DLCT low in all three regions, showing a statistically significant difference compared to DLCT high in the lower and upper abdomen, but not in the thorax (Table 3). It has to be noted that for the thorax, the average tube current-time product of DLCT low was 66.1 mAs, which is close to the minimum mAs of the tube current modulation preset at 65 mAs (Fig. 4). Also, mAs for the upper and lower abdomen was lowest for DLCT low , at approximately 49-51% of SLCT low . There was no significant difference in noise between SLCT low , DLCT low , and DLCT high for the thorax and upper abdomen (p between 0.08 and 0.61). No significant difference in noise was found between SLCT low and DLCT low in the lower abdomen (p = 0.51). In general, noise was lowest for DLCT high , but the difference was only statistically significant in the lower abdomen (p < 0.05) ( Table 4). No significant difference was found for CNR in the thorax and upper abdomen between SLCT low and DLCT low . In the lower abdomen, CNR for DLCT low was significantly lower than that for DLCT high , but no significant difference was found between DLCT low and SLCT low . As for CT numbers, highest values were observed for vessels, e.g. aorta and portal vein, in comparison to organ parenchyma due to the use of contrast agent (Table 5). CT numbers of vessels in SLCT low were significantly higher than those for both DLCT low and DLCT high which can be attributed to higher CT numbers observed for contrast agents at the lower tube potential of 100 kVp versus 120 kVp. CT numbers for DLCT high and DLCT low in organ parenchyma did not show significant differences, except for the liver, which can show considerable contrast agent uptake, where the mean difference was 13.7 HU between SLCT low and DLCT high .

Qualitative analysis
No significant image quality differences were observed between SLCT low and DLCT low (p = 0.97). Overall, best image quality was observed for DLCT high (mean rank ± SD, 1.62 ± 0.62) in comparison to SLCT low (2.2 ± 0.71) and DLCT low (2.18 ± 0.72) with p < 0.01 (Fig. 5).  Fig. 4 Example of tube currenttime product curves for the three evaluated acquisition protocols in one patient. Note the changes in tube current in the different anatomical regions caused by the automated exposure control, as well as the changes during the rotation of the x-ray tube around the patient table ICC for quantitative inter-rater agreement was on average 0.89. Interobserver agreement for qualitative analysis was moderate with Cohen's kappa of 0.42 according to the classification of Landis and Koch [14].

