- Email: [email protected]
Image quality evaluation of dual-layer spectral detector CT of the chest and comparison with conventional CT imaging
European Journal of Radiology 93 (2017) 52–58
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Research papers
Image quality evaluation of dual-layer spectral detector CT of the chest and comparison with conventional CT imaging
MARK
⁎
Jonas Doerner , Myriam Hauger, Tilman Hickethier, Jonathan Byrtus, Christian Wybranski, Nils Große Hokamp, David Maintz, Stefan Haneder Department of Diagnostic and Interventional Radiology, University Hospital of Cologne, Cologne, Germany
A R T I C L E I N F O
A B S T R A C T
Keywords: Dual-layer detector CT Spectral detector CT Virtual mono-energetic imaging Chest
Objectives: To evaluate image quality parameters of virtual mono-energetic (MonoE) and conventional (CR) imaging derived from a dual-layer spectral detector CT (DLCT) in oncological follow-up venous phase imaging of the chest and comparison with conventional multi-detector CT (CRMDCT) imaging. Materials and methods: A total of 55 patients who had oncologic staging with conventional CT and DLCT of the chest in venous phase were included in this study. Established image quality parameters were derived from all datasets in defined thoracic landmarks. Attenuation, image noise, and signal-/contrast- to noise ratios (SNR, CNR) were compared between CRDLCT and MonoE as well as CRMDCT imaging. Two readers performed subjective image analysis. Results: CRMDCT showed significant lower attenuation values compared to CRDLCT and MonoE at 40–70 keV (p ≤ 0.05). Moreover, MonoE at 40–70 keV revealed significantly higher attenuations values compared to CRDLCT (p < 0.001). Noise was statistically lower in CRMDCT compared with CRDLCT and MonoE at 40 keV (11.4 ± 2.3 HU vs. 12.0 ± 3.1 HU vs. 11.7 ± 5.2 HU; p < 0.001). In contrast, all MonoE levels showed significantly lower noise levels compared to CRDLCT (p < 0.001). SNR was not significantly different between CRMDCT and CRDLCT (13.5 ± 3.7 vs. 14.4 ± 5.3; p > 0.99). SNR values were significantly increased for MonoE at 40–80 keV compared to CRMDCT and CRDLCT (p < 0.001). CRDLCT and MonoE (40–70 keV) from DLCT revealed significantly higher CNR values than CRMDCT (p < 0.001). In subjective analysis, MonoE at 40 keV surpassed all other image reconstructions except for noise in MonoE at 70 keV. Conclusion: In dual-layer spectral detector CT, MonoE at low keV showed superior image quality compared to conventional images derived from the same system and may therefore be added to clinical routine imaging protocols. Whether MonoE reconstructions yield additional diagnostic information is still unknown.
1. Introduction Since its clinical introduction, dual-energy or spectral computed tomography (DECT) has evolved as a comprehensive diagnostic tool with several applications [1]. Virtual mono-energetic or mono-chromatic imaging (MonoE) derived from DECT allows to increase soft tissue contrast as well as to reduce beam hardening and scatter artifacts [2–7]. This is reflected by a superior quantitative image quality in several contrast-enhanced DECT angiography (CTA) studies using MonoE images at low kiloelectron volt (keV) compared to conventional/poly-energetic imaging [4,7,8]. These advantages are not restricted to CTA and can also be applied for venous phase imaging. A recent study showed that MonoE at low keV levels generated from venous DECT datasets improved diagnostic accuracy for the detection
⁎
of incidental pulmonary embolism in oncological follow-up staging [9]. From a technical aspect, a high-energy and a low-energy dataset of conventional x-ray spectra are required for calculation of virtual MonoE from DECT [10]. To date, different DECT approaches for the acquisition of these datasets are commercially available, based either on the generation (x-ray tube) or the detection site. In particular, there are currently four CT-tube-based physical concepts available with a high and a low mean energy x-ray spectrum obtained from: a) two consecutive rotations at different tube potentials (dual-spin), b) two independent orthogonally positioned tube-detector systems at different tube potentials (dual-source), c) splitting the output of a single x-ray source using a beam filter resulting in two partial beams with high and low mean energies (split or twin beam), and d) rapid switching of the tube potential of a single x-ray source during a single rotation (kVp
Corresponding author at: University Hospital of Cologne, Department of Diagnostic and Interventional Radiology, Kerpener Str. 62, 50937, Cologne, Germany. E-mail address: [email protected] (J. Doerner).
http://dx.doi.org/10.1016/j.ejrad.2017.05.016 Received 16 February 2017; Received in revised form 9 May 2017; Accepted 15 May 2017 0720-048X/ © 2017 Elsevier B.V. All rights reserved.
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
2.3. MDCT image acquisition and post-processing
switching) [11,12]. The only commercially available detector-based solution, referred to as dual-layer detector computed tomography or spectral detector computed tomography (DLCT), uses a single polychromatic x-ray source and detects the photons of lower energies in the surface and of higher energies in the layer below [11,13]. Unique to the DLCT is the simultaneous measurement of low and high-energy projection data at the exact same spatial and angular location, which facilitates a raw-data based dual-energy post-processing [14]. In contrast, for kVp switching, raw-data post-processing is only possible after angular and temporal interpolation and the other tube-based approaches allow only for dual-energy post-processing in the image domain [15]. Moreover, combination of the projection data from upper and lower detector layer from a DLCT acquisition always provides a true conventional image dataset in addition to the dual-energy data, which may serve as a standard of reference while tube-based DECT systems have to rely on blended images as a surrogate. The aim of this technical feasibility study was twofold: 1. to assess quantitative and qualitative image quality of MonoE in thoracic contrast enhanced DLCT in comparison to conventional reconstructions (CRDLCT) provided by the same scan and 2. to compare these images intra-individually with conventional reconstructions of a conventional multi-detector CT (MDCT) of the same vendor (CRMDCT).
All examinations were performed using a 256-slice MDCT (iCT 256, Philips Healthcare, Best, The Netherlands). Positioning, scanning direction and contrast media application were equivalent to the DLCT system. The following scanning parameters were kept constant in all scans: collimation – 128 × 0.625 mm; rotation time – 0.5 s; pitch – 0.671; tube potential – 120 kVp, matrix – 512 × 512; dose modulation type: DoseRight 3D-DOM (Philips Healthcare, Best, The Netherlands). All axial images were reconstructed (CRMDCT) with a slice thickness of 2 mm and a section increment of 1 mm using a dedicated iterative model-based reconstruction algorithm (IMR, Philips Healthcare, Best, The Netherlands) with a standard strength level of 2 and a sharp kernel (SharpPlus, Philips Healthcare, Best, The Netherlands). Mean mAs as well as the computed tomography dose index (CTDI) were recorded for the whole examination from the patient examination protocol provided by the scanners, respectively. 2.4. Objective image analysis All data sets were evaluated by placing circular regions of interests (ROIs) in the ascending aorta, descending aorta, pulmonary trunk, superior vena cava, left and right ventricular cavity, left and right atrial cavity, erector spinae muscles, subcutaneous fat, and air. All measurements were performed twice and averaged. Absolute attenuation values in Hounsfield units (HU) as well as the standard deviation (SD) were recorded. Contrast to noise ratios (CNR) were calculated using the following formula [4,7,8]:
2. Materials and methods 2.1. Study population Institutional review board approval was obtained for this retrospective study and the requirement to obtain written informed consent was waived. This retrospective study comprised 55 patients (36 males, 19 females) with a mean age of 61.6 ± 13.1 years who were referred to oncological follow-up imaging between June and October 2016 on the DLCT system. All patients had prior imaging on a MDCT system within 3–4 months before (mean days between examinations 130 ± 67). Underlying disease were malignant melanoma (n = 20), lymphoma (n = 7), pancreas cancer (n = 5), esophageal cancer (n = 5), kidney cancer (n = 5), sarcoma (n = 3), lung cancer (n = 2), and others (n = 8).
CNR = (HUlumen − HUmuscle)/image noise In which image noise was defined as the SD of fat. Signal to noise ratio (SNR) was defined as: SNR = HUlumen/SDlumen.
2.5. Qualitative image analysis CRMDCT, CRDLCT as well as MonoE at 40 and 70 keV datasets were evaluated independently by two radiologists in a blinded, randomized manner. Contrast conditions, noise, depiction of vessels within the mediastinum, and overall image quality were assessed and compared using a 5-point Likert-scale, respectively. Grading for contrast conditions was from 1 = substantially lower contrast to 5 = markedly increased contrast. Grading for noise was from 1 = excessive noise to 5 = lowest noise. Grading for depiction of vessels within the mediastinum was from 1 = almost no depiction to 5 = clear and certain depiction. Grading for overall image quality was from 1 = severely impaired image quality due to excessive image noise and/or poor conspicuity of vessel walls to 5 = best image quality with only minimal perception of image noise, no limitations in low contrast resolution, excellent attenuation of the vessel lumen and clear conspicuity of the vessel walls. Readers were explicitly allowed to adjust window settings.
