- Email: [email protected]
Noise characteristics of virtual monoenergetic images from a novel detector-based spectral CT scanner
European Journal of Radiology 98 (2018) 118–125
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Research article
Noise characteristics of virtual monoenergetic images from a novel detectorbased spectral CT scanner Kevin Kalisza, Negin Rassoulia, Amar Dhanantwarib, David Jordana, Prabhakar Rajiaha,c,
T
⁎
a
Department of Radiology, University Hospitals Cleveland Medical Center, Cleveland, OH, United States Philips Healthcare, Cleveland, OH, United States c Cardiothoracic Imaging, Department of Radiology, UT Southwestern Medical Center, Dallas, TX, United States b
A R T I C L E I N F O
A B S T R A C T
Keywords: Spectral Dual energy CT Noise Abdomen
Aim: To evaluate the noise characteristics of virtual monoenergetic images (VMI) obtained from a recently introduced dual-layer detector-based spectral CT (SDCT), both in a phantom and patients. Materials and methods: A cylindrical Catphan® 600 phantom (The Phantom Library, Salem NY, USA) was scanned using the SDCT. Image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were measured in VMI from 40 to 200 keV as well as conventional 120 kVp images. One hundred consecutive patients who had an abdominal CT on the SDCT were then recruited in the study. Noise, SNR and CNR were measured in the liver, pancreas, spleen, kidney, abdominal aorta, portal vein, muscle, bone, and fat, both in VMI (40–200 keV) and conventional 120 kVp images. Qualitative image analysis was performed by an independent reader for vascular enhancement and image quality on a 5 point scale (1-worst, 5-best). Results: On phantom studies, noise was low at all energies of VMI. Noise was highest at 40 keV (5.3 ± 0.2 HU), gradually decreased up to 70 keV (3.6 ± 0.2 HU), after which it remained constant up to 200 keV (3.5 ± 0.2 HU). In the patient cohort, noise was low (< 25 HU) at all the energy levels of VMI for all the regions, with the exception of bone. For example, noise in the liver was highest at 40 keV (13.2 ± 4.6 HU), steadily decreased up to 70 keV (12.0 ± 4.4 HU) and then remained constantly low up to 200 keV (11.6 ± 4.3HU). For liver, pancreas, portal vein, aorta, muscle and fat, noise at all levels of VMI was lower than of conventional images (p < 0.01). For all organs, SNR, and CNR were highest at 40 keV (6.8–34.9; 18.3–44.9, respectively) after which they gradually decreased up to 120 keV (3.4–6.5; 9.5–13.0) and then remained constant to 200 keV (2.6–5.5; 8.5–12.5). Qualitative scores of VMI up to 70 keV were significantly higher than the conventional images (p ≤ 0.01), whereas for VMI ≥ 80 keV, they were lower than conventional images (p < 0.001). Conclusion: VMI obtained from the novel SDCT scanner have low noise across the entire spectrum of energies. There are significant SNR and CNR improvements compared to conventional 120 kVp images.
1. Introduction Dual energy CT (spectral CT/multi-energetic CT) utilizes two different energy spectra at acquisition to provide more detailed material characterization of various tissues, providing information beyond what is possible with conventional single energy CT. Once a predominantly research tool, dual energy CT is now routinely used in clinical practice [1–7]. Dual energy CT technologies operate either at the source or detector level, including dual source, rapid kVp switching, split beam, dual spin, multi-layer detectors, and photon-counting detectors. Several spectral images are generated from the dual energy technology by a process of two or three material decomposition, such as iodine map,
virtual non contrast, effective atomic number and uric acid pair images [8]. Virtual monoenergetic images (VMI) are also generated, which mimic an x-ray beam composed of a single photon energy [9]. VMI at low energies are useful in enhancing vascular contrast due to higher photoelectric attenuation as the energies approach K-edge of iodine [10] and in improving lesion conspicuity [11,12]. VMI at higher energy levels have lower vascular contrast but are useful in reducing several artifacts such as beam hardening, calcium blooming and metallic artifacts [9,13]. VMI is generated by a weighted combination of photoelectric and Compton scatter basis images, and during the process of decomposition into these basis images, anti-correlated noise is introduced. Increased
⁎
Corresponding author at: UT Southwestern Medical Center, Department of Radiology, 5323 Harry Hines Boulevard, Dallas, TX, USA. E-mail addresses: [email protected] (K. Kalisz), [email protected] (N. Rassouli), [email protected] (A. Dhanantwari), [email protected] (D. Jordan), [email protected] (P. Rajiah). https://doi.org/10.1016/j.ejrad.2017.11.005 Received 15 May 2017; Received in revised form 2 November 2017; Accepted 9 November 2017 0720-048X/ © 2017 Elsevier B.V. All rights reserved.
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
2.1.1. Patient cohort The study group compromised of 100 consecutive patients who had an abdominal CT scan in the SDCT scanner from October 2013 to October 2015. This included 53 routine CT abdomen with contrast, 31 CT angiography and 16 TAVI (transcatheter aortic valve implantation) studies. Examinations were performed for several indications, including assessment of abdominal pain, renal or liver mass evaluation, pre-renal and liver transplant assessment, vascular lesions and pre-aortic valve placement evaluation. The contrast dose and timing of contrast administration varied for the different examinations depending on the protocol, body mass index and renal function. Either Isovue 370 (Bracco Diagnostics Inc, Princeton, JU) or Ultravist 350 (Bayer Healthcare, Wayne, NJ) were used with contrast dose ranging from 40 to 150 ml. All patients were scanned using 120 kVp tube voltage with mAs adapted to the body size and automatic tube current modulation. Although based on BMI, some of these patients could have been scanned at 100 kVp in a conventional equivalent scanner, a tube voltage of at least 120 kVp is required for adequate spectral separation in this SDCT scanner. The mAs was reduced correspondingly in these patients to maintain dose neutrality with the conventional scanner. The detector configuration was 64 × 0.625 mm. The pitch ranged from 0.5 to 1.17 and gantry rotation time ranged from 0.3–0.75 s depending on the clinical indication. All the patients were scanned in the supine position. Some image sets also included scans of the chest, depending on the clinical indication.
noise in low-energy VMI has been shown in both phantom studies [14–18] and in various patient cohorts [19–21], which limits its utility and diagnostic capabilities. In a dual source scanner, high noise was shown both at low and high energy VMI, with the optimal low-noise energy varying on the patient size (68, 71, 74, 77 keV for small, medium, large and extra-large phantoms respectively) [17]. Similar results have been shown in rapid kVp switching scanners, with low noise between 67 and 72 keV (least at 69 keV), and higher noise both at low and higher ends of the energy spectrum [18]. Recently noise has been reduced in dual source scanners, by using a second-generation monoenergetic plus algorithm [22,23]. VMI from 40 to 200 keV are also generated by projection space decomposition from the recently introduced detector-based spectral CT (SDCT), which has two layers of detectors, with the top layer absorbing the low energy photons and the bottom layer absorbing the high energy photons [24–26]. In this study, we sought to evaluate the noise charactersitics of these VMI from SDCT, both in phantom and patient studies.
2. Materials and methods This study was a Health Insurance Portability and Accountability Act–compliant study approved by our institutional review board. Informed consent was obtained from all the patients. Patients younger than 18 years and pregnant women were excluded from the study.
2.1.2. Dual energy image processing and monoenergy creation For both phantom and patient studies, conventional polyenergetic images at 120 kVp were generated by using combined data from both the spectral detector layers. These conventional polyenergetic images were reconstructed using iterative reconstruction algorithm (iDose4 Level 3, Philips, Cleveland OH, USA) at 2 mm thickness with 1 mm overlap and a B (standard) filter. VMI were generated from spectral raw data using a dedicated workstation (Intellispace Portal, Philips Healthcare, The Netherlands). VMI were generated at 40, 50, 60, 70, 80, 100, 120, 140, 160, 180, and 200 keV energies, at 2 mm thickness with 1 mm overlap and B (standard) filter.
2.1. Phantom experiment Phantom studies were performed using a cylindrical Catphan® 600 phantom (The Phantom Library, Salem NY, USA). The low contrast module of this phantom (CTP515) has several cylindrical cords of 40 mm length and various diameters with three contrast levels (Fig. 1a). The targets in the phantom as well as background material have equivalent effective atomic numbers, but variable density to change the effective attenuation coefficients. The phantom was scanned ten times on a SDCT prototype scanner (Philips Healthcare, Cleveland, OH, USA) for better statistical representation of the mean values. Scanning parameters were as follows: 120 kVp, 158 mAs, 0.33 s gantry rotation time and 64 × 0.625 mm collimation.
2.1.3. Image analysis Image analysis was performed on a separate workstation (thin-client Spectral Diagnostic Suite, Philips Healthcare). For phantom images, a region of interest (ROIc) was placed in the 15 mm, 1% cylindrical lowcontrast target (Fig. 1) by an independent reader. Noise was calculated as the standard deviation of the pixel values in the ROIc, and the mean Hounsfield Unit (HU) value (HUc) was calculated as the mean pixel value. A background ROI (ROIb) of the same size was also placed and the mean HU value (HUb) was calculated as the mean pixel value within the ROI. Signal to noise ratio (SNR) was calculated as: HUc/noise, and contrast to noise ratio (CNR) was calculated as (HUc − HUb)/noise. This analysis was repeated for all generated monoenergies. For patient images, multiple ROIs were placed within the liver, pancreas, spleen, renal cortex, abdominal aorta, portal vein, paraspinal musculature, vertebral body, and subcutaneous fat by an independent reader with three years experience in CT image analysis. The size of each ROI was 1 cm2, except in the smaller structures, in which case the largest possible ROI was placed. The signal was calculated as the mean HU within the ROI and noise was calculated as the standard deviation of the pixel values. SNR was calculated as: HU/Noise, and CNR was calculated as (HU − HUfat)/Noise for each tissue. This analysis was repeated for VMI at all energy levels. The effective diameter of the abdomen was measured at the location of ROI measurements and calculated as the square root of the product of the anteroposterior and transverse diameters. The patients were divided into three groups for sub analysis-small (less than 28 cm); medium (28–33 cm) and large (greater than 33 cm). Qualitative image analysis was subsequently performed independently
Fig. 1. Phantom study. CT image for evaluating the noise was obtained by scanning the CTP515 low contrast module of a Catphan® 600 phantom. An ROIc (yellow circle) was placed in the 1% cylindrical low-contrast target and ROIb of the same size (red circle) was placed in the background. Signal to noise ratio (SNR) was calculated as: HUc/noise, and contrast to noise ratio (CNR) was calculated as (HUc − HUb)/noise.
