Comparison of image quality and radiation exposure between conventional imaging and gemstone spectral imaging in abdominal CT examination

Abstract

Objective:

To compare patients’ image quality and radiation exposure between gemstone spectral imaging (GSI) with rapid kV switching technique and conventional polychromatic imaging (CPI) performed in abdominal CT examinations.

Methods:

Adult patients who were referred to abdominal CT from October 2015 to March 2016 were enrolled. Unenhanced CT with CPI mode and tri-phase (arterial/portal/delayed phase) contrast-enhanced scan with GSI mode were performed with different protocols respectively. Regions of interest (ROIs) were drawn on muscle and fat. Parametric results of the image noise, signal-to-noise ratio (SNR) and clinical image quality in these regions between the monochromatic images reconstructed at 65 keV and conventional polychromatic images were compared. Radiation dose was also compared between CPI and GSI.

Results:

43 patients were recruited. Compared to conventional imaging, the noise level was generally not significantly different between GSI images in arterial phase and portal phase, and significantly higher (around 10%) in delayed phase. The SNR of GSI in portal phase was significantly higher than that of conventional imaging, and was similar between arterial phase/delayed phase and conventional imaging. The clinical image quality between conventional imaging and GSI was generally not significantly different. The dose length product was reduced by 0.3–20.1% in GSI compared to conventional imaging.

Conclusion:

GSI reduces the radiation exposure slightly, however maintains or even improves image quality. These results may warrant the application of GSI in patients referred for abdominal CT.

Advances in knowledge:

Compared to CPI, GSI reduces the radiation exposure slightly, however maintains or even improves image quality in abdominal CT. These findings may warrant the application of GSI in patients referred for abdominal CT.

Introduction

The recently developed gemstone spectral imaging (GSI) with dual-energy fast kVp-switching technique has led to the commercialization of dual-energy CT.¹ In GSI, a single X-ray tube produces two sets of polychromatic (mostly 80 and 140 kVp) X-rays almost simultaneously, and acquires paired images in almost identical views.² The high-kVp and low-kVp images are mapped into material density images of the selected basis material pair (water and iodine), and then the virtual monochromatic images computed from the two basis material density images.³ Compared to the conventional polychromatic imaging (CPI) which has been widely used currently, GSI may present lower noise and higher contrast-to-noise ratio (CNR),^4–7 and more information such as the virtual monochromatic image, spectral curve and the iodine mapping, which would be of great potential in evaluating various diseases.⁸ For example, in abdominal and pelvic CT examinations which account for more than 30% of all CT scans in the United States in 2006,⁹ compared to CPI mode, GSI could provide additional quantitative parameters to improve the differentiation of lesions. Yang et al showed that quantitative iodine concentration measurement may be used to improve the evaluation for small hepatocellular carcinoma microvascular invasion.¹⁰ Mileto et al showed that monochromatic spectral images can overcome renal cyst pseudo enhancement within the energy level range of 80–140 keV.¹¹ Meanwhile, water density images may effectively substitute for true unenhanced images with lower radiation exposure.¹²

With the increasing use of CT imaging, the awareness and concern regarding the potential risks of radiation continue to escalate. Studies should be performed to achieve “as low as reasonably achievable” (ALARA) radiation exposure, as well as adequate image quality which satisfies the need of clinical diagnosis. Several studies with GSI in abdominal CT have demonstrated better image quality compared to the CPI mode.^{6, 13,14} A recent study reported radiation dose reduction with better image quality, comparing between a group using GSI and adaptive statistical iterative reconstruction (ASIR), and the other group using CPI in abdominal CT patients with high body mass index value.¹⁵ However, to the best of our knowledge, few studies have been performed to compare the radiation dose and image quality between conventional CT and GSI in the same patient group undergoing abdominal CT. Therefore, the purpose of the present investigation was to compare the image quality in terms of noise, signal-to-noise ratio (SNR), clinical image quality score and radiation exposure of abdominal GSI scan with those of CPI within the same patients.

methods and Materials

Patients

This study was approved by the ethics committee and written informed consent was obtained before each scan. Between October 2015 and March 2016, referred adult patients who clinically indicated upper abdominal CT scan (from navel to diaphragm) or abdominal CT scan (from diaphragm to lower edge of symphysis pubis) were enrolled. Inclusion criteria were: age >18 years; no hypersensitivity to iodine contrast agents; and no pregnancy.

