Professional Documents
Culture Documents
A R T I C L E I N F O A B S T R A C T
Keywords: Objectives: To compare the renal parenchyma attenuation of virtual noncontrast (VNC) images derived from
Dual-energy computed tomography multiphase renal dual-energy computed tomography (DECT) with standard noncontrast (SNC) images, and to
Virtual noncontrast determine the optimum phase for VNC images.
Standard noncontrast Materials and methods: Twenty-nine men and 16 women (mean age, 61 ± 13 years; range, 37–89 years) un-
Multiphase renal CT
derwent dynamic renal DECT (100/Sn140 kVp) were included in this institutional review board–approved
Attenuation value
retrospective study. There were four phases of the scan, which included noncontrast, corticomedullary (CMP),
Radiation dose
nephrographic (NP), and excretory phases (EP). The VNC images was generated from CMP, NP and EP. CT
numbers of SNC images and VNC images of each phases were measured in the renal cortex and medulla. Mean
standard deviation of subcutaneous fat was measured as image noise on SNC and VNC images. Radiation dose
was recorded and potential radiation dose reduction was estimated. Results were tested for statistical sig-
nificance using the unpaired t-test and agreement using Bland–Altman plot analysis.
Results: The difference in mean attenuation between SNC and each phase of VNC images were ≤4 HU. The mean
attenuation of renal cortex and medulla was 33.2 ± 4.4 HU, and 34.2 ± 4.8 HU in SNC, 33.6 ± 7.6 HU and
31.1 ± 8.3 HU in VNC of CMP, 34.8 ± 8.6 HU and 35.6 ± 8.5 HU in VNC of NP, 31.5 ± 7.6 HU and
32.4 ± 7.5 HU in VNC of EP. In VNC of CMP, the attenuation of the cortex was higher than the medulla
(p < 0.05), and the attenuation of medulla was significant lower than that of SNC (p < 0.01). In VNC of NP,
the attenuation of renal cortex was higher than SNC (p < 0.05). In VNC of EP, the attenuation of cortex and
medulla were lower than SNC (p < 0.05), and inadequate iodine subtraction in collecting system was noted.
Image noise was significantly greater in SNC (p < 0.001). Mean radiation dose reduction achievable by re-
moving the SNC was 12.3% ± 0.9%.
Conclusions: VNC images from multiphase renal DECT were similar to SNC images. Using the nephrographic
phase can gives more comparable VNC images to SNC images in renal parenchyma than other phases.
Abbreviations: DECT, dual-energy computed tomography; SNC, standard noncontrast; VNC, virtual noncontrast; CMP, corticomedullary phase; NP, nephrographic phase; EP, excretory
phase
⁎
Corresponding author at: Department of Radiology, Taipei Veterans General Hospital, No. 201, Section 2, Shih-Pai Road, Taipei, 11217, Taiwan.
E-mail address: shshen@vghtpe.gov.tw (S.-H. Shen).
https://doi.org/10.1016/j.ejrad.2018.02.001
Received 12 December 2017; Received in revised form 31 January 2018; Accepted 3 February 2018
0720-048X/ © 2018 Elsevier B.V. All rights reserved.
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
Fig. 1. Left kidney of SNC image (A), images of corticomedullary (B–D), nephrographic (E–G), and delayed phases (H–J) in the same patient. The figures from left to right were contrast-
enhanced images, iodine overlay images, and VNC images. White circles demonstrated the placement of ROIs in renal cortex and medulla at level of renal hila in similar location.
Inhomogeneous attenuation of renal parenchyma in VNC images could be noted, which was more obvious in corticomedullary phase (D). And the inadequate iodine subtraction (arrow)
in urinary collecting system was demonstrated in VNC images of excretory phase (J).