119
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
3.3. Noise analysis
by a radiologist with 17 years of experience in CT image analysis, who was blinded to the quantitative results. The conventional and VMI at multiple energies were graded on a five point scale for the degree of vascular enhancement and overall image quality. Vascular enhancement was scored as follows: 1-none, diagnosis impossible; 2-poor-low confidence; 3-moderate, satisfactory for diagnosis; 4-good; 5-excellent. Overall image quality was scored as follows: 1-poor, diagnosis not possible; 2-suboptimal, diagnosis possible with low confidence; 3-moderate, sufficient for diagnosis; 4-good; 5-excellent. Scores for individual organs were not recorded.
Noise was low at all VMI (Fig. 3, Table 2) for all the evaluated regions, with the majority being < 25 HU, with the exception of bone. For each region, the highest noise was seen at 40 keV, following which there was a mild decrease at higher energies, with the noise remaining relatively stable throughout all the energy levels. For example, noise in the liver was greatest at 40 keV (13.2 ± 4.6 HU), steadily decreased up to 70 keV (12.0 ± 4.4 HU) and then remained constantly low up to 200 keV (11.6 ± 4.3HU) (p = 0.20). Although this trend was observed among all structures, a range of noise levels was observed. At 40 keV, the lowest noise level was seen in the muscle (13.0 HU), and the highest noise level was seen in the bone (48.9 HU). At 200 keV, the lowest noise was seen in the spleen (11.0 HU), whereas the highest noise was seen in the bone (23.7 HU). For liver, pancreas, portal vein, aorta, muscle and fat, noise at all levels of VMI was lower than that of conventional images, whereas for kidney noise at energies ≥50 keV was lower than conventional image and for bone noise at energies ≥70 keV were lower than that of conventional images. The noise curve for a representative organ (liver) is shown in Fig. 4.
2.1.4. Statistical analysis All data were analyzed by using dedicated statistical software (Stata version 10.0, Stata Corporation, College station, TX). Quantitative data were expressed as mean ± standard deviation. Paired t tests were used to compare image noise between the VMI and conventional CT images, and one-way analysis of variance (ANOVA) was used to compare noise, SNR, and CNR of the VMI images at more than two energy levels for phantom and patient studies. For analysis of qualitative results, the Wilcoxon rank-sum (Mann–Whitney) and Kruskal-Wallis tests were used in pair-wise and analysis of more than two variables, respectively. A value of p < 0.05 was considered statistically significant.
3.4. SNR and CNR analysis 3. Results
For all organs, SNR and CNR were highest at 40 keV (6.8–34.9; 18.2–44.9, respectively) after which they gradually decreased up to 120 keV (3.4–6.4; 9.5–13.0) and then remained constant up to 200 keV (2.6–5.5; 8.5–12.5) (Tables 3 and 4, Fig. 4). A significant difference in SNR and CNR was observed among tested energy levels (p < 0.001). The SNR of VMI from 40 to 80 keV were higher than conventional image (p < 0.01) for liver, portal vein, pancreas, spleen, aorta, kidney and bone, whereas for muscle, SNR of VMI 40–100 keV were higher than that of conventional image (p < 0.001). At VMI higher than these, the SNR was lower than that of conventional images. At 200 keV, SNR was significantly lower (p < 0.02) than that of conventional images for all organs except the liver, muscle, and fat. The CNR of VMI from 40 to 80 keV was higher than that of conventional image (p < 0.01) for portal vein, pancreas, spleen, aorta, muscle kidney and bone, whereas for liver, CNR of VMI 40–100 keV was higher than that of conventional images (p < 0.001). At energy levels higher than these, the CNR was lower than that of conventional images. A subgroup analysis performed on different body sizes and CT protocols also showed similar trends of noise levels across the energy ranges. Fig. 5 illustrates the noise, SNR and CNR in liver at different body sizes (small, medium, and large) and different CT protocols (routine CT, CT angiography, and TAVI).
3.1. Phantom study The noise was low at all energy levels of VMI, with maximum noise at 40 kev (5.3 ± 0.2). With increasing energy levels, the noise gradually decreased up to 70 keV (3.6 ± 0.2 HU), after which it remained constant up to 200 keV (3.5 ± 0.2 HU) (Table 1, Fig. 2a), Noise at 40 keV was significantly higher than that of the conventional 120 kVp image (4.6 ± 0.2, p < 0.001), while noise at energies of ≥ 50 keV was significantly lower than that of conventional scan (p < 0.001). A significant difference in noise was observed among tested energy levels (p < 0.001). SNR at 40 kev (5.3 ± 0.3), 50 keV (4.3 ± 0.3) and 60 keV (4.3 ± 0.3) VMI were lower than that of conventional image (12.0 ± 0.4, p < 0.001), whereas SNR at energies ≥70 keV were significantly higher than that of the conventional scan (p < 0.001). CNR at 40 keV (2.3 ± 0.2) was lower than conventional (2.4 ± 0.1, p < 0.001), whereas CNR at energies ≥50 keV were significantly higher than that of the conventional scan (p < 0.001). Plots of SNR and CNR are shown in Figs. 2b and 2c. A significant difference in SNR and CNR was observed among tested energy levels (p < 0.001). 3.2. Patient study A total of 100 patients (age 60.3 ± 17.1 years; 40 female, 60 male) were included in this analysis. VMI generation and analysis were successful for all studies. There were 25 small, 52 medium, and 23 large sized patients.
3.5. Qualitative analysis Overall trends in subjective quality for both categories paralled that of quantitative scores of noise, SNR, and CNR described previously. Vascular enhancement and overall image quality scores up to 70 keV were significantly higher (p ≤ 0.01) than that of conventional images, whereas vascular enhancement and overall image quality scores for energies 80–200 keV were significantly less than (p < 0.001) than that of conventional images (Fig. 6, Table 5). A significant difference among energies was noted both in vascular enhancement and overall image quality (p < 0.001).
Table 1 Phantom studies, with mean noise, CNR and SNR at different energy levels obtained as an average of 10 scans of the phantom. Noise Conventional 120 kVp VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
4.6 5.3 4.3 3.9 3.6 3.5 3.6 3.5 3.5 3.5 3.5 3.5
± ± ± ± ± ± ± ± ± ± ± ±
0.2 0.2 0.2 0.3 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
SNR
CNR
12.0 ± 0.4 5.3 ± 0.3 4.3 ± 0.3 11.9 ± 0.8 17.1 ± 1.2 20.4 ± 1.4 23.8 ± 1.7 25.3 ± 1.8 26.4 ± 1.9 27.2 ± 1.9 27.1 ± 2.0 27.9 ± 2.0
2.4 2.3 2.7 2.9 3.0 3.1 3.0 3.0 3.0 3.0 3.0 3.0
± ± ± ± ± ± ± ± ± ± ± ±
0.1 0.2 0.2 0.2 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
4. Discussion Our study is the first to demonstrate the low levels of noise of VMI obtained from the novel detector-based spectral CT scan across the energy levels from 40 to 200 keV, both in phantom and patient experiments. In the patient cohort, an increase in CNR and SNR at lower monoenergies was observed, due to increased contrast between the 120
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 2. Charts showing Noise (Fig. 2a), CNR (Fig. 2b) and SNR (Fig. 2c) of the phantom in conventional images as well as VMI from 40 to 200 keV. The curve for the VMI is shown in blue whereas the 120 kVp conventional image is plotted in red.
amplification of the so called anti-correlated noise. As a result of image decomposition, each basis demonstrates different noise patterns that are anti-correlated i.e, when noise is high in one basis image (e.g photoelectric image), it will be lower in the other basis image (i.e. Compton scatter image) [27–29]. When these basis images are weighted equally (i.e. approximately 70 keV), there is nearly exact “anti-correlation” and thus noise is minimized. However, when there is unequal weighting of basis images, as at low and high monoenergies,
tissue of interest and the fat at lower energies, while the increase in noise is less. In the phantom experiments, this trend was not observed since the contrast between the plastic phantom target and background does not increase at lower monoenergies, and remained largely constant. Therefore, the noise variation across monoenergies are reflected in the measured CNR and SNR. VMI are constructed by weighted linear combination of photoelectric and Compton scatter images and unequal weighting result in an 121
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 3. Axial CT scan of the abdomen at the level of the kidneys with the conventional 120 kVp image shown on the top left panel and the VMI from 40 to 200 keV shown in the following panels. The contrast is higher in the lower-energy levels of VMI and the contrast is lower at higher-energy levels of VMI while the noise is relatively constant across all the energy levels of VMI.
Table 2 Noise levels in patients at conventional and VMI (Means, with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Liver
Portal vein
Pancreas
Spleen
Aorta
Muscle
Kidney
Bone
15.3 (5.6) 13.2 (4.6) 12.6 (4.4) 12.1 (4.4) 12.0 (4.4) 11.9 (4.3) 11.8 (4.3) 11.7 (4.3) 11.67 (4.3) 11.67 (4.3) 11.6 (4.3) 11.6 (4.3)
18.2 (6.7) 17.5 (8.1) 16.0 (6.3) 15.3 (5.7) 15.0 (5.5) 15.0 (6.0) 14.5 (5.3) 14.4 (5.3) 14.4 (5.4) 14.4 (5.4) 14.3 (5.4) 14.34 (5.4)
18.6 17.6 16.2 15.5 15.1 14.8 14.6 14.6 14.5 14.5 14.5 14.4
14.7 14.2 12.8 12.1 11.7 11.4 11.2 11.0 11.1 11.1 11.1 11.0
20.6 18.0 17.2 16.7 16.5 16.4 16.2 16.1 16.1 16.1 16.1 16.1
14.9 13.0 12.4 12.1 11.9 11.8 11.7 11.5 11.7 11.6 11.6 11.6
18.3 21.4 18.0 16.1 15.1 14.5 14.0 13.8 13.6 13.6 13.5 13.6
32.2 48.9 38.2 32.2 28.7 26.8 24.7 23.7 23.2 23.0 22.8 23.7
(6.19) (7.1) (5.9) (5.5) (5.3) (5.2) (5.1) (5.2) (5.1) (5.1) (5.1) (5.1)
(5.6) (8.1) (5.9) (4.9) (4.5) (4.4) (4.3) (4.3) (4.4) (4.3) (4.3) (4.4)
(7.0) (7.8) (6.4) (5.9) (5.7) (5.6) (5.6) (5.5) (5.6) (5.6) (5.6) (5.6)
(5.5) (4.7) (4.4) (4.3) (4.3) (4.2) (4.2) (4.0) (4.2) (4.2) (4.2) (4.2)
(6.1) (12.9) (8.4) (6.3) (5.4) (5.2) (5.0) (5.0) (5.1) (5.0) (5.04) (5.0)
Fat (11.8) (31.9) (21.7) (16.1) (13.0) (11.3) (9.7) (9.0) (8.8) (8.5) (8.4) (12.)