2.2. DLCT image acquisition and post-processing All examinations were performed using a DLCT-scanner (IQon Spectral CT, Philips Healthcare, Best, The Netherlands). Patients were positioned supine and scanned in cranio-caudal direction during breath-hold. The clinical routine protocol for oncological comprised venous phase imaging of the chest and abdomen obtained 70 s after a bolus application of 120 ml non-ionic, iodinated contrast media (Accupaque 350 mg/ ml, GE Healthcare; Little Chalfort, UK) injected via an antecubital vein at a flow rate of 3.5 ml/s followed by a 30-ml saline chaser. For contrast media timing the bolus-tracking technique was activated in all cases. The following scanning parameters were kept constant in all scans: collimation – 64 × 0.625 mm; rotation time – 0.5 s; pitch – 0.671; tube potential – 120 kVp, matrix – 512 × 512; dose modulation type: DoseRight 3D-DOM (Philips Healthcare, Best, The Netherlands). All axial images were reconstructed with a slice thickness of 2 mm and a section increment of 1 mm using a dedicated spectral reconstruction algorithm with a strength level of 3 and a constant kernel (Spectral B, Philips Healthcare, Best, The Netherlands). In addition to the conventional reconstructions (CRDLCT), 11 MonoE datasets were reconstructed using 10 keV intervals from 40 to 120 keV as well as 160 and 200 keV. Image analysis was performed offline on a dedicated workstation (IntelliSpace Portal 6.5, Philips Healthcare, Best, The Netherlands).
2.6. Statistical analysis Statistical analysis was performed using GraphPad Prism (version 7.0b for Macintosh, GraphPad Software, La Jolla, California USA) and SPSS 21 for Mac (IBM SPSS Statistics for Macintosh, Version 21.0, Armonk, NY). Descriptive statistics are summarized as means ± SD. Normal distribution was checked by means of the Shapiro-Wilk test. The equality of variances were checked by means of Levenés test. Friedman’s test and Dunn's multiple comparison test as post hoc were performed as a non-parametric test. To compare dose levels, the paired Student’s t-test was performed. Statistical significance was defined as p ≤ 0.05. Wilcoxon signed-rank test was performed to compare the 53
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
Fig. 1. Graphs show distinct higher absolute attenuation values with low image noise values and consecutive markedly increased contrast to noise (CNR) and signal to noise ratios (SNR) in mono-energetic imaging at low energy. CRMDCT: conventional reconstruction from multi-detector CT; CRDLCT: conventional reconstruction from dual-layer spectral detector CT.
qualitative image parameters. Inter-reader agreement was calculated with quadratic weighted Cohen's kappa coefficients (κ) with values of ≥0.81 indicating excellent, 0.61–0.80 substantial, 0.41–0.60 moderate, 0.21–0.40 fair, and ≤0.20 poor agreement [16,17].
(p < 0.001). Compared to CRDLCT, all MonoE levels showed statistically significant lower noise levels (p < 0.001).
3. Results
No significant differences were found between CRMDCT and CRDLCT (13.5 ± 3.7 vs. 14.4 ± 5.3; p > 0.99). For MonoE at 40–70 keV (p < 0.001), SNR values were significantly higher compared to both conventional reconstructions, CRMDCT and CRDLCT. For MonoE greater or equal 90 keV, SNR values were significantly lower compared to CRMDCT and CRDLCT (p ≤ 0.03). Compared to CRMDCT, the relative increase was 106% for CRDLCT and 333%, 239%, 176%, 135%, and 109% for MonoE at 40, 50, 60, 70 and 80 keV, respectively.
3.3. Signal to noise ratio (SNR)
3.1. Objective analysis Attenuation: Absolute attenuation values of MonoE revealed a stepwise decrease from low to high energy levels (see Fig. 1). CRMDCT showed significantly lower attenuation values compared to CRDLCT and MonoE at 40, 50, 60, and 70 KeV (p ≤ 0.05). Contrary, MonoE greater or equal 80 keV showed significantly lower attenuation values compared to CRMDCT (p < 0.001). MonoE at 40–60 keV revealed significantly higher attenuations values compared to CRDLCT (p < 0.001). MonoE at 70 keV was comparable to CRDLCT (p = 0.79). Again, MonoE greater or equal 80 keV showed significantly lower attenuation values compared to CRDLCT (p < 0.001). Compared to CRMDCT, the relative increase was 110% for CRDLCT and 311%, 208%, and 147% for MonoE at 40, 50 and 60 keV, respectively. All values are shown in Tables 1 and 2.
3.4. Contrast to noise (CNR) CRDLCT revealed significantly higher CNR values compared to CRMDCT (11.8 ± 5.6 vs. 8.6 ± 3.6; p < 0.001). CNR values for MonoE at 40–80 keV were superior compared to CRMDCT (p < 0.001). Compared to CRDLCT, CNR values for MonoE at 40–60 keV were superior (p < 0.001). Contrary, CNR values for MonoE greater than or equal to 90 keV and 80 keV were significantly lower compared to CRMDCT and CRDLCT, respectively (p ≤ 0.005). Compared to CRMDCT, the relative increase was 138% for CRDLCT and 491%, 326%, 230%, 162%, and 115% for MonoE at 40–80 keV, respectively.
3.2. Noise CRMDCT revealed significantly lower noise levels compared to CRDLCT and MonoE at 40 keV (11.4 ± 2.3 HU vs. 12.0 ± 3.1 HU vs. 11.7 ± 5.2 HU; p < 0.001). MonoE at 50 keV showed lower, although not statistically significant, levels compared to CRMDCT (10.6 ± 3.8 HU vs. 11.4 ± 2.3 HU; p > 0.99). MonoE equal or greater 60 keV showed significantly lower noise levels than CRMDCT
3.5. Dose Compared to the MDCT, mean mAs as well as the CTDI was significantly lower on the DLCT (124.1 ± 41.2 mAs vs. 54
108.5 ± 20.5 9.5 ± 2.7 12.3 ± 4.4 7.3 ± 3.8
90 keV
148.2 ± 26.3 11.4 ± 2.3 13.5 ± 3.7 8.6 ± 3.6
93.5 ± 16.3 9.4 ± 2.6 10.6 ± 3.7 5.4 ± 2.9
<0.001 <0.001 <0.03 <0.005
<0.001 <0.001 <0.001 <0.001
p-value
461.0 ± 122.8 11.7 ± 5.2 45.0 ± 18.7 42.0 ± 21.8
<0.05 <0.001 >0.99 <0.001
100 keV
40 keV
p-value
p-value
161.7 ± 34.7 12.0 ± 3.1 14.4 ± 5.3 11.8 ± 5.6
CRDLCT
<0.001 <0.001 <0.001 <0.001
p-value
308.2 ± 78.6 10.6 ± 3.8 32.3 ± 13.0 27.9 ± 14.5
50 keV
82.9 ± 13.4 9.4 ± 2.6 9.5 ± 3.2 4.0 ± 2.4
110 keV
<0.001 <0.001 <0.001 <0.001
p-value <0.001 <0.001 <0.001 <0.001
p-value
<0.001 <0.001 <0.001 <0.001
p-value
218.3 ± 52.3 10.1 ± 3.1 23.9 ± 9.3 19.7 ± 9.8
60 keV
75.3 ± 11.4 9.3 ± 2.6 8.6 ± 2.9 3.1 ± 2.0
120 keV
<0.001 >0.99 <0.001 <0.001
p-value
59.1 ± 7.6 9.3 ± 2.6 6.8 ± 2.2 1.0 ± 1.2
160 keV
164.3 ± 36.4 9.7 ± 2.9 18.4 ± 7.0 13.9 ± 6.9
70 keV
<0.001 <0.001 <0.001 <0.001
p-value
<0.001 <0.001 <0.001 <0.001
p-value
55
461.0 ± 122.8 11.7 ± 5.2 45.0 ± 18.7 42.0 ± 21.8
40 keV
70 keV
< 0.001 164.3 ± 36.4 < 0.001 9.7 ± 2.9 < 0.001 18.4 ± 7.0 < 0.001 13.9 ± 6.9
60 keV p-value
< 0.001 218.3 ± 52.3 < 0.001 10.1 ± 3.1 < 0.001 23.9 ± 9.3 < 0.001 19.7 ± 9.8
50 keV p-value
< 0.001 308.2 ± 78.6 < 0.001 10.6 ± 3.8 < 0.001 32.3 ± 13.0 < 0.001 27.9 ± 14.5
p-value
80 keV
130.5 ± 26.9 < 0.001 9.6 ± 2.7 < 0.001 14.8 ± 5.4 0.07 9.9 ± 5.0
0.79
p-value
52.4 ± 6.3 9.3 ± 2.6 6.0 ± 1.8 0.2 ± 1.0
<0.001 <0.001 <0.001 <0.001
p-value
<0.001 <0.001 <0.02 >0.99
p-value
< 0.001
< 0.001
< 0.001
< 0.001
200 keV p-value < 0.001 52.4 ± 6.3 < 0.001 9.3 ± 2.6 < 0.001 6.0 ± 1.8 < 0.001 0.2 ± 1.0
160 keV p-value < 0.001 59.1 ± 7.6 < 0.001 9.3 ± 2.6 < 0.001 6.8 ± 2.2 < 0.001 1.0 ± 1.2
120 keV p-value < 0.001 75.3 ± 11.4 < 0.001 9.3 ± 2.6 < 0.001 8.6 ± 2.9 < 0.001 3.1 ± 2.0
110 keV p-value < 0.001 82.9 ± 13.4 < 0.001 9.4 ± 2.6 < 0.001 9.5 ± 3.2 < 0.001 4.0 ± 2.4
100 keV p-value
< 0.001 93.5 ± 16.3 < 0.001 9.4 ± 2.6 < 0.001 10.6 ± 3.7 < 0.001 5.4 ± 2.9
90 keV p-value
< 0.001 108.5 ± 20.5 < 0.001 9.5 ± 2.7 > 0.99 12.3 ± 4.4 0.001 7.3 ± 3.8
p-value
Note – Bold indicating statistical significance; SNR = signal-to-noise ratio, CNR = contrast-to-noise ratio; CR = conventional reconstruction; DLCT = dual-layer spectral detector CT.