Scan protocol

All patients underwent plain scan [unenhanced phase (UEP)] with CPI mode and tri-phase contrast-enhanced scan with GSI mode on Discovery 750HD CT scanner (GE Healthcare, Milwaukee, WI). After UEP, i.v. administration of a contrast medium (Iopromide, Ultravist 300, Bayer Schering Pharma, Berlin-wedding, Germany) was performed at a dose of 450 mg kg^–1 by using a power injector at a rate of 3 ml s⁻¹. An enhanced scan was then performed with GSI mode in three phases including the arterial phase (AP), portal phase (PP) and delayed phase (DP) with delay time of 30 s, 60 s and 180 s following the injection of contrast medium, respectively. Some scanning parameters were kept constant in these four phases, including: pitch of 0.984, section thickness of 0.625 mm, matrix of 512 × 512, field of view (FOV) of 35 cm and reconstruction thickness of 5 mm. For the unenhanced CPI scan (UEP), the tube potential was set to 120 kVp, and the tube current was defined by the noise index (NI) which was set to 10. The enhanced GSI scan (AP, PP and DP) was performed with fast kVp-switching between 80 kVp and 140 kVp (GSI mode) with different tube current (in mA) choices: AP, maximum mA value in the UEP; PP, mA setting according to the automatic spectral imaging protocol selection technique, which aims to individualize patients’ dose by offering assistance in the selection of the optimal protocol in spectral imaging,¹⁶ with targeted NI of 10; and DP, mean mA [(maximum + minimum)/2] in the UEP. The scanning parameters are summarized in Table 1.

Table 1.

Scan protocol

Scanning Parameters	CPI (unenhanced scan)	GSI (enhanced scan)
Tube voltage (kVp)	120	80 and 140
Tube current (mA)
UEP	(defined by NI of 10, 150~400 mA)
AP		400
PP		(defined by NI of 10)
DP		275
Gantry rotation time (s)	0.7	0.7
Pitch	0.984:1	0.984:1
FOV (cm)	35	35
Matrix	512 × 512	512 × 512
Delay time (s)
AP		30
PP		60
DP		180
Flow rate (ml s−1)	N/A	3
Contrast dose (mg Kg−1)	N/A	450
Section thickness (mm)	0.625	0.625
Reconstruction thickness (mm)	5	5

AP, arterial phase; CPI, conventional polychromatic imaging; DP, delayed phase; FOV, field of view; GSI, gemstone spectral imaging; NI, noise index; PP, portal phase; UEP, unenhanced phase.

All CT images were reconstructed with filtered back projection (FBP) technique. Monochromatic energy level of 65 keV was the default setting used in our hospital for abdominal CT. Meanwhile, studies have shown that monochromatic images of 65 keV yield best image quality for diagnosis.^{17, 18} Thus, for GSI mode, the monochromatic images of 65 keV were reconstructed for the following image quality comparisons.

Assessment of image quality and radiation exposure

To quantitatively and objectively assess the CT image quality, regions of interest (ROIs) were drawn by two experienced radiologists by consensus. All image data were randomized and blinded to the radiologists, and reviewed on Advantage Workstation (v. 4.6; GE Healthcare, Milwaukee, WI). All ROIs were the same in terms of size (approximately 91 mm²) and window width/level. These circular ROIs were placed in four sites in the anterior abdominal wall, including three in the erector spinae muscle (lumbar vertebrae L2 and L4, and thoracic vertebrae T12) and one in the abdominal fat in lumbar vertebrae L2 (Figure 1). Each type of ROI was drawn three times in the different sites and the average of the parameters was calculated. The objective image quality of muscle and fat, compared to those of aorta and the other organs in abdomen, were not significantly affected by the contrast media, and hence were used for the comparison of objective image quality. The standard deviation of the CT numbers (in Hounsfield unit, HU) in each ROI was recorded as the objective image noise. The SNR was calculated using the following formula:

Figure 1.