14.3 12.9 11.8 11.2 10.8 10.6 10.4 10.3 10.4 10.2 10.2 10.2
(4.8) (4.7) (4.0) (3.6) (3.5) (3.4) (3.3) (3.2) (3.4) (3.2) (3.2) (3.2)
correlated noise reduction algorithm in the reconstruction process [27,36,37]. Total variation minimization approached has been shown to be efficient for de-noising CT images [36,37]. A modified total variation minimization approach has been recently introduced and has been shown to be effective in reducing anti-correlated noise in basis images from detector based spectral systems [37]. A direct benefit of the reduction of anti-correlated noise is the reduction of the noise variation across the energies of VMI, especially at low and high keVs, where the anti-correlated noise would lead to amplified image noise. This low noise across the spectrum will potentially help in exploiting the full benefits of VMI, including improved contrast signal, improved lesion conspicuity and decreased artifact. The low noise also results in increased SNR and CNR as shown in our study. There are some limitations in our study. This is a single center study with a retrospective design. The patient cohort and CT acquisition protocols were heterogeneous; however, they reflect the routine clinical practice. In this study, we did not compare the noise, SNR and CNR of SDCT VMI with VMI obtained from other vendors. Data on this is however available in the literature as discussed above. Quantitative measurements were made only once at each level. Qualitative analysis was focused on vascular enhancement and overall image quality with a vascular perspective. We did not evaluate the image quality specific to
the noise of the generated image is higher, particularly at low energies as demonstrated in earlier studies on dual source and rapid kVp switching scanners [17,18]. Several techniques have been developed to reduce the noise in VMI. In the dual-source technology, a second-generation monoenergetic algorithm (monoenergetic plus) has been developed. This is a spatial frequency-split technique that combines the high contrast information from low energy VMI (i.e 40 keV) with the low noise of medium energy levels (i.e 70 keV) [22,23]. This has been shown to improve the SNR and CNR at low energy levels compared to first generation monoenergetic algorithm [22,23] and has been used to improve vascular contrast and lesion conspicuity in several regions including abdomen, chest, head and neck [30–34]. An energy domain noise reduction algorithm has also been used in phantom studies to reduce the image noise up to 59% and improve the CNR by 64% by exploiting information redundancies in the energy domain [35]. These techniques shift the maximal CNR from 70 to 80 keV to 40–50 keV, which is comparable to or higher than that of single energy images at the optimal tube potential [31,35]. In the dual layer SDCT system, due to perfectly registered spatial data, basis decomposition is performed in the projection domain. The anti-correlated noise is accounted for and reduced by using an anti122
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 4. Noise, SNR, CNR of the liver in the conventional images and VMI from 40 to 200 keV.
Table 3 SNR in patients at conventional and VMI (Means, with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Liver
Portal vein
Pancreas
Spleen
Aorta
Muscle
Kidney
Bone
Fat
5.8 (3.1) 13.4 (8.9) 10.6 (6.5) 8.7 (4.9) 7.5 (4.0) 6.7 (3.4) 5.8 (2.7) 5.4 (2.4) 5.2 (2.2) 5.0 (2.1) 4.9 (2.1) 5.3 (4.9)
6.8 (4.1) 21.5 (15.0) 15.4 (10.3) 11.2 (7.3) 9.6 (9.4) 7.1 (4.2) 5.3 (2.9) 4.3 (2.2) 4.4 (5.4) 3.5 (1.7) 3.3 (1.6) 3.0 (1.4)
4.9 (5.1) 14.0 (9.6) 10.3 (6.5) 7.8 (4.7) 6.3 (3.6) 5.2 (2.9) 4.0 (2.1) 3.4 (1.8) 3.0 (1.6) 2.8 (1.5) 2.7 (1.4) 2.6 (1.4)
7.5 (4.4) 20.7 (13.5) 15.9 (10.0) 12.2 (7.5) 9.9 (5.9) 8.4 (4.7) 6.7 (3.4) 5.8 (2.7) 5.3 (2.4) 5.0 (2.1) 4.8 (2.0) 4.7 (1.9)
9.5 (5.6) 34.9 (22.9) 24.6 (15.6) 17.4 (10.8) 13.0 (7.7) 10.1 (5.8) 7.1 (3.8) 5.3 (2.6) 4.5 (2.1) 3.9 (1.8) 3.6 (1.6) 3.3 (1.4)
3.6 6.8 6.4 5.0 5.1 4.3 4.1 3.8 3.7 3.6 3.6 3.6
7.6 (4.0) 22.7 (12.7) 16.8 (9.0) 12.8 (6.5) 10.1 (4.9) 7.8 (4.5) 6.3 (4.3) 4.7 (2.1) 3.9 (2.7) 3.6 (1.6) 3.4 (1.5) 3.4 (3.1)
7.4 (3.6) 13.8 (8.2) 12.0 (6.9) 10.0 (5.2) 8.9 (4.5) 8.1 (4.1) 7.0 (3.7) 6.5 (3.4) 6.0 (3.4) 5.9 (3.3) 5.7 (3.3) 5.5 (3.3)
−8.0 (3.0) 14.1 (5.23) 12.3 (4.4) 10.9 (4.2) 10.2 (3.6) −9.7 (3.3) −9.0 (3.1) −8.6 (3.4) −8.4 (2.8) −8.4 (2.8) −8.4 (2.8) −8.3 (2.8)
(1.6) (3.5) (7.7) (2.3) (5.5) (1.8) (2.3) (1.6) (1.6) (1.5) (1.5) (1.5)
Table 4 CNR in patients at conventional and VMI (Means, with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Liver
Portal vein
Pancreas
Spleen
Aorta
Muscle
Kidney
Bone
13.2 (5.1) 26.7 (10.7) 21.9 (8.4) 18.6 (7.0) 16.67 (6.0) 15.4 (5.3) 13.8 (4.6) 13.0 (4.6) 12.7 (4.2) 12.5 (4.1) 12.3 (4.0) 12.6 (6.0)
13.1 (5.4) 32.2 (16.7) 24.5 (11.7) 19.2 (8.7) 17.0 (10.4) 14.0 (6.0) 11.7 (4.2) 10.4 (3.9) 10.6 (6.2) 9.5 (3.1) 9.3 (3.1) 9.0 (3.0)
11.0 (4.3) 24.4 (12.1) 19.3 (8.6) 15.7 (6.7) 13.6 (5.4) 12.1 (4.5) 10.4 (3.4) 9.5 (3.6) 9.1 (3.1) 8.8 (3.0) 8.6 (2.9) 8.5 (2.9)
15.23 (6.3) 34.0 (16.1) 27.4 (12.2) 22.3 (9.9) 19.4 (8.1) 17.4 (6.9) 15.0 (5.1) 13.9 (5.2) 13.3 (4.6) 12.9 (4.4) 12.6 (4.3) 12.5 (4.1)
15.0 (6.7) 44.9 (23.9) 32.9 (16.5) 24.6 (11.9) 19.6 (8.8) 16.3 (6.8) 12.8 (4.7) 10.8 (3.9) 9.9 (3.3) 9.2 (3.0) 8.9 (2.8) 8.6 (2.7)
11.2 20.5 18.0 15.0 14.3 13.0 12.1 11.4 11.3 11.1 11.0 10.9
13.8 (6.0) 32.2 (15.3) 25.4 (11.2) 20.6 (8.7) 17.5 (6.9) 15.0 (5.9) 13.1 (5.8) 11.2 (4.3) 10.5 (4.1) 10.2 (3.8) 9.9 (3.6) 9.9 (4.6)
11.0 (4.3) 18.3 (9.7) 16.3 (8.0) 14.2 (6.1) 13.0 (5.3) 12.1 (4.8) 10.9 (4.3) 10.3 (4.2) 9.9 (4.0) 9.7 (4.0) 9.6 (3.9) 9.3 (3.9)
(3.9) (6.6) (9.2) (5.2) (6.7) (4.1) (4.1) (3.9) (3.6) (3.5) (3.5) (3.5)
5. Conclusion
each organ, since this is highly variable and not the focus of this study. We did not evaluate the impact of the noise in specific clinical scenarios, such as contrast improvement, lesion conspicuity or artifact reduction, and in specific patient populations and pathologies. Future work will concentrate on evaluation of this detector based scanner platform to produce qualitatively improved images aiding in evaluation of these clinical scenarios.
VMI obtained from the novel detector based spectral CT scanner has low noise across the entire spectrum of energies and is comparable to conventional images. There are significant SNR and CNR improvements compared to conventional polyenergetic images. The low noise of VMI in SDCT at various energy levels makes the VMI usable at all energy
123
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 5. Noise, CNR, SNR of the liver at different body sizes (A) and in different CT protocols (B). Body sizes were small, medium and large. CT protocols were routine CT, CT angiography and TAVI.
Fig. 6. Chart showing subjective assessment scores, ie vascular enhancement (blue) and overall image quality (grey) for the conventional (conv) images as well as VMI from 40 to 200 keV.