Attenuation 161.7 ± 34.7 Noise 12.0 ± 3.1 SNR 14.4 ± 5.3 CNR 11.8 ± 5.6
CRDLCT
Table 2 Objective image analysis: comparison with conventional reconstructions from DLCT.
130.5 ± 26.9 9.6 ± 2.7 14.8 ± 5.4 9.9 ± 5.0
80 keV
200 keV
Note – Bold indicating statistical significance; SNR = signal to noise ratio, CNR = contrast to noise ratio; CR = conventional reconstruction; MDCT = multi-detector CT; DLCT = dual-layer spectral detector CT.
Attenuation Noise SNR CNR
Attenuation Noise SNR CNR
CRMDCT
Table 1 Objective image analysis: comparison with conventional reconstructions from MDCT.
J. Doerner et al.
European Journal of Radiology 93 (2017) 52–58
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
Table 3 Subjective image analysis: comparison and inter-reader agreement.
Contrast Noise Vessel depiction Quality
CRMDCT
Kappa
CRDLCT
3.1 3.6 3.3 3.3
1 0.778 1 0.968
3.1 2.9 2.9 3.0
± ± ± ±
0.7 0.5 0.6 0.7
± ± ± ±
0.8 0.5 0.5 0.6
Kappa
40 keV
0.971 1 0.923 0.964
5.0 4.1 4.9 4.9
± ± ± ±
0 0.6 0.1 0.3
Kappa
70 keV
n/a 0.930 1 0–730
3.4 4.1 3.4 3.6
± ± ± ±
0.7 0.5 0.5 0.5
Kappa
CRMDCT vs. CRDLCT
CRMDCT vs. 40 keV
CRMDCTvs. 70 keV
CRDLCT vs. 40 keV
CRDLCT vs. 70 keV
40 keV vs. 70 keV
0.969 0.958 1 0.964
1.0 0.0001 0.0001 0.009
0.0001 0.0001 0.0001 0.001
0.024 0.0001 0.728 0.008
0.0001 0.0001 0.0001 0.0001
0.0001 0.0001 0.0001 0.0001
0.0001 0.79 0.0001 0.0001
Note – Bold indicating statistical significance, CR = conventional reconstruction; MDCT = multi-detector CT; DLCT = dual-layer spectral detector CT, n/a = not applicable.
Fig. 2. Representative image series of a patient in coronal reconstructions showing all images with standard window settings except for MonoE at 40 keV, which is also shown with adjusted window settings. Note the superior delineation of vascular structures in combination with low image noise in the MonoE at 40 keV adjusted compared with CRDLCT and CRMDCT. MonoE: virtual mono-energetic imaging, CRMDCT: conventional reconstruction from multi-detector CT; CRDLCT: conventional reconstruction from dual-layer spectral detector CT.
182.9 ± 60.1 mAs; p < 0.0001 12.5 ± 4.1 mGy: p < 0.0001).
and
11.3 ± 3.8 mGy
4. Discussion
vs.
This study aimed to evaluate objective and subjective image parameters of mono-energetic imaging of the chest derived from a dual-layer spectral detector CT. Therefore, an intra-individual comparison of conventional and MonoE reconstructions as well as a comparison with conventional CT reconstructions from a MDCT was performed. We found that MonoE reconstructions at low to mid energy levels derived from the DLCT system yielded significantly higher CNR and SNR values than the conventional reconstructions of both scanners – DLCT and MDCT. This was achieved by an attenuation boost for virtual mono-energetic reconstructions at low keV levels near the kedge of iodine (33 keV). The most relevant new aspect is the substantial low image noise level for MonoE reconstructions over the complete
3.6. Subjective analysis Subjective analysis revealed that CRMDCT was statistically superior compared to CRDLCT in all categories except for contrast conditions. However, MonoE at 40 keV surpassed all other image reconstructions except for the category noise in MonoE at 70 keV. MonoE at 40 keV revealed the highest contrast with the best depiction of vascular structures within the mediastinum. Inter-reader agreements were excellent for all reconstructions. All values are shown in Table 3. Representative images are shown in Fig. 2. 56
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
subjectively rather than investigating thoracic pathologies, which should be the next step. Therefore, this study aimed to be a technical investigation of the DLCT system. Second, conventional MDCT images were reconstructed using a theoretically advanced, iterative modelbased model reconstruction algorithm rather than the iterative statistical-based algorithm used for conventional DLCT and MonoE. Third, it should be mentioned that quantitative noise measurements are not exactly physically correct in the setting of nonlinear iterative reconstruction techniques, but no better alternative is available to date.