Previous studies have shown that dual-energy computed tomo- [11–14], adrenal gland [15], and kidney [3,4], it is still a controversy
graphy (DECT) can provide high-quality VNC images and a reasonable with different vendors, applications, and protocol. The aim of our study
approximation to SNC images in multiphase contrast-enhanced imaging is to determine the optimum phase to compute VNC images from a
[10,11]. Although some studies have attempted to identify the optimal multiphase CT scan of the kidneys.
phase for generating VNC images in multiphase DECT of the liver
104
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
3. Results
2.3. Image reconstruction and VNC postprocessing
Among the 45 patients analysed (29 men, 16 women; mean age
The noncontrast raw data were reconstructed using a thickness of 61 ± 13 years; range 37–89 years), 160 renal cortex and medulla ROIs
1 mm and an interval of 1 mm with kernel of I30f. The 100 kV images, were derived from each SNC and VNC image acquired in the CMP, NP,
Sn140 kV images and mixed images with ratio of 0.5 were generated and EP (overall, 1280 measurements; 10 patients received nephrectomy
from dual energy raw data, using a thickness of 1 mm and an interval of at either side of kidney). Among these patients, there were 5 patients
1 mm with kernel of Q30f. These images were imported into a dedi- with renal angiomyolipomas, 2 with complicated cysts, 9 with clear cell
cated dual-energy postprocessing software package (syngo.via, version type renal cell carcinomas, 1 with chromophobe renal cell carcinomas,
VB10B, Siemens Healthcare). The VNC images were generated from and 1 with urothelial cell carcinomas. The rest had normal finding in
each of the contrast-enhanced sequences using Liver VNC with the renal parenchyma.
manufacturer’s default setting, which is a standard three-material de- The average attenuation values of SNC and VNC images in different
composition algorithm for producing VNC images. The SNC and VNC phases were demonstrated in Table 1 and Fig. 2, and the measurements
images were then relayed to a picture archiving and communication between reader 1 and 2 gave the similar results. Statistical comparison
system (PACS) for review. between SNC and VNC images was demonstrated in Tables 1 and 2, and
105
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
Table 1
Comparison of standard noncontrast (SNC) images, virtual noncontrast (VNC) images of corticomedullary, nephrographic, and excretory phases using unpaired t-test.
Reader 1: measurements of 45 patients, Reader 2: measurement of 20 patients from the same patient group.
SNC, Standard noncontrast images.
VNC, Virtual noncontrast images.
* =p < 0.05.
** =p < 0.01.
showed that the mean differences did not exceed approximately 4 HU; measurements to compare SNC and VNC images (Fig. 3 and Table 2).
the largest difference appeared in the medulla in the CMP, with the Although most points lay within the two error bars (mean bias of 95%
VNC attenuation being lower than the SNC attenuation by 3.1 HU. The confidence interval), the distance between the upper and lower error
attenuation values of VNC images acquired in the NP were slightly bars was approximately 50 HU. The distribution of the points exhibited
higher than those of SNC images—on average 1.6 HU higher in the a trend where the difference HU (y-axis) decreased as the mean HU (x-
cortex and 1.4 HU higher in the medulla. By contrast, the attenuation axis) increased. There was a moderate significant correlation for dif-
values of VNC images acquired in the EP were slightly lower than those ference HU and mean HU, with a Pearson’s correlation coefficient from
of SNC images—on average 1.7 HU lower in the cortex and 1.8 HU −0.412 to −0.583 (Table 2).
lower in the medulla. Generally, a larger standard deviation of VNC The objective analysis of noise was based on the standard deviation
attenuation values was noted, reflecting the wider variability of at- of the attenuation values in each ROI drawn on subcutaneous fat. The
tenuation values than that of SNC images (Fig. 2). average standard deviation within these ROIs as the measurement of
For the cortex, the attenuation values of VNC images showed no image noise was 14.5 ± 1.4 for SNC images, 12.3 ± 1.8 for VNC
significant difference from SNC images in the CMP, while they were images acquired in the CMP, 12.7 ± 2.0 for VNC images acquired in
different in the other two phases. For the medulla, the attenuation the NP, and 10.8 ± 1.7 for VNC images acquired in the EP. The noise
values of VNC images in the CMP and EP were significantly lower than of SNC images was significantly higher than that of VNC images
those of SNC images. On comparing the attenuation values of the cortex (p < 0.001).