124
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Table 5 Image quality scores for conventional and VMI (means with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Vascular enhancement
Overall image quality
2.9 4.9 4.6 3.8 3.0 2.6 1.8 1.4 1.2 1.1 1.0 1.0
2.8 3.9 4.3 3.7 3.2 2.6 1.8 1.4 1.2 1.1 1.0 1.0
(1.4) (0.6) (0.8) (1.2) (1.3) (1.3) (1.0) (0.7) (0.4) (0.3) (0.1) (0.1)
[14] K.M. Brown, S. Zabic, T. Koeher, Comparison of ML iterative reconstruction and TVminimization for noise reduction in CT images, Proc. Fully 3D (2011) 443–446. [15] R. Alvarez, E. Seppi, A comparison of noise and dose in conventional and energy selective computed tomography, IEEE Trans. Nucl. Sci. 26 (1979) 2853–2856. [16] W.A. Kalender, P. Deak, M. Kellermeier, M. van Straten, S.V. Vollmar, pplication and patient size dependent optimization of x-ray spectra for CT, Med. Phys. 36 (2009) 993–1007. [17] L. Yu, J.A. Christner, S. Leng, J. Wang, J.G. Fletcher, C.H. McCollough, Virtual monochromatic imaging in dual-source dual-energy CT: radiation dose and image quality, Med. Phys. 38 (2011) 6371–6379. [18] K. Matsumoto, M. Jinzaki, Y. Tanami, A. Ueno, M. Yamada, S. Kuribayashi, Virtual monochromatic spectral imaging with fast kilovoltage switching: improved image quality as compared with that obtained with conventional 120-kVp CT, Radiology 259 (201) (2016) 257–262. [19] M.C.B. Godoy, S.L. Heller, D.P. Naidich, B. Assadourian, C. Leidecker, B. Schmidt, I. Vlahos, Dual-energy MDCT: comparison of pulmonary artery enhancement on dedicated CT pulmonary angiography, routine and low contrast volume studies, Eur. J. Radiol. 79 (2011) e11–e17. [20] D.F. Pinho, N.M. Kulkarni, A. Krishnaraj, S.P. Kalva, D.V. Sahani, Initial experience with single-source dual-energy CT abdominal angiography and comparison with single-energy CT angiography: image quality, enhancement, diagnosis and radiation dose, Eur. Radiol. 23 (2013) 351–359. [21] D. Hu, T. Yu, X. Duan, Y. Peng, R. Zhai, Determination of the optimal energy level in spectral CT imaging for displaying abdominal vessels in pediatric patients, Eur. J. Radiol. 83 (2014) 589–594. [22] K.L. Grant, T.G. Flohr, B. Krauss, et al., Assessment of an advanced image-based technique to calculate virtual monoenergetic computed tomographic images from a dual-energy examination to improve contrast-to-noise ratio in examinations using iodinated contrast media, Invest. Radiol. 49 (2014) 586–592. [23] C. Schabel, M. Bongers, M. Sedlmair, et al., Assessment of the hepatic veins in poor contrast conditions using dual energy CT: Evaluation of a novel monoenergetic extrapolation software algorithm, Rofo 186 (2014) 591–597. [24] R.H. Wellenberg, M.F. Boomsma, J.A. van Osch, et al., Quantifying metal artifact reduction using virtual monochromatic dual-layer detector spectral CT imaging in unilateral and bilateral total hip prostheses, Eur. J. Radiol. 88 (2017) 61–70. [25] T. Hicthethier, B. Baebler, J.R. Kroeger, et al., Monoenergetic reconstructions for imaging of coronary artery stents using spectral detector CT: in-vitro experience and comparison to conventional images, J. Cardiovasc. Comput. Tomogr. 11 (2017) 33–39. [26] N. Rassouli, H. Chalian, P. Rajiah, A. Dhanantwari, L. Landeras, Assessment of 70keV virtual monoenergetic spectral images in abdominal imaging: a comparison study to conventional polychromatic 120-kVp images, Abdom. Radiol. (NY) (April 18) (2017), http://dx.doi.org/10.1007/s00261-017-1151-2 (PMID:28421243). [27] R.E. Alvarez, A. Macovski, Energy-selective reconstructions in X-ray computerised tomography, Phys. Med. Biol. 21 (1976) 733. [28] W.A. Kalender, E. Klotz, L. Kostaridou, An algorithm for noise suppression in dual energy CT material density images, IEEE Trans. Med. Imaging 7 (1988) 218–224. [29] J.T. Dobbins, J.R. Wells, Correlated-polarity noise reduction: feasibility of a new statistical approach to reduce image noise, Proc Physics of Medical Imaging, Vol 7961 of Proceedings of SPIE (2011). [30] K.L. Grant, T.G. Flohr, B. Krauss, et al., Assessment of an advanced image-based technique to calculate virtual monoenergetic computed tomographic images from a dual-energy examination to improve contrast-to-noise ratio in examinations using iodinated contrast media, Invest. Radiol. 49 (2014) 586–592. [31] D. Marin, J.C. Ramirez-Giraldo, S. Gupta, et al., Effect of a noise-optimized secondgeneration monoenergetic algorithm on image noise and conspicuity of hypervascular liver tumors: an in vitro and in vivo study, AJR Am. J. Roentgenol. 206 (2016) 1222–1232. [32] M.H. Albrecht, J.E. Scholtz, K. Husers, et al., Advanced image based virtual monoenergetic dual-energy CT angiography of the abdomen: optimization of kiloelectron volt settings to improve image contrast, Eur. Radiol. 26 (2016) 1863–1870. [33] A. Meier, M. Wurning, L. Desboilles, et al., Advanced virtual monoenergetic images: improving the contrast of dual-energy CT pulmonary angiography, Clin. Radiol. 70 (2015) 1244–1251. [34] M.H. Albrecht, J.E. Scholtz, J. Kraft, et al., Assessment of an advanced monoenergetic reconstruction technique in dual-energy computed tomography of head and neck cancer, Eur. Radiol. 25 (2015) 2493–2501. [35] S. Leng, L. Yu, J.G. Fletcher, C.H. McCollough, Maximizing iodine contrast-to-noise ratios in abdominal CT imaging through use of energy domain noise reduction and virtual monoenergetic dual-energy CT, Radiology 276 (2015) 562–570. [36] S. Zabic, Acceleration of ML iterative algorithms for CT by the use of fast start images, Proceedings of SPIE 8313, Medical Imaging (2012) (Physics of Medical Imaging, 831339). [37] K.M. Brown, S. Zabic, S. Shechter, Impact of spectral separation in dual-energy CT with anti-correlated statistical reconstruction, Proc. Fully 3D (2015) 491–494.
(1.1) (0.6) (0.8) (1.0) (1.8) (1.1) (1.0) (0.7) (0.4) (0.4) (0.1) (0.1)
levels, particularly at low energies for enhancing vascular contrast or improving lesion conspicuity and at higher energies for decreasing artifacts. Conflict of interest
• Amar Dhanantwari is an employee of Philips healthcare • Prabhakar Rajiah has received in the past (more than 2 years ago), honoraria from Philips Healthcare. • The study was funded by institutional research grant from Philips healthcare.
References [1] C. McCollough, S. Leng, L. Yu, J.G. Fletcher, Dual- and multi-energy computed tomography: principles, technical approaches, and clinical applications, Radiology 276 (2015) 637–653. [2] T.J. Vrtiska, N. Takahashi, J.G. Fletcher, R.P. Hartman, L. Yu, A. Kawashima, Genitourinary applications of dual-energy CT, AJR Am. J. Roentgenol. 194 (2010) 1434–1442. [3] G. Girish, K.N. Glazebrook, J.A. Jacobson, Advanced imaging in gout, AJR Am. J. Roentgenol. 201 (2013) 515–525. [4] J. Paul, T.J. Vogl, E.C. Mbalisike, Oncological applications of dual-energy computed tomography imaging, J. Comput. Assist. Tomogr. 38 (2014) 834–842. [5] I. Vlahos, M.C.B. Godoy, D.P. Naidich, Dual-energy computed tomography imaging of the aorta, J. Thorac. Imaging 25 (2010) 289–300. [6] A.A. Postma, P.A.M. Hofman, A.A.R. Stadler, R.J. van Oostenbrugge, M.P.M. Tijssen, J.E. Wildberger, Dual-energy CT of the brain and intracranial vessels, AJR Am. J. Roentgenol. 199 (2012) S26–S33. [7] I. Danad, Z.A. Fayad, M.J. Willemink, J.K. Min, New applications of cardiac computed tomography: dual-energy spectral, and molecular CT imaging, JACC Cardiovasc. Imaging 8 (2015) 710–723. [8] L.-J. Zhang, J. Peng, S.-Y. Wu, et al., Liver virtual non-enhanced CT with dualsource: dual-energy CT: a preliminary study, Eur. Radiol. 20 (2010) 2257–2264. [9] E. Pessis, R. Campagna, J.-M. Sverzut, et al., Virtual monochromatic spectral imaging with fast kilovoltage switching: reduction of metal artifacts at CT, Radiogr. Rev. Publ. Radiol. Soc. N. Am. Inc. 33 (2013) 573–583. [10] R. Yuan, W.P. Shuman, J.P. Earls, et al., Reduced iodine load at CT pulmonary angiography with dual-energy monochromatic imaging: comparison with standard CT pulmonary angiography–a prospective randomized trial, Radiology 262 (2012) 290–297. [11] L. Yu, S. Leng, C.H. McCollough, Dual energy CT-based monochromatic imaging, AJR Am. J. Roentgenol. 199 (Suppl. 5) (2012) S9. [12] W.P. Shuman, D.E. Green, J.M. Busey, et al., Dual-energy liver CT: effect of monochromatic imaging on lesion detection conspicuity, and contrast-to-noise ratio of hypervascular lesions on late arterial phase, AJR Am. J. Roentgenol. 203 (2014) 601–606. [13] F. Bamberg, A. Dierks, K. Nikolaou, M.F. Reiser, C.R. Becker, T.R.C. Johnson, Metal artifact reduction by dual energy computed tomography using monoenergetic extrapolation, Eur. Radiol. 21 (2011) 1424–1429.