energy spectrum, including the low keV levels, compared to conventional reconstructions of the DLCT dataset. This results contrast with previous investigations, which used distinctly different dual energy technologies to derive MonoE [7,8,17]. Our results can be mainly explained by the new detector technique. DLCT provides almost simultaneously acquired (spatially and temporally) measurements of high and low energy projection datasets in the two detector layers. This facilitates the exploitation of anti-correlated noise suppression, which is particularly possible for detector-based DECT systems where, due to the perfectly matched raw-data, the anticorrelation is not lost in spatial or temporal interpolation [18]. Additionally, DLCT allows the implementation of recent iterative reconstruction techniques, as high and low energy projection datasets are used for conventional spectral reconstruction leading to a further reduction in image noise [19,20]. From the other available DECT systems, which operate either with modulation of the x-ray tube voltage or beam hardening only the kVp switching technology allows for rawdata post-processing in the projection domain to derive MonoE datasets, although angular and temporal interpolation is required. All other available scanners generate MonoE reconstructions in the image domain, causing a significant increase of image noise at low energy reconstruction levels [15]. Recently introduced vendor specific software algorithms have been applied to overcome this issue but only worked sufficiently for mid to low keV [3,7]. A newly developed algorithm was recently introduced for third generation dual energy CTs, which rendered MonoE at low keV levels diagnostic usable and shifted the highest CNR and SNR values to the lowest keV level [8,9,17]. The CRMDCT used in this study was performed with the latest iterative model-based reconstruction algorithm (IMR, Philips Healthcare, Best, The Netherlands). Based on novel model-based optimization processes, algorithms such as IMR reduce noise levels even further than the previous generation of statistical iterative reconstruction methods [19–21]. This is reflected by significantly lower noise levels of the CRMDCT compared with the CRDLCT. In this context, it is interesting that CRDLCT outperformed conventional CRMDCT in regard to CNR and attenuation levels although the DLCT datasets were reconstructed with a model-based reconstruction algorithm that takes into account anti-correlated noise features, which is an advanced reconstruction method comparable to the iterative model-based reconstruction algorithm for conventional images but not equivalent. Therefore, this effect is also mostly attributed to the new detector technology with the inherent features described above. Nevertheless, subjective image analysis revealed that conventional CT with its IMR reconstruction algorithm was rated superior than conventional reconstructions of DLCT regarding overall image quality, vessel depiction and noise. However, MonoE at 40 keV outperformed both, CRDLCT and CRMDCT, in all assessed subjective image quality categories mostly because of the significant contrast increase accompanied with low image noise, which leads to a clearer delineation of vascular structures within the mediastinum. The imaging properties derived from the DLCT were achieved with less radiation dose compared to the MDCT. This is mainly achieved by lowering the tube current, which in turn is possible because of the noise features of the DLCT detector. As a limitation of the DLCT system it shall be noted that for reliable DECT material decomposition not only the noise is relevant, but also the spectral separation between high and low energy data. This separation is potentially higher for tube-based DECT systems where e.g. tin filtration can be used to remove low energy x-ray photons from the high energy acquisition [22]. Several limitations of this technical feasibility study have to be discussed: First, only image quality was assessed objectively and
5. Conclusion In dual-layer spectral detector CT, MonoE at low keV showed superior image quality compared to conventional images derived from the same system and may therefore be added to clinical routine imaging protocols. Whether MonoE reconstructions yield additional diagnostic information is still unknown. Conflicts of interest Myriam Hauger was supported by the Koeln Fortune Program/ Faculty of Medicine, University of Cologne, Cologne, Germany. David Maintz is on the speaker’s bureau for Philips Healthcare. For the remaining authors, no conflicts of interest were declared. References [1] T.G. Flohr, C.H. McCollough, H. Bruder, M. Petersilka, K. Gruber, C. Suss, M. Grasruck, K. Stierstorfer, B. Krauss, R. Raupach, A.N. Primak, A. Kuttner, S. Achenbach, C. Becker, A. Kopp, B.M. Ohnesorge, First performance evaluation of a dual-source CT (DSCT) system, Eur. Radiol. 16 (2) (2006) 256–268. [2] S.R. Pomerantz, S. Kamalian, D. Zhang, R. Gupta, O. Rapalino, D.V. Sahani, M.H. Lev, Virtual monochromatic reconstruction of dual-energy unenhanced head CT at 65–75 keV maximizes image quality compared with conventional polychromatic CT, Radiology 266 (1) (2013) 318–325. [3] S. Sudarski, P. Apfaltrer, J.W. Nance Jr., D. Schneider, M. Meyer, S.O. Schoenberg, C. Fink, T. Henzler, Optimization of keV-settings in abdominal and lower extremity dual-source dual-energy CT angiography determined with virtual monoenergetic imaging, Eur. J. Radiol. 82 (10) (2013) e574–e581. [4] D. Schneider, P. Apfaltrer, S. Sudarski, J.W. Nance Jr., H. Haubenreisser, C. Fink, S.O. Schoenberg, T. Henzler, Optimization of kiloelectron volt settings in cerebral and cervical dual-energy CT angiography determined with virtual monoenergetic imaging, Acad. Radiol. 21 (4) (2014) 431–436. [5] A. Meier, M. Wurnig, L. Desbiolles, S. Leschka, T. Frauenfelder, H. Alkadhi, Advanced virtual monoenergetic images: improving the contrast of dual-energy CT pulmonary angiography, Clin. Radiol. 70 (11) (2015) 1244–1251. [6] F. Bamberg, A. Dierks, K. Nikolaou, M.F. Reiser, C.R. Becker, T.R. Johnson, Metal artifact reduction by dual energy computed tomography using monoenergetic extrapolation, Eur. Radiol. 21 (7) (2011) 1424–1429. [7] P. Apfaltrer, S. Sudarski, D. Schneider, J.W. Nance Jr., H. Haubenreisser, C. Fink, S.O. Schoenberg, T. Henzler, Value of monoenergetic low-kV dual energy CT datasets for improved image quality of CT pulmonary angiography, Eur. J. Radiol. 83 (2) (2014) 322–328. [8] M.H. Albrecht, J.E. Scholtz, K. Husers, M. Beeres, A.M. Bucher, M. Kaup, S.S. Martin, S. Fischer, B. Bodelle, R.W. Bauer, T. Lehnert, T.J. Vogl, J.L. Wichmann, Advanced image-based virtual monoenergetic dual-energy CT angiography of the abdomen: optimization of kiloelectron volt settings to improve image contrast, Eur. Radiol. 26 (6) (2016) 1863–1870. [9] J. Weiss, M. Notohamiprodjo, M. Bongers, C. Schabel, S. Mangold, K. Nikolaou, F. Bamberg, A.E. Othman, Effect of noise-optimized monoenergetic postprocessing on diagnostic accuracy for detecting incidental pulmonary embolism in portalvenous phase dual-energy computed tomography, Invest. Radiol. 52 (3) (2017) 142–147. [10] R.E. Alvarez, A. Macovski, Energy-selective reconstructions in X-ray computerized tomography, Phys. Med. Biol. 21 (5) (1976) 733–744. [11] A.C. Silva, B.G. Morse, A.K. Hara, R.G. Paden, N. Hongo, W. Pavlicek, Dual-energy (spectral) CT: applications in abdominal imaging, Radiographics: Rev. Publ. Radiol. Soc. North Am. Inc. 31 (4) (2011) 1031–1046 discussion 1047-50. [12] M. Patino, A. Prochowski, M.D. Agrawal, F.J. Simeone, R. Gupta, P.F. Hahn, D.V. Sahani, Material separation using dual-energy CT: current and emerging applications, Radiographics: Rev. Publ. Radiol. Soc. North Am. Inc. 36 (4) (2016) 1087–1105.
57
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
[18] W.A. Kalender, E. Klotz, L. Kostaridou, An algorithm for noise suppression in dual energy CT material density images, IEEE Trans. Med. Imaging 7 (3) (1988) 218–224. [19] W. Chang, J.M. Lee, K. Lee, J.H. Yoon, M.H. Yu, J.K. Han, B.I. Choi, Assessment of a model-based, iterative reconstruction algorithm (MBIR) regarding image quality and dose reduction in liver computed tomography, Invest. Radiol. 48 (8) (2013) 598–606. [20] K.M. Brown, S. Zabic, T. Koehler, Acceleration of ML iterative algorithms for CT by the use of fast start images, SPIE Med. Imaging Int. Soc. Opt. Photonics (2012) 831339–831339-7. [21] D. Mehta, R. Thompson, T. Morton, A. Dhanantwari, E. Shefer, Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality, Med. Phys. Int. J. 2 (1) (2013) 147–155. [22] C. Thomas, B. Krauss, D. Ketelsen, I. Tsiflikas, A. Reimann, M. Werner, D. Schilling, J. Hennenlotter, C.D. Claussen, H.P. Schlemmer, M. Heuschmid, Differentiation of urinary calculi with dual energy CT: effect of spectral shaping by high energy tin filtration, Invest. Radiol. 45 (7) (2010) 393–398.
[13] C.H. McCollough, S. Leng, L. Yu, J.G. Fletcher, Dual- and multi-energy CT: principles, technical approaches, and clinical applications, Radiology 276 (3) (2015) 637–653. [14] F.A. Dilmanian, X.Y. Wu, E.C. Parsons, B. Ren, J. Kress, T.M. Button, L.D. Chapman, J.A. Coderre, F. Giron, D. Greenberg, D.J. Krus, Z. Liang, S. Marcovici, M.J. Petersen, C.T. Roque, M. Shleifer, D.N. Slatkin, W.C. Thomlinson, K. Yamamoto, Z. Zhong, Single-and dual-energy CT with monochromatic synchrotron x-rays, Phys. Med. Biol. 42 (2) (1997) 371–387. [15] L. Yu, S. Leng, C.H. McCollough, Dual-energy CT-based monochromatic imaging, AJR Am. J. Roentgenol. 199 (Suppl. 5) (2012) S9–S15. [16] J. Cohen, Weighted kappa: nominal scale agreement with provision for scaled disagreement or partial credit, Psychol. Bull. 70 (4) (1968) 213–220. [17] M.H. Albrecht, J. Trommer, J.L. Wichmann, J.E. Scholtz, S.S. Martin, T. Lehnert, T.J. Vogl, B. Bodelle, Comprehensive comparison of virtual monoenergetic and linearly blended reconstruction techniques in third-generation dual-source dualenergy computed tomography angiography of the thorax and abdomen, Invest. Radiol. 51 (9) (2016) 582–590.