and medulla in the same phase, no significant difference was found The mean CTDlvol, DLP and effective dose for noncontrast, CMP, NP
between SNC and VNC images in both the NP and EP. However, the and EP were demonstrated in Table 3. The mean total effective dose of
attenuation values of the medulla were significantly lower than those of standard multiphase renal DECT was 17.2 ± 3.0 (range, 10.7–30.7)
the cortex in the CMP (Fig. 1). In comparisons of the three phases, NP mSv. The effective doses for the CMP, NP, and EP were higher than
showed the least difference between VNC and SNC images as well as those for SNC images. If SNC images were omitted, the average theo-
between the cortex and medulla. In VNC images acquired in the EP, retical dose reduction calculated as an effective dose value was
contrast medium retained in the collecting system was noted (Fig. 1). 12.3% ± 0.9% in this study.
Bland–Altman plot analyses were performed on the attenuation
Fig. 2. Box and whisker’s plot comparing the renal cortex and medulla Hounsfield unit (HU) values for SNC and VNC images. Note the greater attenuation variability in VNC images.
106
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
Fig. 3. Bland–Altman graphs show magnitude of attenuation difference between VNC of corticomedullary, nephrographic, and excretory phases and SNC images on renal cortex and
medulla. The mean attenuation differences were less than 4 HU, and the majority of points lie within the two error bars ( ± 1.96 standard deviations), which the distance was around 50
HU.
4. Discussion can serve as surrogates for SNC images can be produced by subtracting
the iodine from contrast-enhanced images. Previous studies have shown
VNC images from DECT can reduce radiation exposure in compar- that in isolated organ systems, DECT allowed for successful subtraction
ison with SNC images in patients requiring both noncontrast and con- of iodine-containing contrast material from single-phase contrast-en-
trast-enhanced imaging, especially in multiphase CT. VNC images that hanced images [10,17,18]. However, the reliability of VNC imaging
107
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
Table 2
Mean attenuation difference between virtual noncontrast images and standard noncontrast images using Bland–Altman analysis and Pearson’s correlation between mean attenuation and
mean attenuation difference to SNC.
Mean Attenuation Difference to Correlation of mean attenuation Mean Attenuation Difference to Correlation of mean attenuation and
SNC (Mean Bias with 95% CI) and mean difference to SNC SNC (Mean Bias with 95% CI) mean difference to SNC
Corticomedullary VNC −0.4 (22.7 to −21.7) r = −0.495, p < 0.001 3.1 (21.2 to −30) r = − 0.503, p < 0.001
Nephrographic VNC −1.6 (22.2 to −30.5) r = −0.583, p < 0.001 -1.4 (19.8 to −27.1) r = − 0.514, p < 0.001
Excretory VNC 1.7 (25.1 to −22.8) r = −0.491, p < 0.001 1.8 (26.6 to −22.8) r = − 0.412, p < 0.001
SNC, Standard noncontrast images VNC, Virtual noncontrast images CI, Confidence interval.
remains uncertain in multiorgan and multiphase applications. In mul- subtract iodine in kidney.
tiphase studies using second-generation dual-source DECT, the at- Second, we found that the mean attenuation of the renal cortex was
tenuation values of VNC images of the liver, spleen, kidney, and aorta higher than that of the medulla in VNC images acquired in the CMP.
have been reported to be overestimated in the arterial and portal phases This discrepancy between the cortex and medulla has been noted in
[12–14]. In a multiphase renal imaging study, VNC images derived previous studies with VNC images acquired in the arterial and portal
from the NP and EP exhibited overestimated attenuation values in the phases. De Cecco et al. [11,13] reported that attenuation values of the
liver and spleen, but underestimated values in the renal parenchyma renal cortex in VNC images acquired in the arterial phase were higher
[3]. In a study using third-generation dual-source DECT, the attenua- than those from the portal venous phase, whereas the medulla showed
tion values of VNC images of liver were similar to those of SNC images higher attenuation in the portal venous phase than in the arterial phase.