125
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Research article
Noise characteristics of virtual monoenergetic images from a novel detectorbased spectral CT scanner Kevin Kalisza, Negin Rassoulia, Amar Dhanantwarib, David Jordana, Prabhakar Rajiaha,c,
T
⁎
a
Department of Radiology, University Hospitals Cleveland Medical Center, Cleveland, OH, United States Philips Healthcare, Cleveland, OH, United States c Cardiothoracic Imaging, Department of Radiology, UT Southwestern Medical Center, Dallas, TX, United States b
A R T I C L E I N F O
A B S T R A C T
Keywords: Spectral Dual energy CT Noise Abdomen
Aim: To evaluate the noise characteristics of virtual monoenergetic images (VMI) obtained from a recently introduced dual-layer detector-based spectral CT (SDCT), both in a phantom and patients. Materials and methods: A cylindrical Catphan® 600 phantom (The Phantom Library, Salem NY, USA) was scanned using the SDCT. Image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were measured in VMI from 40 to 200 keV as well as conventional 120 kVp images. One hundred consecutive patients who had an abdominal CT on the SDCT were then recruited in the study. Noise, SNR and CNR were measured in the liver, pancreas, spleen, kidney, abdominal aorta, portal vein, muscle, bone, and fat, both in VMI (40–200 keV) and conventional 120 kVp images. Qualitative image analysis was performed by an independent reader for vascular enhancement and image quality on a 5 point scale (1-worst, 5-best). Results: On phantom studies, noise was low at all energies of VMI. Noise was highest at 40 keV (5.3 ± 0.2 HU), gradually decreased up to 70 keV (3.6 ± 0.2 HU), after which it remained constant up to 200 keV (3.5 ± 0.2 HU). In the patient cohort, noise was low (< 25 HU) at all the energy levels of VMI for all the regions, with the exception of bone. For example, noise in the liver was highest at 40 keV (13.2 ± 4.6 HU), steadily decreased up to 70 keV (12.0 ± 4.4 HU) and then remained constantly low up to 200 keV (11.6 ± 4.3HU). For liver, pancreas, portal vein, aorta, muscle and fat, noise at all levels of VMI was lower than of conventional images (p < 0.01). For all organs, SNR, and CNR were highest at 40 keV (6.8–34.9; 18.3–44.9, respectively) after which they gradually decreased up to 120 keV (3.4–6.5; 9.5–13.0) and then remained constant to 200 keV (2.6–5.5; 8.5–12.5). Qualitative scores of VMI up to 70 keV were significantly higher than the conventional images (p ≤ 0.01), whereas for VMI ≥ 80 keV, they were lower than conventional images (p < 0.001). Conclusion: VMI obtained from the novel SDCT scanner have low noise across the entire spectrum of energies. There are significant SNR and CNR improvements compared to conventional 120 kVp images.
1. Introduction Dual energy CT (spectral CT/multi-energetic CT) utilizes two different energy spectra at acquisition to provide more detailed material characterization of various tissues, providing information beyond what is possible with conventional single energy CT. Once a predominantly research tool, dual energy CT is now routinely used in clinical practice [1–7]. Dual energy CT technologies operate either at the source or detector level, including dual source, rapid kVp switching, split beam, dual spin, multi-layer detectors, and photon-counting detectors. Several spectral images are generated from the dual energy technology by a process of two or three material decomposition, such as iodine map,
virtual non contrast, effective atomic number and uric acid pair images [8]. Virtual monoenergetic images (VMI) are also generated, which mimic an x-ray beam composed of a single photon energy [9]. VMI at low energies are useful in enhancing vascular contrast due to higher photoelectric attenuation as the energies approach K-edge of iodine [10] and in improving lesion conspicuity [11,12]. VMI at higher energy levels have lower vascular contrast but are useful in reducing several artifacts such as beam hardening, calcium blooming and metallic artifacts [9,13]. VMI is generated by a weighted combination of photoelectric and Compton scatter basis images, and during the process of decomposition into these basis images, anti-correlated noise is introduced. Increased
⁎
Corresponding author at: UT Southwestern Medical Center, Department of Radiology, 5323 Harry Hines Boulevard, Dallas, TX, USA. E-mail addresses: [email protected] (K. Kalisz), [email protected] (N. Rassouli), [email protected] (A. Dhanantwari), [email protected] (D. Jordan), [email protected] (P. Rajiah). https://doi.org/10.1016/j.ejrad.2017.11.005 Received 15 May 2017; Received in revised form 2 November 2017; Accepted 9 November 2017 0720-048X/ © 2017 Elsevier B.V. All rights reserved.
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
2.1.1. Patient cohort The study group compromised of 100 consecutive patients who had an abdominal CT scan in the SDCT scanner from October 2013 to October 2015. This included 53 routine CT abdomen with contrast, 31 CT angiography and 16 TAVI (transcatheter aortic valve implantation) studies. Examinations were performed for several indications, including assessment of abdominal pain, renal or liver mass evaluation, pre-renal and liver transplant assessment, vascular lesions and pre-aortic valve placement evaluation. The contrast dose and timing of contrast administration varied for the different examinations depending on the protocol, body mass index and renal function. Either Isovue 370 (Bracco Diagnostics Inc, Princeton, JU) or Ultravist 350 (Bayer Healthcare, Wayne, NJ) were used with contrast dose ranging from 40 to 150 ml. All patients were scanned using 120 kVp tube voltage with mAs adapted to the body size and automatic tube current modulation. Although based on BMI, some of these patients could have been scanned at 100 kVp in a conventional equivalent scanner, a tube voltage of at least 120 kVp is required for adequate spectral separation in this SDCT scanner. The mAs was reduced correspondingly in these patients to maintain dose neutrality with the conventional scanner. The detector configuration was 64 × 0.625 mm. The pitch ranged from 0.5 to 1.17 and gantry rotation time ranged from 0.3–0.75 s depending on the clinical indication. All the patients were scanned in the supine position. Some image sets also included scans of the chest, depending on the clinical indication.
noise in low-energy VMI has been shown in both phantom studies [14–18] and in various patient cohorts [19–21], which limits its utility and diagnostic capabilities. In a dual source scanner, high noise was shown both at low and high energy VMI, with the optimal low-noise energy varying on the patient size (68, 71, 74, 77 keV for small, medium, large and extra-large phantoms respectively) [17]. Similar results have been shown in rapid kVp switching scanners, with low noise between 67 and 72 keV (least at 69 keV), and higher noise both at low and higher ends of the energy spectrum [18]. Recently noise has been reduced in dual source scanners, by using a second-generation monoenergetic plus algorithm [22,23]. VMI from 40 to 200 keV are also generated by projection space decomposition from the recently introduced detector-based spectral CT (SDCT), which has two layers of detectors, with the top layer absorbing the low energy photons and the bottom layer absorbing the high energy photons [24–26]. In this study, we sought to evaluate the noise charactersitics of these VMI from SDCT, both in phantom and patient studies.
2. Materials and methods This study was a Health Insurance Portability and Accountability Act–compliant study approved by our institutional review board. Informed consent was obtained from all the patients. Patients younger than 18 years and pregnant women were excluded from the study.
2.1.2. Dual energy image processing and monoenergy creation For both phantom and patient studies, conventional polyenergetic images at 120 kVp were generated by using combined data from both the spectral detector layers. These conventional polyenergetic images were reconstructed using iterative reconstruction algorithm (iDose4 Level 3, Philips, Cleveland OH, USA) at 2 mm thickness with 1 mm overlap and a B (standard) filter. VMI were generated from spectral raw data using a dedicated workstation (Intellispace Portal, Philips Healthcare, The Netherlands). VMI were generated at 40, 50, 60, 70, 80, 100, 120, 140, 160, 180, and 200 keV energies, at 2 mm thickness with 1 mm overlap and B (standard) filter.
2.1. Phantom experiment Phantom studies were performed using a cylindrical Catphan® 600 phantom (The Phantom Library, Salem NY, USA). The low contrast module of this phantom (CTP515) has several cylindrical cords of 40 mm length and various diameters with three contrast levels (Fig. 1a). The targets in the phantom as well as background material have equivalent effective atomic numbers, but variable density to change the effective attenuation coefficients. The phantom was scanned ten times on a SDCT prototype scanner (Philips Healthcare, Cleveland, OH, USA) for better statistical representation of the mean values. Scanning parameters were as follows: 120 kVp, 158 mAs, 0.33 s gantry rotation time and 64 × 0.625 mm collimation.
2.1.3. Image analysis Image analysis was performed on a separate workstation (thin-client Spectral Diagnostic Suite, Philips Healthcare). For phantom images, a region of interest (ROIc) was placed in the 15 mm, 1% cylindrical lowcontrast target (Fig. 1) by an independent reader. Noise was calculated as the standard deviation of the pixel values in the ROIc, and the mean Hounsfield Unit (HU) value (HUc) was calculated as the mean pixel value. A background ROI (ROIb) of the same size was also placed and the mean HU value (HUb) was calculated as the mean pixel value within the ROI. Signal to noise ratio (SNR) was calculated as: HUc/noise, and contrast to noise ratio (CNR) was calculated as (HUc − HUb)/noise. This analysis was repeated for all generated monoenergies. For patient images, multiple ROIs were placed within the liver, pancreas, spleen, renal cortex, abdominal aorta, portal vein, paraspinal musculature, vertebral body, and subcutaneous fat by an independent reader with three years experience in CT image analysis. The size of each ROI was 1 cm2, except in the smaller structures, in which case the largest possible ROI was placed. The signal was calculated as the mean HU within the ROI and noise was calculated as the standard deviation of the pixel values. SNR was calculated as: HU/Noise, and CNR was calculated as (HU − HUfat)/Noise for each tissue. This analysis was repeated for VMI at all energy levels. The effective diameter of the abdomen was measured at the location of ROI measurements and calculated as the square root of the product of the anteroposterior and transverse diameters. The patients were divided into three groups for sub analysis-small (less than 28 cm); medium (28–33 cm) and large (greater than 33 cm). Qualitative image analysis was subsequently performed independently
Fig. 1. Phantom study. CT image for evaluating the noise was obtained by scanning the CTP515 low contrast module of a Catphan® 600 phantom. An ROIc (yellow circle) was placed in the 1% cylindrical low-contrast target and ROIb of the same size (red circle) was placed in the background. Signal to noise ratio (SNR) was calculated as: HUc/noise, and contrast to noise ratio (CNR) was calculated as (HUc − HUb)/noise.