58
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Research papers
Image quality evaluation of dual-layer spectral detector CT of the chest and comparison with conventional CT imaging
MARK
⁎
Jonas Doerner , Myriam Hauger, Tilman Hickethier, Jonathan Byrtus, Christian Wybranski, Nils Große Hokamp, David Maintz, Stefan Haneder Department of Diagnostic and Interventional Radiology, University Hospital of Cologne, Cologne, Germany
A R T I C L E I N F O
A B S T R A C T
Keywords: Dual-layer detector CT Spectral detector CT Virtual mono-energetic imaging Chest
Objectives: To evaluate image quality parameters of virtual mono-energetic (MonoE) and conventional (CR) imaging derived from a dual-layer spectral detector CT (DLCT) in oncological follow-up venous phase imaging of the chest and comparison with conventional multi-detector CT (CRMDCT) imaging. Materials and methods: A total of 55 patients who had oncologic staging with conventional CT and DLCT of the chest in venous phase were included in this study. Established image quality parameters were derived from all datasets in defined thoracic landmarks. Attenuation, image noise, and signal-/contrast- to noise ratios (SNR, CNR) were compared between CRDLCT and MonoE as well as CRMDCT imaging. Two readers performed subjective image analysis. Results: CRMDCT showed significant lower attenuation values compared to CRDLCT and MonoE at 40–70 keV (p ≤ 0.05). Moreover, MonoE at 40–70 keV revealed significantly higher attenuations values compared to CRDLCT (p < 0.001). Noise was statistically lower in CRMDCT compared with CRDLCT and MonoE at 40 keV (11.4 ± 2.3 HU vs. 12.0 ± 3.1 HU vs. 11.7 ± 5.2 HU; p < 0.001). In contrast, all MonoE levels showed significantly lower noise levels compared to CRDLCT (p < 0.001). SNR was not significantly different between CRMDCT and CRDLCT (13.5 ± 3.7 vs. 14.4 ± 5.3; p > 0.99). SNR values were significantly increased for MonoE at 40–80 keV compared to CRMDCT and CRDLCT (p < 0.001). CRDLCT and MonoE (40–70 keV) from DLCT revealed significantly higher CNR values than CRMDCT (p < 0.001). In subjective analysis, MonoE at 40 keV surpassed all other image reconstructions except for noise in MonoE at 70 keV. Conclusion: In dual-layer spectral detector CT, MonoE at low keV showed superior image quality compared to conventional images derived from the same system and may therefore be added to clinical routine imaging protocols. Whether MonoE reconstructions yield additional diagnostic information is still unknown.
1. Introduction Since its clinical introduction, dual-energy or spectral computed tomography (DECT) has evolved as a comprehensive diagnostic tool with several applications [1]. Virtual mono-energetic or mono-chromatic imaging (MonoE) derived from DECT allows to increase soft tissue contrast as well as to reduce beam hardening and scatter artifacts [2–7]. This is reflected by a superior quantitative image quality in several contrast-enhanced DECT angiography (CTA) studies using MonoE images at low kiloelectron volt (keV) compared to conventional/poly-energetic imaging [4,7,8]. These advantages are not restricted to CTA and can also be applied for venous phase imaging. A recent study showed that MonoE at low keV levels generated from venous DECT datasets improved diagnostic accuracy for the detection
⁎
of incidental pulmonary embolism in oncological follow-up staging [9]. From a technical aspect, a high-energy and a low-energy dataset of conventional x-ray spectra are required for calculation of virtual MonoE from DECT [10]. To date, different DECT approaches for the acquisition of these datasets are commercially available, based either on the generation (x-ray tube) or the detection site. In particular, there are currently four CT-tube-based physical concepts available with a high and a low mean energy x-ray spectrum obtained from: a) two consecutive rotations at different tube potentials (dual-spin), b) two independent orthogonally positioned tube-detector systems at different tube potentials (dual-source), c) splitting the output of a single x-ray source using a beam filter resulting in two partial beams with high and low mean energies (split or twin beam), and d) rapid switching of the tube potential of a single x-ray source during a single rotation (kVp
Corresponding author at: University Hospital of Cologne, Department of Diagnostic and Interventional Radiology, Kerpener Str. 62, 50937, Cologne, Germany. E-mail address: [email protected] (J. Doerner).
http://dx.doi.org/10.1016/j.ejrad.2017.05.016 Received 16 February 2017; Received in revised form 9 May 2017; Accepted 15 May 2017 0720-048X/ © 2017 Elsevier B.V. All rights reserved.
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
2.3. MDCT image acquisition and post-processing
switching) [11,12]. The only commercially available detector-based solution, referred to as dual-layer detector computed tomography or spectral detector computed tomography (DLCT), uses a single polychromatic x-ray source and detects the photons of lower energies in the surface and of higher energies in the layer below [11,13]. Unique to the DLCT is the simultaneous measurement of low and high-energy projection data at the exact same spatial and angular location, which facilitates a raw-data based dual-energy post-processing [14]. In contrast, for kVp switching, raw-data post-processing is only possible after angular and temporal interpolation and the other tube-based approaches allow only for dual-energy post-processing in the image domain [15]. Moreover, combination of the projection data from upper and lower detector layer from a DLCT acquisition always provides a true conventional image dataset in addition to the dual-energy data, which may serve as a standard of reference while tube-based DECT systems have to rely on blended images as a surrogate. The aim of this technical feasibility study was twofold: 1. to assess quantitative and qualitative image quality of MonoE in thoracic contrast enhanced DLCT in comparison to conventional reconstructions (CRDLCT) provided by the same scan and 2. to compare these images intra-individually with conventional reconstructions of a conventional multi-detector CT (MDCT) of the same vendor (CRMDCT).
All examinations were performed using a 256-slice MDCT (iCT 256, Philips Healthcare, Best, The Netherlands). Positioning, scanning direction and contrast media application were equivalent to the DLCT system. The following scanning parameters were kept constant in all scans: collimation – 128 × 0.625 mm; rotation time – 0.5 s; pitch – 0.671; tube potential – 120 kVp, matrix – 512 × 512; dose modulation type: DoseRight 3D-DOM (Philips Healthcare, Best, The Netherlands). All axial images were reconstructed (CRMDCT) with a slice thickness of 2 mm and a section increment of 1 mm using a dedicated iterative model-based reconstruction algorithm (IMR, Philips Healthcare, Best, The Netherlands) with a standard strength level of 2 and a sharp kernel (SharpPlus, Philips Healthcare, Best, The Netherlands). Mean mAs as well as the computed tomography dose index (CTDI) were recorded for the whole examination from the patient examination protocol provided by the scanners, respectively. 2.4. Objective image analysis All data sets were evaluated by placing circular regions of interests (ROIs) in the ascending aorta, descending aorta, pulmonary trunk, superior vena cava, left and right ventricular cavity, left and right atrial cavity, erector spinae muscles, subcutaneous fat, and air. All measurements were performed twice and averaged. Absolute attenuation values in Hounsfield units (HU) as well as the standard deviation (SD) were recorded. Contrast to noise ratios (CNR) were calculated using the following formula [4,7,8]:
2. Materials and methods 2.1. Study population Institutional review board approval was obtained for this retrospective study and the requirement to obtain written informed consent was waived. This retrospective study comprised 55 patients (36 males, 19 females) with a mean age of 61.6 ± 13.1 years who were referred to oncological follow-up imaging between June and October 2016 on the DLCT system. All patients had prior imaging on a MDCT system within 3–4 months before (mean days between examinations 130 ± 67). Underlying disease were malignant melanoma (n = 20), lymphoma (n = 7), pancreas cancer (n = 5), esophageal cancer (n = 5), kidney cancer (n = 5), sarcoma (n = 3), lung cancer (n = 2), and others (n = 8).
CNR = (HUlumen − HUmuscle)/image noise In which image noise was defined as the SD of fat. Signal to noise ratio (SNR) was defined as: SNR = HUlumen/SDlumen.
2.5. Qualitative image analysis CRMDCT, CRDLCT as well as MonoE at 40 and 70 keV datasets were evaluated independently by two radiologists in a blinded, randomized manner. Contrast conditions, noise, depiction of vessels within the mediastinum, and overall image quality were assessed and compared using a 5-point Likert-scale, respectively. Grading for contrast conditions was from 1 = substantially lower contrast to 5 = markedly increased contrast. Grading for noise was from 1 = excessive noise to 5 = lowest noise. Grading for depiction of vessels within the mediastinum was from 1 = almost no depiction to 5 = clear and certain depiction. Grading for overall image quality was from 1 = severely impaired image quality due to excessive image noise and/or poor conspicuity of vessel walls to 5 = best image quality with only minimal perception of image noise, no limitations in low contrast resolution, excellent attenuation of the vessel lumen and clear conspicuity of the vessel walls. Readers were explicitly allowed to adjust window settings.