in the arterial and portal phases, but were underestimated in the The factors that have been postulated to affect iodine subtraction in-
kidney, pancreas, adrenal gland, and aorta [11]. clude large patient size, areas of beam-hardening artefacts, very high
In this study, we used second-generation dual-source DECT, focused iodine concentrations, scatter and partial volume effect [10,21–23]. In
on multiphase renal imaging, and evaluated the estimated attenuation the CMP, the renal cortex shows intense enhancement due to pre-
of VNC images in the cortex and medulla separately. We used small size ferential arterial flow in the cortex and glomerular filtration of the
of ROI (0.2–0.5 cm2) to measure cortex and medulla. Although at- contrast material, whereas the medulla remains relatively less enhanced
tenuation measurements are known to change with different size of ROI [24,25]. This discrepant attenuation may be related to the higher iodine
[19,20], the constant size and location of ROIs were selected in our concentration in the cortex, creating a beam-hardening effect and
study to lessen the impact. thereby producing the lower VNC attenuation in the medulla. This
The principle of creating VNC images is based on three-material phenomenon raises a concern about the reliability of attenuation
deposition algorithm, assuming that each voxel was composed of fat, measurement on VNC images of lesions with high contrast enhance-
soft tissue and iodine. Our results showed that the mean attenuation ment differences, and further investigation is necessary.
difference between VNC and SNC images was less than 4 HU in all Third, we found that the noise of SNC images was significantly
phases for both the cortex and medulla; however, the mean bias was higher than the noise of VNC images. Similar observations have been
greater than 20 HU in Bland–Altman spot (Table 2, Fig. 3). This indicate reported previously [10–14,21,22]. The lower noise of VNC images is
that the attenuation values difference between SNC and VNC were likely due to smoothing induced by the postprocessing algorithm
greater than 20 HU in some measurements, which may challenge the [12,13]. Moreover, the higher tube voltage was used in postcontrast
routine substitution of VNC images in clinical applications that rely on scanning, which gives the lower noise comparing to noncontrast scan-
attenuation value measurements. This finding is comparable with ning. Although the image noise was lower in VNC images, the in-
Miller et al. [21], who showed the substantial variations in inter- adequate iodine subtraction, mildly blurred organ margins, possibly
individual VNC attenuation in different organs. They proposed a pixel- attributable to imperfect image fusion between the two datasets due to
by-pixel organ based algorithm to correct the variations caused by motion artefacts or to the denoising algorithm applied in postproces-
different organ perfusion. Moreover, in Bland–Altman analysis, a sing, were also found in our study.
moderate negative linear relationship was found between mean at- Although it has been reported that the contrast material phase does
tenuation and mean attenuation difference of SNC and VNC (Table 2). not have an effect on VNC attenuation values [3,11,12,14], there is no
This indicates the overestimated attenuation of VNC in higher esti- agreement on which phase is optimal for generating VNC images in
mated attenuation while underestimated attenuation of VNC in lower multiphase CT scan. In liver images, different authors have re-
estimated attenuation. This finding is also consistent with previous commended the arterial phase [11,14] or the portovenous phase
research showing the difference between SNC and VNC attenuation [12,13] for generating VNC images. In the present study, we found a
appeared linearly dependent on SNC attenuation [21]. These sub- trend of overestimated attenuation values of VNC images in the NP and
stantial interindividual variations in virtual unenhanced attenuation underestimated values in the EP. Additionally, for the renal medulla,
reflect the fact that the algorithm for iodine subtraction is not perfect, CMP and EP VNC attenuation values were both significantly lower than
because the kidney is not only composed of fat and soft tissue, but also SNC attenuation values. In the EP, inadequate iodine subtraction was
blood, and urine. Therefore, other algorithms based on different three noted in the urinary collecting system (Fig. 1). Our result confirmed
materials may more accurately decompose DECT data and successfully that the scanning phase, the enhancement pattern of interest organ, and
Table 3
CTDlvol, DLP and effective dose of noncontrast, corticomedullary, nephrographic, and excretory phases. Data are means ± standard deviations (range).