Discussion
The necessity to increase the tube potential to at least 120 kVp for spectral imaging with current DLCT results in an increase of patient radiation exposure, compared with conventional single-layer acquisitions performed at 100 kVp. As a compensation tube current-time product might be adapted to achieve equivalent CTDI vol to standard-, reduced-, or low-dose SLCT protocols. The results in this study show that patient radiation exposure can be restored to levels close to that of reduceddose SLCT with comparable image quality by adjusting automated exposure control (i.e., DRI). Image quality was comparable between DLCT low and SLCT low despite the changes in tube potential and the resulting changes in tube current-time product. Thus, a notable dose reduction due to adjusted automatic exposure control can be achieved in DLCT protocols compared to standard protocols, but further reductions might be possible by fine-tuning the lower threshold for tube current-time product.
Previously, Van Ommen et al [15] had already shown that DLCT can be used for a broad range of applications without increasing patient radiation exposure compared to normaldose SLCT. However, neither quantitative nor qualitative image quality was evaluated in this study. Another aspect of our Table 5 Results of the quantitative image quality evaluation: CT numbers of evaluated organs and vessels in the thorax, upper abdomen, and lower abdomen. Statistical significance of results is indicated by italics Quantitative parameters SLCT low (*) DLCT low (**) DLCT high (***) p value Anatomic location * vs. ** * vs. *** ** vs. ***  study was the adaptation of DLCT to match previous lowdose SLCT low with DLCT low , which shows lower radiation doses than the current literature [15,16]. Moreover, the results show that even further dose reduction could be possible, especially in the thorax region where the automated exposure control reached the lower threshold of tube current-time product set in the acquisition protocol. According to the presented results, further dose reduction might be achieved by reducing the lower limit of the tube current from 65 to 55 mAs. According to Nagayama et al [17], DLCT showed a lower radiation dose than SLCT at equal tube potential of 120 kVp. However, with regard to image quality of conventional DLCT images, which is currently still the most frequently used reconstruction in clinical settings, no comparison was made with SLCT. Conventional CT images of dose equivalent DLCT and SLCT were already compared in a phantom study showing higher CNR but higher noise for DLCT compared to SLCT for 80-140 kVp, which was partly in agreement with the results presented here [18]. It should also be noted that despite matched CTDI vol between SLCT low and DLCT low , as configured in the acquisition protocol (Table 1), the resulting CTDI vol for DLCT low was higher than that for the SLCT low (Table 2), which could be connected to the lower limit for the tube current mentioned above. In general, the aspect of patient radiation exposure in spectral imaging has already been discussed in previous studies. No radiation dose increase is necessary for dual-source, dualenergy scans without compromises in image quality of the thorax and abdomen [19][20][21]. Rapid voltage switching dualenergy acquisition has shown ambiguous results. Singh et al showed dose equivalence to dual-source, dual-energy acquisition but with inferior image quality whereas other authors state that rapid voltage switching acquisition results in higher patient's radiation [22,23].
One limitation of the study is that the patients included in the quantitative evaluation were not scanned with all acquisition protocols, as 6 out of 31 patients were only scanned with two of the three protocols. As the quantitative evaluation did not rely on pairwise comparisons, but used an unbalanced design instead, it was possible to include patients where one of the acquisitions was missing as well as to include multiple acquisitions of the same patient. This was done to increase the statistical power of the results, as the overall number of patients included in this study was relatively low, which is another limitation. Because of the limited number of patients, the transferability of the obtained results to clinical practice may be limited as well. One of the reasons for the limited number of patients in this study is the restriction to a single acquisition protocol, which, even though it is the most frequently used acquisition protocol at that CT scanner, limits the overall number of eligible patients and also limits the transferability of the obtained results to other acquisition protocols. However, comparing different protocols at once would have led to an unnecessary complexity of the evaluation, as additional variables would have to be considered. This is also true for the inclusion of arterial-phase imaging, which is more susceptible to changes in patient habitus, cardiac output, and underlying disease (e.g., differences in enhancement patterns for different tumors, as evident in pancreatic carcinoma [24]). While the results achieved in this study are specific for the acquisition protocol and body region, and obtained from a limited number of patients, and reductions in patient radiation exposure may be larger or smaller for other protocols or patient collectives, which has to be the subject of further evaluation in the future, we do expect that the results are nonetheless fairly generalizable, as we chose to evaluate a single, widely applicable acquisition protocol that makes up the bulk of acquisitions at the evaluated CT scanner. Even though there is no clear consensus on the definition of low dose, with an average patient dose of 60% below the applicable DRL, the evaluated acquisition protocol in this study qualifies as "reduced dose" [25]. with best image quality observed for DLCT high . Image quality for DLCT low was equal to SLCT low Another limitation of this study and the DLCT acquisition protocols with increased tube potential might be the inferior conspicuity of vessel structures at higher tube potential, as mean photon energies closer to the k-edge of iodine provide improved image contrast [26]. However, with the additional possibility of spectral post-processing provided by DLCT, vessel delineation might be improved with virtual monochromatic images at lower virtual photon energies [27]. Furthermore, only conventional images of DLCT were analyzed in this study, even though spectral imaging was the main reason for the tube potential increase. Nonetheless, the number of potential spectral post-processing applications is very large and the quantification of the potential benefit from the additional spectral information for clinical routine is beyond the scope of this study.
Regarding the results, the inter-reader agreement for qualitative analysis was found to be moderate. However, this might be a consequence of the forced choice method used for evaluation combined with the low number of available choice (3 images). In consequence, a disagreement of ranking for one patient always consists of at least two further mismatches, resulting in a moderate Cohen's kappa. As all CT images were of diagnostic quality, the forced choice method was able to provide a side-by-side comparison, allowing to identify even small differences in image quality that might otherwise remain hidden with more conventional methods, like rating image quality on a Likert scale. Moreover, the quantitative analysis found differences in the delineation of vasculature as shown by the differences in CT numbers measured in the vessels. These differences did not lead to significant differences in subjective image quality, as delineation of vasculature is but one aspect of overall image quality. While Singh et al or Tabari et al [22,28] suggest to link the evaluation of image quality to the clinical task of lesion detection, this approach was not applicable to this study as not all patients had lesions in all three evaluated body regions. In consequence, future studies may evaluate different aspects of image quality to provide a more complete evaluation for different clinical tasks.

Conclusion
Overall, image quality of DLCT low was comparable to SLCT low and a low-dose acquisition was possible using dual-layer technique, which has the additional benefit of providing spectral information. With DLCT low , CTDI vol could be reduced by 28% compared to DLCT high , but remained above that of SLCT low . Further reduction of patient radiation exposure for the thorax seems possible.