2.2. DLCT image acquisition and post-processing All examinations were performed using a DLCT-scanner (IQon Spectral CT, Philips Healthcare, Best, The Netherlands). Patients were positioned supine and scanned in cranio-caudal direction during breath-hold. The clinical routine protocol for oncological comprised venous phase imaging of the chest and abdomen obtained 70 s after a bolus application of 120 ml non-ionic, iodinated contrast media (Accupaque 350 mg/ ml, GE Healthcare; Little Chalfort, UK) injected via an antecubital vein at a flow rate of 3.5 ml/s followed by a 30-ml saline chaser. For contrast media timing the bolus-tracking technique was activated in all cases. The following scanning parameters were kept constant in all scans: collimation – 64 × 0.625 mm; rotation time – 0.5 s; pitch – 0.671; tube potential – 120 kVp, matrix – 512 × 512; dose modulation type: DoseRight 3D-DOM (Philips Healthcare, Best, The Netherlands). All axial images were reconstructed with a slice thickness of 2 mm and a section increment of 1 mm using a dedicated spectral reconstruction algorithm with a strength level of 3 and a constant kernel (Spectral B, Philips Healthcare, Best, The Netherlands). In addition to the conventional reconstructions (CRDLCT), 11 MonoE datasets were reconstructed using 10 keV intervals from 40 to 120 keV as well as 160 and 200 keV. Image analysis was performed offline on a dedicated workstation (IntelliSpace Portal 6.5, Philips Healthcare, Best, The Netherlands).
2.6. Statistical analysis Statistical analysis was performed using GraphPad Prism (version 7.0b for Macintosh, GraphPad Software, La Jolla, California USA) and SPSS 21 for Mac (IBM SPSS Statistics for Macintosh, Version 21.0, Armonk, NY). Descriptive statistics are summarized as means ± SD. Normal distribution was checked by means of the Shapiro-Wilk test. The equality of variances were checked by means of Levenés test. Friedman’s test and Dunn's multiple comparison test as post hoc were performed as a non-parametric test. To compare dose levels, the paired Student’s t-test was performed. Statistical significance was defined as p ≤ 0.05. Wilcoxon signed-rank test was performed to compare the 53
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
Fig. 1. Graphs show distinct higher absolute attenuation values with low image noise values and consecutive markedly increased contrast to noise (CNR) and signal to noise ratios (SNR) in mono-energetic imaging at low energy. CRMDCT: conventional reconstruction from multi-detector CT; CRDLCT: conventional reconstruction from dual-layer spectral detector CT.
qualitative image parameters. Inter-reader agreement was calculated with quadratic weighted Cohen's kappa coefficients (κ) with values of ≥0.81 indicating excellent, 0.61–0.80 substantial, 0.41–0.60 moderate, 0.21–0.40 fair, and ≤0.20 poor agreement [16,17].
(p < 0.001). Compared to CRDLCT, all MonoE levels showed statistically significant lower noise levels (p < 0.001).
3. Results
No significant differences were found between CRMDCT and CRDLCT (13.5 ± 3.7 vs. 14.4 ± 5.3; p > 0.99). For MonoE at 40–70 keV (p < 0.001), SNR values were significantly higher compared to both conventional reconstructions, CRMDCT and CRDLCT. For MonoE greater or equal 90 keV, SNR values were significantly lower compared to CRMDCT and CRDLCT (p ≤ 0.03). Compared to CRMDCT, the relative increase was 106% for CRDLCT and 333%, 239%, 176%, 135%, and 109% for MonoE at 40, 50, 60, 70 and 80 keV, respectively.
3.3. Signal to noise ratio (SNR)
3.1. Objective analysis Attenuation: Absolute attenuation values of MonoE revealed a stepwise decrease from low to high energy levels (see Fig. 1). CRMDCT showed significantly lower attenuation values compared to CRDLCT and MonoE at 40, 50, 60, and 70 KeV (p ≤ 0.05). Contrary, MonoE greater or equal 80 keV showed significantly lower attenuation values compared to CRMDCT (p < 0.001). MonoE at 40–60 keV revealed significantly higher attenuations values compared to CRDLCT (p < 0.001). MonoE at 70 keV was comparable to CRDLCT (p = 0.79). Again, MonoE greater or equal 80 keV showed significantly lower attenuation values compared to CRDLCT (p < 0.001). Compared to CRMDCT, the relative increase was 110% for CRDLCT and 311%, 208%, and 147% for MonoE at 40, 50 and 60 keV, respectively. All values are shown in Tables 1 and 2.
3.4. Contrast to noise (CNR) CRDLCT revealed significantly higher CNR values compared to CRMDCT (11.8 ± 5.6 vs. 8.6 ± 3.6; p < 0.001). CNR values for MonoE at 40–80 keV were superior compared to CRMDCT (p < 0.001). Compared to CRDLCT, CNR values for MonoE at 40–60 keV were superior (p < 0.001). Contrary, CNR values for MonoE greater than or equal to 90 keV and 80 keV were significantly lower compared to CRMDCT and CRDLCT, respectively (p ≤ 0.005). Compared to CRMDCT, the relative increase was 138% for CRDLCT and 491%, 326%, 230%, 162%, and 115% for MonoE at 40–80 keV, respectively.
3.2. Noise CRMDCT revealed significantly lower noise levels compared to CRDLCT and MonoE at 40 keV (11.4 ± 2.3 HU vs. 12.0 ± 3.1 HU vs. 11.7 ± 5.2 HU; p < 0.001). MonoE at 50 keV showed lower, although not statistically significant, levels compared to CRMDCT (10.6 ± 3.8 HU vs. 11.4 ± 2.3 HU; p > 0.99). MonoE equal or greater 60 keV showed significantly lower noise levels than CRMDCT
3.5. Dose Compared to the MDCT, mean mAs as well as the CTDI was significantly lower on the DLCT (124.1 ± 41.2 mAs vs. 54
108.5 ± 20.5 9.5 ± 2.7 12.3 ± 4.4 7.3 ± 3.8
90 keV
148.2 ± 26.3 11.4 ± 2.3 13.5 ± 3.7 8.6 ± 3.6
93.5 ± 16.3 9.4 ± 2.6 10.6 ± 3.7 5.4 ± 2.9
<0.001 <0.001 <0.03 <0.005
<0.001 <0.001 <0.001 <0.001
p-value
461.0 ± 122.8 11.7 ± 5.2 45.0 ± 18.7 42.0 ± 21.8
<0.05 <0.001 >0.99 <0.001
100 keV
40 keV
p-value
p-value
161.7 ± 34.7 12.0 ± 3.1 14.4 ± 5.3 11.8 ± 5.6
CRDLCT
<0.001 <0.001 <0.001 <0.001
p-value
308.2 ± 78.6 10.6 ± 3.8 32.3 ± 13.0 27.9 ± 14.5
50 keV
82.9 ± 13.4 9.4 ± 2.6 9.5 ± 3.2 4.0 ± 2.4
110 keV
<0.001 <0.001 <0.001 <0.001
p-value <0.001 <0.001 <0.001 <0.001
p-value
<0.001 <0.001 <0.001 <0.001
p-value
218.3 ± 52.3 10.1 ± 3.1 23.9 ± 9.3 19.7 ± 9.8
60 keV
75.3 ± 11.4 9.3 ± 2.6 8.6 ± 2.9 3.1 ± 2.0
120 keV
<0.001 >0.99 <0.001 <0.001
p-value
59.1 ± 7.6 9.3 ± 2.6 6.8 ± 2.2 1.0 ± 1.2
160 keV
164.3 ± 36.4 9.7 ± 2.9 18.4 ± 7.0 13.9 ± 6.9
70 keV
<0.001 <0.001 <0.001 <0.001
p-value
<0.001 <0.001 <0.001 <0.001
p-value
55
461.0 ± 122.8 11.7 ± 5.2 45.0 ± 18.7 42.0 ± 21.8
40 keV
70 keV
< 0.001 164.3 ± 36.4 < 0.001 9.7 ± 2.9 < 0.001 18.4 ± 7.0 < 0.001 13.9 ± 6.9
60 keV p-value
< 0.001 218.3 ± 52.3 < 0.001 10.1 ± 3.1 < 0.001 23.9 ± 9.3 < 0.001 19.7 ± 9.8
50 keV p-value
< 0.001 308.2 ± 78.6 < 0.001 10.6 ± 3.8 < 0.001 32.3 ± 13.0 < 0.001 27.9 ± 14.5
p-value
80 keV
130.5 ± 26.9 < 0.001 9.6 ± 2.7 < 0.001 14.8 ± 5.4 0.07 9.9 ± 5.0
0.79
p-value
52.4 ± 6.3 9.3 ± 2.6 6.0 ± 1.8 0.2 ± 1.0
<0.001 <0.001 <0.001 <0.001
p-value
<0.001 <0.001 <0.02 >0.99
p-value
< 0.001
< 0.001
< 0.001
< 0.001
200 keV p-value < 0.001 52.4 ± 6.3 < 0.001 9.3 ± 2.6 < 0.001 6.0 ± 1.8 < 0.001 0.2 ± 1.0
160 keV p-value < 0.001 59.1 ± 7.6 < 0.001 9.3 ± 2.6 < 0.001 6.8 ± 2.2 < 0.001 1.0 ± 1.2
120 keV p-value < 0.001 75.3 ± 11.4 < 0.001 9.3 ± 2.6 < 0.001 8.6 ± 2.9 < 0.001 3.1 ± 2.0
110 keV p-value < 0.001 82.9 ± 13.4 < 0.001 9.4 ± 2.6 < 0.001 9.5 ± 3.2 < 0.001 4.0 ± 2.4
100 keV p-value
< 0.001 93.5 ± 16.3 < 0.001 9.4 ± 2.6 < 0.001 10.6 ± 3.7 < 0.001 5.4 ± 2.9
90 keV p-value
< 0.001 108.5 ± 20.5 < 0.001 9.5 ± 2.7 > 0.99 12.3 ± 4.4 0.001 7.3 ± 3.8
p-value
Note – Bold indicating statistical significance; SNR = signal-to-noise ratio, CNR = contrast-to-noise ratio; CR = conventional reconstruction; DLCT = dual-layer spectral detector CT.