Noncontrast phase 7.2 ± 1.6 (4.2–11.7) 141.6 ± 35.6 (82.1–246.7) 2.1 ± 0.5 (1.2–3.7)
Corticomedullary phase 8.9 ± 1.9 (5.4–14.2) 177.6 ± 45.5 (106.8–315) 2.6 ± 0.7 (1.6–4.7)
Nephrographic phase 9.5 ± 1.8 (6.7–16.3) 426.1 ± 101.6 (151.3–753.3) 6.4 ± 1.5 (2.3–11.3)
Excretory phase 9.9 ± 1.9 (6.9–17.3) 402.0 ± 87.1 (259–729.4) 6.0 ± 1.3 (3.9–11.)
108
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
109
Y.-M. Lin et al. European Journal of Radiology 101 (2018) 103–110
[24] S. Sheth, E.K. Fishman, Multi-detector row CT of the kidneys and urinary tract: can it tell us, and when can we use it in abdominopelvic imaging? Radiographics 30
techniques and applications in the diagnosis of benign diseases, Radiographics 24 (4) (2010) 1037–1055 a review publication of the Radiological Society of North
(2) (2004) e20 a review publication of the Radiological Society of North America, America, Inc.
Inc. [29] T. Henzler, C. Fink, S.O. Schoenberg, U.J. Schoepf, Dual-energy CT: radiation dose
[25] B.I. Yuh, R.H. Cohan, Different phases of renal enhancement: role in detecting and aspects, AJR Am. J. Roentgenol. 199 (5 Suppl) (2012) S16–25.
characterizing renal masses during helical CT, AJR Am. J. Roentgenol. 173 (3) [30] T.R. Johnson, B. Krauss, M. Sedlmair, M. Grasruck, H. Bruder, D. Morhard, C. Fink,
(1999) 747–755. S. Weckbach, M. Lenhard, B. Schmidt, T. Flohr, M.F. Reiser, C.R. Becker, Material
[26] N. Takahashi, T.J. Vrtiska, A. Kawashima, R.P. Hartman, A.N. Primak, J.G. Fletcher, differentiation by dual energy CT: initial experience, Eur. Radiol. 17 (6) (2007)
C.H. McCollough, Detectability of urinary stones on virtual nonenhanced images 1510–1517.
generated at pyelographic-Phase dual-Energy CT, Radiology 256 (1) (2010) [31] K.H. Lee, J.M. Lee, S.K. Moon, J.H. Baek, J.H. Park, T.G. Flohr, K.W. Kim, S.J. Kim,
184–190. J.K. Han, B.I. Choi, Attenuation-based automatic tube voltage selection and tube
[27] H. Scheffel, P. Stolzmann, T. Frauenfelder, T. Schertler, L. Desbiolles, S. Leschka, current modulation for dose reduction at contrast-enhanced liver CT, Radiology 265
B. Marincek, H. Alkadhi, Dual-energy contrast-enhanced computed tomography for (2) (2012) 437–447.
the detection of urinary stone disease, Investig. Radiol. 42 (12) (2007) 823–829. [32] A. Graser, T.R. Johnson, H. Chandarana, M. Macari, Dual energy CT: preliminary
[28] C.A. Coursey, R.C. Nelson, D.T. Boll, E.K. Paulson, L.M. Ho, A.M. Neville, D. Marin, observations and potential clinical applications in the abdomen, Eur. Radiol. 19 (1)
R.T. Gupta, S.T. Schindera, Dual-energy multidetector CT: how does it work, what (2009) 13–23.
110