119
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
3.3. Noise analysis
by a radiologist with 17 years of experience in CT image analysis, who was blinded to the quantitative results. The conventional and VMI at multiple energies were graded on a five point scale for the degree of vascular enhancement and overall image quality. Vascular enhancement was scored as follows: 1-none, diagnosis impossible; 2-poor-low confidence; 3-moderate, satisfactory for diagnosis; 4-good; 5-excellent. Overall image quality was scored as follows: 1-poor, diagnosis not possible; 2-suboptimal, diagnosis possible with low confidence; 3-moderate, sufficient for diagnosis; 4-good; 5-excellent. Scores for individual organs were not recorded.
Noise was low at all VMI (Fig. 3, Table 2) for all the evaluated regions, with the majority being < 25 HU, with the exception of bone. For each region, the highest noise was seen at 40 keV, following which there was a mild decrease at higher energies, with the noise remaining relatively stable throughout all the energy levels. For example, noise in the liver was greatest at 40 keV (13.2 ± 4.6 HU), steadily decreased up to 70 keV (12.0 ± 4.4 HU) and then remained constantly low up to 200 keV (11.6 ± 4.3HU) (p = 0.20). Although this trend was observed among all structures, a range of noise levels was observed. At 40 keV, the lowest noise level was seen in the muscle (13.0 HU), and the highest noise level was seen in the bone (48.9 HU). At 200 keV, the lowest noise was seen in the spleen (11.0 HU), whereas the highest noise was seen in the bone (23.7 HU). For liver, pancreas, portal vein, aorta, muscle and fat, noise at all levels of VMI was lower than that of conventional images, whereas for kidney noise at energies ≥50 keV was lower than conventional image and for bone noise at energies ≥70 keV were lower than that of conventional images. The noise curve for a representative organ (liver) is shown in Fig. 4.
2.1.4. Statistical analysis All data were analyzed by using dedicated statistical software (Stata version 10.0, Stata Corporation, College station, TX). Quantitative data were expressed as mean ± standard deviation. Paired t tests were used to compare image noise between the VMI and conventional CT images, and one-way analysis of variance (ANOVA) was used to compare noise, SNR, and CNR of the VMI images at more than two energy levels for phantom and patient studies. For analysis of qualitative results, the Wilcoxon rank-sum (Mann–Whitney) and Kruskal-Wallis tests were used in pair-wise and analysis of more than two variables, respectively. A value of p < 0.05 was considered statistically significant.
3.4. SNR and CNR analysis 3. Results
For all organs, SNR and CNR were highest at 40 keV (6.8–34.9; 18.2–44.9, respectively) after which they gradually decreased up to 120 keV (3.4–6.4; 9.5–13.0) and then remained constant up to 200 keV (2.6–5.5; 8.5–12.5) (Tables 3 and 4, Fig. 4). A significant difference in SNR and CNR was observed among tested energy levels (p < 0.001). The SNR of VMI from 40 to 80 keV were higher than conventional image (p < 0.01) for liver, portal vein, pancreas, spleen, aorta, kidney and bone, whereas for muscle, SNR of VMI 40–100 keV were higher than that of conventional image (p < 0.001). At VMI higher than these, the SNR was lower than that of conventional images. At 200 keV, SNR was significantly lower (p < 0.02) than that of conventional images for all organs except the liver, muscle, and fat. The CNR of VMI from 40 to 80 keV was higher than that of conventional image (p < 0.01) for portal vein, pancreas, spleen, aorta, muscle kidney and bone, whereas for liver, CNR of VMI 40–100 keV was higher than that of conventional images (p < 0.001). At energy levels higher than these, the CNR was lower than that of conventional images. A subgroup analysis performed on different body sizes and CT protocols also showed similar trends of noise levels across the energy ranges. Fig. 5 illustrates the noise, SNR and CNR in liver at different body sizes (small, medium, and large) and different CT protocols (routine CT, CT angiography, and TAVI).
3.1. Phantom study The noise was low at all energy levels of VMI, with maximum noise at 40 kev (5.3 ± 0.2). With increasing energy levels, the noise gradually decreased up to 70 keV (3.6 ± 0.2 HU), after which it remained constant up to 200 keV (3.5 ± 0.2 HU) (Table 1, Fig. 2a), Noise at 40 keV was significantly higher than that of the conventional 120 kVp image (4.6 ± 0.2, p < 0.001), while noise at energies of ≥ 50 keV was significantly lower than that of conventional scan (p < 0.001). A significant difference in noise was observed among tested energy levels (p < 0.001). SNR at 40 kev (5.3 ± 0.3), 50 keV (4.3 ± 0.3) and 60 keV (4.3 ± 0.3) VMI were lower than that of conventional image (12.0 ± 0.4, p < 0.001), whereas SNR at energies ≥70 keV were significantly higher than that of the conventional scan (p < 0.001). CNR at 40 keV (2.3 ± 0.2) was lower than conventional (2.4 ± 0.1, p < 0.001), whereas CNR at energies ≥50 keV were significantly higher than that of the conventional scan (p < 0.001). Plots of SNR and CNR are shown in Figs. 2b and 2c. A significant difference in SNR and CNR was observed among tested energy levels (p < 0.001). 3.2. Patient study A total of 100 patients (age 60.3 ± 17.1 years; 40 female, 60 male) were included in this analysis. VMI generation and analysis were successful for all studies. There were 25 small, 52 medium, and 23 large sized patients.
3.5. Qualitative analysis Overall trends in subjective quality for both categories paralled that of quantitative scores of noise, SNR, and CNR described previously. Vascular enhancement and overall image quality scores up to 70 keV were significantly higher (p ≤ 0.01) than that of conventional images, whereas vascular enhancement and overall image quality scores for energies 80–200 keV were significantly less than (p < 0.001) than that of conventional images (Fig. 6, Table 5). A significant difference among energies was noted both in vascular enhancement and overall image quality (p < 0.001).
Table 1 Phantom studies, with mean noise, CNR and SNR at different energy levels obtained as an average of 10 scans of the phantom. Noise Conventional 120 kVp VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
4.6 5.3 4.3 3.9 3.6 3.5 3.6 3.5 3.5 3.5 3.5 3.5
± ± ± ± ± ± ± ± ± ± ± ±
0.2 0.2 0.2 0.3 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
SNR
CNR
12.0 ± 0.4 5.3 ± 0.3 4.3 ± 0.3 11.9 ± 0.8 17.1 ± 1.2 20.4 ± 1.4 23.8 ± 1.7 25.3 ± 1.8 26.4 ± 1.9 27.2 ± 1.9 27.1 ± 2.0 27.9 ± 2.0
2.4 2.3 2.7 2.9 3.0 3.1 3.0 3.0 3.0 3.0 3.0 3.0
± ± ± ± ± ± ± ± ± ± ± ±
0.1 0.2 0.2 0.2 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
4. Discussion Our study is the first to demonstrate the low levels of noise of VMI obtained from the novel detector-based spectral CT scan across the energy levels from 40 to 200 keV, both in phantom and patient experiments. In the patient cohort, an increase in CNR and SNR at lower monoenergies was observed, due to increased contrast between the 120
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 2. Charts showing Noise (Fig. 2a), CNR (Fig. 2b) and SNR (Fig. 2c) of the phantom in conventional images as well as VMI from 40 to 200 keV. The curve for the VMI is shown in blue whereas the 120 kVp conventional image is plotted in red.
amplification of the so called anti-correlated noise. As a result of image decomposition, each basis demonstrates different noise patterns that are anti-correlated i.e, when noise is high in one basis image (e.g photoelectric image), it will be lower in the other basis image (i.e. Compton scatter image) [27–29]. When these basis images are weighted equally (i.e. approximately 70 keV), there is nearly exact “anti-correlation” and thus noise is minimized. However, when there is unequal weighting of basis images, as at low and high monoenergies,
tissue of interest and the fat at lower energies, while the increase in noise is less. In the phantom experiments, this trend was not observed since the contrast between the plastic phantom target and background does not increase at lower monoenergies, and remained largely constant. Therefore, the noise variation across monoenergies are reflected in the measured CNR and SNR. VMI are constructed by weighted linear combination of photoelectric and Compton scatter images and unequal weighting result in an 121
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 3. Axial CT scan of the abdomen at the level of the kidneys with the conventional 120 kVp image shown on the top left panel and the VMI from 40 to 200 keV shown in the following panels. The contrast is higher in the lower-energy levels of VMI and the contrast is lower at higher-energy levels of VMI while the noise is relatively constant across all the energy levels of VMI.
Table 2 Noise levels in patients at conventional and VMI (Means, with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Liver
Portal vein
Pancreas
Spleen
Aorta
Muscle
Kidney
Bone
15.3 (5.6) 13.2 (4.6) 12.6 (4.4) 12.1 (4.4) 12.0 (4.4) 11.9 (4.3) 11.8 (4.3) 11.7 (4.3) 11.67 (4.3) 11.67 (4.3) 11.6 (4.3) 11.6 (4.3)
18.2 (6.7) 17.5 (8.1) 16.0 (6.3) 15.3 (5.7) 15.0 (5.5) 15.0 (6.0) 14.5 (5.3) 14.4 (5.3) 14.4 (5.4) 14.4 (5.4) 14.3 (5.4) 14.34 (5.4)
18.6 17.6 16.2 15.5 15.1 14.8 14.6 14.6 14.5 14.5 14.5 14.4
14.7 14.2 12.8 12.1 11.7 11.4 11.2 11.0 11.1 11.1 11.1 11.0
20.6 18.0 17.2 16.7 16.5 16.4 16.2 16.1 16.1 16.1 16.1 16.1
14.9 13.0 12.4 12.1 11.9 11.8 11.7 11.5 11.7 11.6 11.6 11.6
18.3 21.4 18.0 16.1 15.1 14.5 14.0 13.8 13.6 13.6 13.5 13.6
32.2 48.9 38.2 32.2 28.7 26.8 24.7 23.7 23.2 23.0 22.8 23.7
(6.19) (7.1) (5.9) (5.5) (5.3) (5.2) (5.1) (5.2) (5.1) (5.1) (5.1) (5.1)
(5.6) (8.1) (5.9) (4.9) (4.5) (4.4) (4.3) (4.3) (4.4) (4.3) (4.3) (4.4)
(7.0) (7.8) (6.4) (5.9) (5.7) (5.6) (5.6) (5.5) (5.6) (5.6) (5.6) (5.6)
(5.5) (4.7) (4.4) (4.3) (4.3) (4.2) (4.2) (4.0) (4.2) (4.2) (4.2) (4.2)
(6.1) (12.9) (8.4) (6.3) (5.4) (5.2) (5.0) (5.0) (5.1) (5.0) (5.04) (5.0)
Fat (11.8) (31.9) (21.7) (16.1) (13.0) (11.3) (9.7) (9.0) (8.8) (8.5) (8.4) (12.)