Attenuation 161.7 ± 34.7 Noise 12.0 ± 3.1 SNR 14.4 ± 5.3 CNR 11.8 ± 5.6
CRDLCT
Table 2 Objective image analysis: comparison with conventional reconstructions from DLCT.
130.5 ± 26.9 9.6 ± 2.7 14.8 ± 5.4 9.9 ± 5.0
80 keV
200 keV
Note – Bold indicating statistical significance; SNR = signal to noise ratio, CNR = contrast to noise ratio; CR = conventional reconstruction; MDCT = multi-detector CT; DLCT = dual-layer spectral detector CT.
Attenuation Noise SNR CNR
Attenuation Noise SNR CNR
CRMDCT
Table 1 Objective image analysis: comparison with conventional reconstructions from MDCT.
J. Doerner et al.
European Journal of Radiology 93 (2017) 52–58
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
Table 3 Subjective image analysis: comparison and inter-reader agreement.
Contrast Noise Vessel depiction Quality
CRMDCT
Kappa
CRDLCT
3.1 3.6 3.3 3.3
1 0.778 1 0.968
3.1 2.9 2.9 3.0
± ± ± ±
0.7 0.5 0.6 0.7
± ± ± ±
0.8 0.5 0.5 0.6
Kappa
40 keV
0.971 1 0.923 0.964
5.0 4.1 4.9 4.9
± ± ± ±
0 0.6 0.1 0.3
Kappa
70 keV
n/a 0.930 1 0–730
3.4 4.1 3.4 3.6
± ± ± ±
0.7 0.5 0.5 0.5
Kappa
CRMDCT vs. CRDLCT
CRMDCT vs. 40 keV
CRMDCTvs. 70 keV
CRDLCT vs. 40 keV
CRDLCT vs. 70 keV
40 keV vs. 70 keV
0.969 0.958 1 0.964
1.0 0.0001 0.0001 0.009
0.0001 0.0001 0.0001 0.001
0.024 0.0001 0.728 0.008
0.0001 0.0001 0.0001 0.0001
0.0001 0.0001 0.0001 0.0001
0.0001 0.79 0.0001 0.0001
Note – Bold indicating statistical significance, CR = conventional reconstruction; MDCT = multi-detector CT; DLCT = dual-layer spectral detector CT, n/a = not applicable.
Fig. 2. Representative image series of a patient in coronal reconstructions showing all images with standard window settings except for MonoE at 40 keV, which is also shown with adjusted window settings. Note the superior delineation of vascular structures in combination with low image noise in the MonoE at 40 keV adjusted compared with CRDLCT and CRMDCT. MonoE: virtual mono-energetic imaging, CRMDCT: conventional reconstruction from multi-detector CT; CRDLCT: conventional reconstruction from dual-layer spectral detector CT.
182.9 ± 60.1 mAs; p < 0.0001 12.5 ± 4.1 mGy: p < 0.0001).
and
11.3 ± 3.8 mGy
4. Discussion
vs.
This study aimed to evaluate objective and subjective image parameters of mono-energetic imaging of the chest derived from a dual-layer spectral detector CT. Therefore, an intra-individual comparison of conventional and MonoE reconstructions as well as a comparison with conventional CT reconstructions from a MDCT was performed. We found that MonoE reconstructions at low to mid energy levels derived from the DLCT system yielded significantly higher CNR and SNR values than the conventional reconstructions of both scanners – DLCT and MDCT. This was achieved by an attenuation boost for virtual mono-energetic reconstructions at low keV levels near the kedge of iodine (33 keV). The most relevant new aspect is the substantial low image noise level for MonoE reconstructions over the complete
3.6. Subjective analysis Subjective analysis revealed that CRMDCT was statistically superior compared to CRDLCT in all categories except for contrast conditions. However, MonoE at 40 keV surpassed all other image reconstructions except for the category noise in MonoE at 70 keV. MonoE at 40 keV revealed the highest contrast with the best depiction of vascular structures within the mediastinum. Inter-reader agreements were excellent for all reconstructions. All values are shown in Table 3. Representative images are shown in Fig. 2. 56
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
subjectively rather than investigating thoracic pathologies, which should be the next step. Therefore, this study aimed to be a technical investigation of the DLCT system. Second, conventional MDCT images were reconstructed using a theoretically advanced, iterative modelbased model reconstruction algorithm rather than the iterative statistical-based algorithm used for conventional DLCT and MonoE. Third, it should be mentioned that quantitative noise measurements are not exactly physically correct in the setting of nonlinear iterative reconstruction techniques, but no better alternative is available to date.
energy spectrum, including the low keV levels, compared to conventional reconstructions of the DLCT dataset. This results contrast with previous investigations, which used distinctly different dual energy technologies to derive MonoE [7,8,17]. Our results can be mainly explained by the new detector technique. DLCT provides almost simultaneously acquired (spatially and temporally) measurements of high and low energy projection datasets in the two detector layers. This facilitates the exploitation of anti-correlated noise suppression, which is particularly possible for detector-based DECT systems where, due to the perfectly matched raw-data, the anticorrelation is not lost in spatial or temporal interpolation [18]. Additionally, DLCT allows the implementation of recent iterative reconstruction techniques, as high and low energy projection datasets are used for conventional spectral reconstruction leading to a further reduction in image noise [19,20]. From the other available DECT systems, which operate either with modulation of the x-ray tube voltage or beam hardening only the kVp switching technology allows for rawdata post-processing in the projection domain to derive MonoE datasets, although angular and temporal interpolation is required. All other available scanners generate MonoE reconstructions in the image domain, causing a significant increase of image noise at low energy reconstruction levels [15]. Recently introduced vendor specific software algorithms have been applied to overcome this issue but only worked sufficiently for mid to low keV [3,7]. A newly developed algorithm was recently introduced for third generation dual energy CTs, which rendered MonoE at low keV levels diagnostic usable and shifted the highest CNR and SNR values to the lowest keV level [8,9,17]. The CRMDCT used in this study was performed with the latest iterative model-based reconstruction algorithm (IMR, Philips Healthcare, Best, The Netherlands). Based on novel model-based optimization processes, algorithms such as IMR reduce noise levels even further than the previous generation of statistical iterative reconstruction methods [19–21]. This is reflected by significantly lower noise levels of the CRMDCT compared with the CRDLCT. In this context, it is interesting that CRDLCT outperformed conventional CRMDCT in regard to CNR and attenuation levels although the DLCT datasets were reconstructed with a model-based reconstruction algorithm that takes into account anti-correlated noise features, which is an advanced reconstruction method comparable to the iterative model-based reconstruction algorithm for conventional images but not equivalent. Therefore, this effect is also mostly attributed to the new detector technology with the inherent features described above. Nevertheless, subjective image analysis revealed that conventional CT with its IMR reconstruction algorithm was rated superior than conventional reconstructions of DLCT regarding overall image quality, vessel depiction and noise. However, MonoE at 40 keV outperformed both, CRDLCT and CRMDCT, in all assessed subjective image quality categories mostly because of the significant contrast increase accompanied with low image noise, which leads to a clearer delineation of vascular structures within the mediastinum. The imaging properties derived from the DLCT were achieved with less radiation dose compared to the MDCT. This is mainly achieved by lowering the tube current, which in turn is possible because of the noise features of the DLCT detector. As a limitation of the DLCT system it shall be noted that for reliable DECT material decomposition not only the noise is relevant, but also the spectral separation between high and low energy data. This separation is potentially higher for tube-based DECT systems where e.g. tin filtration can be used to remove low energy x-ray photons from the high energy acquisition [22]. Several limitations of this technical feasibility study have to be discussed: First, only image quality was assessed objectively and
5. Conclusion In dual-layer spectral detector CT, MonoE at low keV showed superior image quality compared to conventional images derived from the same system and may therefore be added to clinical routine imaging protocols. Whether MonoE reconstructions yield additional diagnostic information is still unknown. Conflicts of interest Myriam Hauger was supported by the Koeln Fortune Program/ Faculty of Medicine, University of Cologne, Cologne, Germany. David Maintz is on the speaker’s bureau for Philips Healthcare. For the remaining authors, no conflicts of interest were declared. References [1] T.G. Flohr, C.H. McCollough, H. Bruder, M. Petersilka, K. Gruber, C. Suss, M. Grasruck, K. Stierstorfer, B. Krauss, R. Raupach, A.N. Primak, A. Kuttner, S. Achenbach, C. Becker, A. Kopp, B.M. Ohnesorge, First performance evaluation of a dual-source CT (DSCT) system, Eur. Radiol. 16 (2) (2006) 256–268. [2] S.R. Pomerantz, S. Kamalian, D. Zhang, R. Gupta, O. Rapalino, D.V. Sahani, M.H. Lev, Virtual monochromatic reconstruction of dual-energy unenhanced head CT at 65–75 keV maximizes image quality compared with conventional polychromatic CT, Radiology 266 (1) (2013) 318–325. [3] S. Sudarski, P. Apfaltrer, J.W. Nance Jr., D. Schneider, M. Meyer, S.O. Schoenberg, C. Fink, T. Henzler, Optimization of keV-settings in abdominal and lower extremity dual-source dual-energy CT angiography determined with virtual monoenergetic imaging, Eur. J. Radiol. 82 (10) (2013) e574–e581. [4] D. Schneider, P. Apfaltrer, S. Sudarski, J.W. Nance Jr., H. Haubenreisser, C. Fink, S.O. Schoenberg, T. Henzler, Optimization of kiloelectron volt settings in cerebral and cervical dual-energy CT angiography determined with virtual monoenergetic imaging, Acad. Radiol. 21 (4) (2014) 431–436. [5] A. Meier, M. Wurnig, L. Desbiolles, S. Leschka, T. Frauenfelder, H. Alkadhi, Advanced virtual monoenergetic images: improving the contrast of dual-energy CT pulmonary angiography, Clin. Radiol. 70 (11) (2015) 1244–1251. [6] F. Bamberg, A. Dierks, K. Nikolaou, M.F. Reiser, C.R. Becker, T.R. Johnson, Metal artifact reduction by dual energy computed tomography using monoenergetic extrapolation, Eur. Radiol. 21 (7) (2011) 1424–1429. [7] P. Apfaltrer, S. Sudarski, D. Schneider, J.W. Nance Jr., H. Haubenreisser, C. Fink, S.O. Schoenberg, T. Henzler, Value of monoenergetic low-kV dual energy CT datasets for improved image quality of CT pulmonary angiography, Eur. J. Radiol. 83 (2) (2014) 322–328. [8] M.H. Albrecht, J.E. Scholtz, K. Husers, M. Beeres, A.M. Bucher, M. Kaup, S.S. Martin, S. Fischer, B. Bodelle, R.W. Bauer, T. Lehnert, T.J. Vogl, J.L. Wichmann, Advanced image-based virtual monoenergetic dual-energy CT angiography of the abdomen: optimization of kiloelectron volt settings to improve image contrast, Eur. Radiol. 26 (6) (2016) 1863–1870. [9] J. Weiss, M. Notohamiprodjo, M. Bongers, C. Schabel, S. Mangold, K. Nikolaou, F. Bamberg, A.E. Othman, Effect of noise-optimized monoenergetic postprocessing on diagnostic accuracy for detecting incidental pulmonary embolism in portalvenous phase dual-energy computed tomography, Invest. Radiol. 52 (3) (2017) 142–147. [10] R.E. Alvarez, A. Macovski, Energy-selective reconstructions in X-ray computerized tomography, Phys. Med. Biol. 21 (5) (1976) 733–744. [11] A.C. Silva, B.G. Morse, A.K. Hara, R.G. Paden, N. Hongo, W. Pavlicek, Dual-energy (spectral) CT: applications in abdominal imaging, Radiographics: Rev. Publ. Radiol. Soc. North Am. Inc. 31 (4) (2011) 1031–1046 discussion 1047-50. [12] M. Patino, A. Prochowski, M.D. Agrawal, F.J. Simeone, R. Gupta, P.F. Hahn, D.V. Sahani, Material separation using dual-energy CT: current and emerging applications, Radiographics: Rev. Publ. Radiol. Soc. North Am. Inc. 36 (4) (2016) 1087–1105.
57
European Journal of Radiology 93 (2017) 52–58
J. Doerner et al.
[18] W.A. Kalender, E. Klotz, L. Kostaridou, An algorithm for noise suppression in dual energy CT material density images, IEEE Trans. Med. Imaging 7 (3) (1988) 218–224. [19] W. Chang, J.M. Lee, K. Lee, J.H. Yoon, M.H. Yu, J.K. Han, B.I. Choi, Assessment of a model-based, iterative reconstruction algorithm (MBIR) regarding image quality and dose reduction in liver computed tomography, Invest. Radiol. 48 (8) (2013) 598–606. [20] K.M. Brown, S. Zabic, T. Koehler, Acceleration of ML iterative algorithms for CT by the use of fast start images, SPIE Med. Imaging Int. Soc. Opt. Photonics (2012) 831339–831339-7. [21] D. Mehta, R. Thompson, T. Morton, A. Dhanantwari, E. Shefer, Iterative model reconstruction: simultaneously lowered computed tomography radiation dose and improved image quality, Med. Phys. Int. J. 2 (1) (2013) 147–155. [22] C. Thomas, B. Krauss, D. Ketelsen, I. Tsiflikas, A. Reimann, M. Werner, D. Schilling, J. Hennenlotter, C.D. Claussen, H.P. Schlemmer, M. Heuschmid, Differentiation of urinary calculi with dual energy CT: effect of spectral shaping by high energy tin filtration, Invest. Radiol. 45 (7) (2010) 393–398.
[13] C.H. McCollough, S. Leng, L. Yu, J.G. Fletcher, Dual- and multi-energy CT: principles, technical approaches, and clinical applications, Radiology 276 (3) (2015) 637–653. [14] F.A. Dilmanian, X.Y. Wu, E.C. Parsons, B. Ren, J. Kress, T.M. Button, L.D. Chapman, J.A. Coderre, F. Giron, D. Greenberg, D.J. Krus, Z. Liang, S. Marcovici, M.J. Petersen, C.T. Roque, M. Shleifer, D.N. Slatkin, W.C. Thomlinson, K. Yamamoto, Z. Zhong, Single-and dual-energy CT with monochromatic synchrotron x-rays, Phys. Med. Biol. 42 (2) (1997) 371–387. [15] L. Yu, S. Leng, C.H. McCollough, Dual-energy CT-based monochromatic imaging, AJR Am. J. Roentgenol. 199 (Suppl. 5) (2012) S9–S15. [16] J. Cohen, Weighted kappa: nominal scale agreement with provision for scaled disagreement or partial credit, Psychol. Bull. 70 (4) (1968) 213–220. [17] M.H. Albrecht, J. Trommer, J.L. Wichmann, J.E. Scholtz, S.S. Martin, T. Lehnert, T.J. Vogl, B. Bodelle, Comprehensive comparison of virtual monoenergetic and linearly blended reconstruction techniques in third-generation dual-source dualenergy computed tomography angiography of the thorax and abdomen, Invest. Radiol. 51 (9) (2016) 582–590.
58