14.3 12.9 11.8 11.2 10.8 10.6 10.4 10.3 10.4 10.2 10.2 10.2
(4.8) (4.7) (4.0) (3.6) (3.5) (3.4) (3.3) (3.2) (3.4) (3.2) (3.2) (3.2)
correlated noise reduction algorithm in the reconstruction process [27,36,37]. Total variation minimization approached has been shown to be efficient for de-noising CT images [36,37]. A modified total variation minimization approach has been recently introduced and has been shown to be effective in reducing anti-correlated noise in basis images from detector based spectral systems [37]. A direct benefit of the reduction of anti-correlated noise is the reduction of the noise variation across the energies of VMI, especially at low and high keVs, where the anti-correlated noise would lead to amplified image noise. This low noise across the spectrum will potentially help in exploiting the full benefits of VMI, including improved contrast signal, improved lesion conspicuity and decreased artifact. The low noise also results in increased SNR and CNR as shown in our study. There are some limitations in our study. This is a single center study with a retrospective design. The patient cohort and CT acquisition protocols were heterogeneous; however, they reflect the routine clinical practice. In this study, we did not compare the noise, SNR and CNR of SDCT VMI with VMI obtained from other vendors. Data on this is however available in the literature as discussed above. Quantitative measurements were made only once at each level. Qualitative analysis was focused on vascular enhancement and overall image quality with a vascular perspective. We did not evaluate the image quality specific to
the noise of the generated image is higher, particularly at low energies as demonstrated in earlier studies on dual source and rapid kVp switching scanners [17,18]. Several techniques have been developed to reduce the noise in VMI. In the dual-source technology, a second-generation monoenergetic algorithm (monoenergetic plus) has been developed. This is a spatial frequency-split technique that combines the high contrast information from low energy VMI (i.e 40 keV) with the low noise of medium energy levels (i.e 70 keV) [22,23]. This has been shown to improve the SNR and CNR at low energy levels compared to first generation monoenergetic algorithm [22,23] and has been used to improve vascular contrast and lesion conspicuity in several regions including abdomen, chest, head and neck [30–34]. An energy domain noise reduction algorithm has also been used in phantom studies to reduce the image noise up to 59% and improve the CNR by 64% by exploiting information redundancies in the energy domain [35]. These techniques shift the maximal CNR from 70 to 80 keV to 40–50 keV, which is comparable to or higher than that of single energy images at the optimal tube potential [31,35]. In the dual layer SDCT system, due to perfectly registered spatial data, basis decomposition is performed in the projection domain. The anti-correlated noise is accounted for and reduced by using an anti122
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 4. Noise, SNR, CNR of the liver in the conventional images and VMI from 40 to 200 keV.
Table 3 SNR in patients at conventional and VMI (Means, with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Liver
Portal vein
Pancreas
Spleen
Aorta
Muscle
Kidney
Bone
Fat
5.8 (3.1) 13.4 (8.9) 10.6 (6.5) 8.7 (4.9) 7.5 (4.0) 6.7 (3.4) 5.8 (2.7) 5.4 (2.4) 5.2 (2.2) 5.0 (2.1) 4.9 (2.1) 5.3 (4.9)
6.8 (4.1) 21.5 (15.0) 15.4 (10.3) 11.2 (7.3) 9.6 (9.4) 7.1 (4.2) 5.3 (2.9) 4.3 (2.2) 4.4 (5.4) 3.5 (1.7) 3.3 (1.6) 3.0 (1.4)
4.9 (5.1) 14.0 (9.6) 10.3 (6.5) 7.8 (4.7) 6.3 (3.6) 5.2 (2.9) 4.0 (2.1) 3.4 (1.8) 3.0 (1.6) 2.8 (1.5) 2.7 (1.4) 2.6 (1.4)
7.5 (4.4) 20.7 (13.5) 15.9 (10.0) 12.2 (7.5) 9.9 (5.9) 8.4 (4.7) 6.7 (3.4) 5.8 (2.7) 5.3 (2.4) 5.0 (2.1) 4.8 (2.0) 4.7 (1.9)
9.5 (5.6) 34.9 (22.9) 24.6 (15.6) 17.4 (10.8) 13.0 (7.7) 10.1 (5.8) 7.1 (3.8) 5.3 (2.6) 4.5 (2.1) 3.9 (1.8) 3.6 (1.6) 3.3 (1.4)
3.6 6.8 6.4 5.0 5.1 4.3 4.1 3.8 3.7 3.6 3.6 3.6
7.6 (4.0) 22.7 (12.7) 16.8 (9.0) 12.8 (6.5) 10.1 (4.9) 7.8 (4.5) 6.3 (4.3) 4.7 (2.1) 3.9 (2.7) 3.6 (1.6) 3.4 (1.5) 3.4 (3.1)
7.4 (3.6) 13.8 (8.2) 12.0 (6.9) 10.0 (5.2) 8.9 (4.5) 8.1 (4.1) 7.0 (3.7) 6.5 (3.4) 6.0 (3.4) 5.9 (3.3) 5.7 (3.3) 5.5 (3.3)
−8.0 (3.0) 14.1 (5.23) 12.3 (4.4) 10.9 (4.2) 10.2 (3.6) −9.7 (3.3) −9.0 (3.1) −8.6 (3.4) −8.4 (2.8) −8.4 (2.8) −8.4 (2.8) −8.3 (2.8)
(1.6) (3.5) (7.7) (2.3) (5.5) (1.8) (2.3) (1.6) (1.6) (1.5) (1.5) (1.5)
Table 4 CNR in patients at conventional and VMI (Means, with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Liver
Portal vein
Pancreas
Spleen
Aorta
Muscle
Kidney
Bone
13.2 (5.1) 26.7 (10.7) 21.9 (8.4) 18.6 (7.0) 16.67 (6.0) 15.4 (5.3) 13.8 (4.6) 13.0 (4.6) 12.7 (4.2) 12.5 (4.1) 12.3 (4.0) 12.6 (6.0)
13.1 (5.4) 32.2 (16.7) 24.5 (11.7) 19.2 (8.7) 17.0 (10.4) 14.0 (6.0) 11.7 (4.2) 10.4 (3.9) 10.6 (6.2) 9.5 (3.1) 9.3 (3.1) 9.0 (3.0)
11.0 (4.3) 24.4 (12.1) 19.3 (8.6) 15.7 (6.7) 13.6 (5.4) 12.1 (4.5) 10.4 (3.4) 9.5 (3.6) 9.1 (3.1) 8.8 (3.0) 8.6 (2.9) 8.5 (2.9)
15.23 (6.3) 34.0 (16.1) 27.4 (12.2) 22.3 (9.9) 19.4 (8.1) 17.4 (6.9) 15.0 (5.1) 13.9 (5.2) 13.3 (4.6) 12.9 (4.4) 12.6 (4.3) 12.5 (4.1)
15.0 (6.7) 44.9 (23.9) 32.9 (16.5) 24.6 (11.9) 19.6 (8.8) 16.3 (6.8) 12.8 (4.7) 10.8 (3.9) 9.9 (3.3) 9.2 (3.0) 8.9 (2.8) 8.6 (2.7)
11.2 20.5 18.0 15.0 14.3 13.0 12.1 11.4 11.3 11.1 11.0 10.9
13.8 (6.0) 32.2 (15.3) 25.4 (11.2) 20.6 (8.7) 17.5 (6.9) 15.0 (5.9) 13.1 (5.8) 11.2 (4.3) 10.5 (4.1) 10.2 (3.8) 9.9 (3.6) 9.9 (4.6)
11.0 (4.3) 18.3 (9.7) 16.3 (8.0) 14.2 (6.1) 13.0 (5.3) 12.1 (4.8) 10.9 (4.3) 10.3 (4.2) 9.9 (4.0) 9.7 (4.0) 9.6 (3.9) 9.3 (3.9)
(3.9) (6.6) (9.2) (5.2) (6.7) (4.1) (4.1) (3.9) (3.6) (3.5) (3.5) (3.5)
5. Conclusion
each organ, since this is highly variable and not the focus of this study. We did not evaluate the impact of the noise in specific clinical scenarios, such as contrast improvement, lesion conspicuity or artifact reduction, and in specific patient populations and pathologies. Future work will concentrate on evaluation of this detector based scanner platform to produce qualitatively improved images aiding in evaluation of these clinical scenarios.
VMI obtained from the novel detector based spectral CT scanner has low noise across the entire spectrum of energies and is comparable to conventional images. There are significant SNR and CNR improvements compared to conventional polyenergetic images. The low noise of VMI in SDCT at various energy levels makes the VMI usable at all energy
123
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Fig. 5. Noise, CNR, SNR of the liver at different body sizes (A) and in different CT protocols (B). Body sizes were small, medium and large. CT protocols were routine CT, CT angiography and TAVI.
Fig. 6. Chart showing subjective assessment scores, ie vascular enhancement (blue) and overall image quality (grey) for the conventional (conv) images as well as VMI from 40 to 200 keV.
124
European Journal of Radiology 98 (2018) 118–125
K. Kalisz et al.
Table 5 Image quality scores for conventional and VMI (means with standard deviations in parentheses).
Conventional VMI 40 keV VMI 50 keV VMI 60 keV VMI 70 keV VMI 80 keV VMI 100 keV VMI 120 keV VMI 140 keV VMI 160 keV VMI 180 keV VMI 200 keV
Vascular enhancement
Overall image quality
2.9 4.9 4.6 3.8 3.0 2.6 1.8 1.4 1.2 1.1 1.0 1.0
2.8 3.9 4.3 3.7 3.2 2.6 1.8 1.4 1.2 1.1 1.0 1.0
(1.4) (0.6) (0.8) (1.2) (1.3) (1.3) (1.0) (0.7) (0.4) (0.3) (0.1) (0.1)
[14] K.M. Brown, S. Zabic, T. Koeher, Comparison of ML iterative reconstruction and TVminimization for noise reduction in CT images, Proc. Fully 3D (2011) 443–446. [15] R. Alvarez, E. Seppi, A comparison of noise and dose in conventional and energy selective computed tomography, IEEE Trans. Nucl. Sci. 26 (1979) 2853–2856. [16] W.A. Kalender, P. Deak, M. Kellermeier, M. van Straten, S.V. Vollmar, pplication and patient size dependent optimization of x-ray spectra for CT, Med. Phys. 36 (2009) 993–1007. [17] L. Yu, J.A. Christner, S. Leng, J. Wang, J.G. Fletcher, C.H. McCollough, Virtual monochromatic imaging in dual-source dual-energy CT: radiation dose and image quality, Med. Phys. 38 (2011) 6371–6379. [18] K. Matsumoto, M. Jinzaki, Y. Tanami, A. Ueno, M. Yamada, S. Kuribayashi, Virtual monochromatic spectral imaging with fast kilovoltage switching: improved image quality as compared with that obtained with conventional 120-kVp CT, Radiology 259 (201) (2016) 257–262. [19] M.C.B. Godoy, S.L. Heller, D.P. Naidich, B. Assadourian, C. Leidecker, B. Schmidt, I. Vlahos, Dual-energy MDCT: comparison of pulmonary artery enhancement on dedicated CT pulmonary angiography, routine and low contrast volume studies, Eur. J. Radiol. 79 (2011) e11–e17. [20] D.F. Pinho, N.M. Kulkarni, A. Krishnaraj, S.P. Kalva, D.V. Sahani, Initial experience with single-source dual-energy CT abdominal angiography and comparison with single-energy CT angiography: image quality, enhancement, diagnosis and radiation dose, Eur. Radiol. 23 (2013) 351–359. [21] D. Hu, T. Yu, X. Duan, Y. Peng, R. Zhai, Determination of the optimal energy level in spectral CT imaging for displaying abdominal vessels in pediatric patients, Eur. J. Radiol. 83 (2014) 589–594. [22] K.L. Grant, T.G. Flohr, B. Krauss, et al., Assessment of an advanced image-based technique to calculate virtual monoenergetic computed tomographic images from a dual-energy examination to improve contrast-to-noise ratio in examinations using iodinated contrast media, Invest. Radiol. 49 (2014) 586–592. [23] C. Schabel, M. Bongers, M. Sedlmair, et al., Assessment of the hepatic veins in poor contrast conditions using dual energy CT: Evaluation of a novel monoenergetic extrapolation software algorithm, Rofo 186 (2014) 591–597. [24] R.H. Wellenberg, M.F. Boomsma, J.A. van Osch, et al., Quantifying metal artifact reduction using virtual monochromatic dual-layer detector spectral CT imaging in unilateral and bilateral total hip prostheses, Eur. J. Radiol. 88 (2017) 61–70. [25] T. Hicthethier, B. Baebler, J.R. Kroeger, et al., Monoenergetic reconstructions for imaging of coronary artery stents using spectral detector CT: in-vitro experience and comparison to conventional images, J. Cardiovasc. Comput. Tomogr. 11 (2017) 33–39. [26] N. Rassouli, H. Chalian, P. Rajiah, A. Dhanantwari, L. Landeras, Assessment of 70keV virtual monoenergetic spectral images in abdominal imaging: a comparison study to conventional polychromatic 120-kVp images, Abdom. Radiol. (NY) (April 18) (2017), http://dx.doi.org/10.1007/s00261-017-1151-2 (PMID:28421243). [27] R.E. Alvarez, A. Macovski, Energy-selective reconstructions in X-ray computerised tomography, Phys. Med. Biol. 21 (1976) 733. [28] W.A. Kalender, E. Klotz, L. Kostaridou, An algorithm for noise suppression in dual energy CT material density images, IEEE Trans. Med. Imaging 7 (1988) 218–224. [29] J.T. Dobbins, J.R. Wells, Correlated-polarity noise reduction: feasibility of a new statistical approach to reduce image noise, Proc Physics of Medical Imaging, Vol 7961 of Proceedings of SPIE (2011). [30] K.L. Grant, T.G. Flohr, B. Krauss, et al., Assessment of an advanced image-based technique to calculate virtual monoenergetic computed tomographic images from a dual-energy examination to improve contrast-to-noise ratio in examinations using iodinated contrast media, Invest. Radiol. 49 (2014) 586–592. [31] D. Marin, J.C. Ramirez-Giraldo, S. Gupta, et al., Effect of a noise-optimized secondgeneration monoenergetic algorithm on image noise and conspicuity of hypervascular liver tumors: an in vitro and in vivo study, AJR Am. J. Roentgenol. 206 (2016) 1222–1232. [32] M.H. Albrecht, J.E. Scholtz, K. Husers, et al., Advanced image based virtual monoenergetic dual-energy CT angiography of the abdomen: optimization of kiloelectron volt settings to improve image contrast, Eur. Radiol. 26 (2016) 1863–1870. [33] A. Meier, M. Wurning, L. Desboilles, et al., Advanced virtual monoenergetic images: improving the contrast of dual-energy CT pulmonary angiography, Clin. Radiol. 70 (2015) 1244–1251. [34] M.H. Albrecht, J.E. Scholtz, J. Kraft, et al., Assessment of an advanced monoenergetic reconstruction technique in dual-energy computed tomography of head and neck cancer, Eur. Radiol. 25 (2015) 2493–2501. [35] S. Leng, L. Yu, J.G. Fletcher, C.H. McCollough, Maximizing iodine contrast-to-noise ratios in abdominal CT imaging through use of energy domain noise reduction and virtual monoenergetic dual-energy CT, Radiology 276 (2015) 562–570. [36] S. Zabic, Acceleration of ML iterative algorithms for CT by the use of fast start images, Proceedings of SPIE 8313, Medical Imaging (2012) (Physics of Medical Imaging, 831339). [37] K.M. Brown, S. Zabic, S. Shechter, Impact of spectral separation in dual-energy CT with anti-correlated statistical reconstruction, Proc. Fully 3D (2015) 491–494.
(1.1) (0.6) (0.8) (1.0) (1.8) (1.1) (1.0) (0.7) (0.4) (0.4) (0.1) (0.1)
levels, particularly at low energies for enhancing vascular contrast or improving lesion conspicuity and at higher energies for decreasing artifacts. Conflict of interest
• Amar Dhanantwari is an employee of Philips healthcare • Prabhakar Rajiah has received in the past (more than 2 years ago), honoraria from Philips Healthcare. • The study was funded by institutional research grant from Philips healthcare.
References [1] C. McCollough, S. Leng, L. Yu, J.G. Fletcher, Dual- and multi-energy computed tomography: principles, technical approaches, and clinical applications, Radiology 276 (2015) 637–653. [2] T.J. Vrtiska, N. Takahashi, J.G. Fletcher, R.P. Hartman, L. Yu, A. Kawashima, Genitourinary applications of dual-energy CT, AJR Am. J. Roentgenol. 194 (2010) 1434–1442. [3] G. Girish, K.N. Glazebrook, J.A. Jacobson, Advanced imaging in gout, AJR Am. J. Roentgenol. 201 (2013) 515–525. [4] J. Paul, T.J. Vogl, E.C. Mbalisike, Oncological applications of dual-energy computed tomography imaging, J. Comput. Assist. Tomogr. 38 (2014) 834–842. [5] I. Vlahos, M.C.B. Godoy, D.P. Naidich, Dual-energy computed tomography imaging of the aorta, J. Thorac. Imaging 25 (2010) 289–300. [6] A.A. Postma, P.A.M. Hofman, A.A.R. Stadler, R.J. van Oostenbrugge, M.P.M. Tijssen, J.E. Wildberger, Dual-energy CT of the brain and intracranial vessels, AJR Am. J. Roentgenol. 199 (2012) S26–S33. [7] I. Danad, Z.A. Fayad, M.J. Willemink, J.K. Min, New applications of cardiac computed tomography: dual-energy spectral, and molecular CT imaging, JACC Cardiovasc. Imaging 8 (2015) 710–723. [8] L.-J. Zhang, J. Peng, S.-Y. Wu, et al., Liver virtual non-enhanced CT with dualsource: dual-energy CT: a preliminary study, Eur. Radiol. 20 (2010) 2257–2264. [9] E. Pessis, R. Campagna, J.-M. Sverzut, et al., Virtual monochromatic spectral imaging with fast kilovoltage switching: reduction of metal artifacts at CT, Radiogr. Rev. Publ. Radiol. Soc. N. Am. Inc. 33 (2013) 573–583. [10] R. Yuan, W.P. Shuman, J.P. Earls, et al., Reduced iodine load at CT pulmonary angiography with dual-energy monochromatic imaging: comparison with standard CT pulmonary angiography–a prospective randomized trial, Radiology 262 (2012) 290–297. [11] L. Yu, S. Leng, C.H. McCollough, Dual energy CT-based monochromatic imaging, AJR Am. J. Roentgenol. 199 (Suppl. 5) (2012) S9. [12] W.P. Shuman, D.E. Green, J.M. Busey, et al., Dual-energy liver CT: effect of monochromatic imaging on lesion detection conspicuity, and contrast-to-noise ratio of hypervascular lesions on late arterial phase, AJR Am. J. Roentgenol. 203 (2014) 601–606. [13] F. Bamberg, A. Dierks, K. Nikolaou, M.F. Reiser, C.R. Becker, T.R.C. Johnson, Metal artifact reduction by dual energy computed tomography using monoenergetic extrapolation, Eur. Radiol. 21 (2011) 1424–1429.
125