ROIs drawn on arterial phase images obtained in the same patients. These circular ROIs were drawn on the anterior abdominal wall, including three in the erector spinae muscle (lumbar vertebrae L2 and L4 and thoracic vertebrae T12) and one in the abdominal fat in lumbar vertebrae L2. ROIs, regions of interest.

$${\displaystyle \mathrm{S}\mathrm{N}\mathrm{R}=\frac{\mathrm{C}\mathrm{T}\mathrm{n}\mathrm{u}\mathrm{m}\mathrm{b}\mathrm{e}\mathrm{r}}{\mathrm{S}\mathrm{D}}}$$

where CT number denotes the mean value of corresponding ROI, and SD (standard deviation)denotes the mean value of corresponding objective image noise.

We also carried out an experiment to explore how SNR was affected by the contrast medium. In this experiment, four patients underwent plain scan (UEP) and tri-phase contrast-enhanced scan on Discovery 750HD CT scanner. Both unenhanced phase and enhanced phases were performed with GSI mode and the scanning protocols for these four phases were the same as in Table 1. The ROIs were also drawn the same as above and each type of ROI was drawn three times in different sites. The objective image noise, SNR of each enhanced phase were compared against those of UEP by using paired Student t-test. A p-value of less than 0.05 was considered statistically significant.

Clinical image quality was assessed independently by two radiologists with 17 and 10 years of experience in interpreting abdominal CT images. Subjective image noise was evaluated on a 5-point scale; artifacts, on a 5-point scale; anatomical structure, on a 5-point scale; overall image quality, on a 5-point scale; as described in Table 2. The averages of these two individual scores were then calculated as the clinical image quality assessment.

Table 2.

Grading scale for clinical image quality

Score	Image noise	Artifacts	Anatomical structure	Overall quality
5	Minimum	Minimum	Clear	Superior
4	Less than average	Partial artifacts, not affecting diagnosis	Better than average, not very clear	Above average
3	Average	Artifacts in the whole body, but diagnosis still possible	Average	Average
2	Above average	Artifacts affecting diagnosis	Poorer than average	Suboptimal
1	Unacceptable	Unacceptable	Very Poor	Unacceptable

The dose-length products (DLPs) data of these four phases in the dose report after each scan were recorded as the radiation exposure measurement. The CT dose index (CTDI) data on their own were not used, as the scan length can be different among patients, and hence the DLPs can be different even though the CTDIs are the same. The CTDI values of the scanner are measured and validated by the medical physicists in Panyu Central Hospital in routine quality assurance.

Statistical analysis

The inter-observer agreement was estimated using kappa statistics between the two radiologists for the assessment of clinical image quality. The scale was as follows: less than 0.20, poor; 0.21 to 0.40, fair; 0.41 to 0.60, moderate; 0.61 to 0.80, substantial; and 0.81 to 1.00, almost perfect.^{19, 20} The CT number, objective image noise, SNR, clinical image quality score and DLP of each enhanced phase were compared against those of UEP by using the paired Student t-test. A p-value of less than 0.05 was considered statistically significant. All statistical analysis was performed using a commercially available software package (SPSS v. 21.0; IBM, Chicago, IL).

Results

43 patients (29 male, 14 female) were recruited. The mean age was 51.7 years (24–88 years; standard deviation, 14.7 years). Diagnostic tasks in these patients included the diagnosis of hepatic cyst, renal cyst, hepatic hemangioma, nephrolithiasis, cholecystitis and gallstone.

Objective image quality

The detailed results of CT number, image noise, SNR comparisons are summarized in Table 3. CT number of muscle and fat was generally significantly higher in the three phases (AP, PP and DP) of enhanced GSI scan compared to the unenhanced CPI scan (UEP). The noise level was generally not significantly different between UEP and AP, or between UEP and PP. Significantly higher (around 10%) noise level was observed in DP compared to UEP in fat and muscle ROIs. The SNR in enhanced scan in PP was significantly higher than the unenhanced scan in UEP, and was generally similar between AP/DP and UEP. The effect of the contrast on SNR calculation are summarized in Table 4. The objective image quality were not significantly different between unenhanced phase and enhanced phases in terms of noise and SNR.

Table 3.

Comparison of CT number, noise, SNR between CPI and GSI in the CT

ROI	CT number (HU)				Noise				SNR
L2 Muscle	L2 FAT	L4 Muscle	T12 Muscle	L2 Muscle	L2 FAT	L4 Muscle	T12 Muscle	L2 Muscle	L2 FAT	L4 Muscle	T12 Muscle
UEP
Mean (SD)	54.31  (8.55)	−95.28  (17.45)	57.33  (7.89)	53.78  (8.42)	9.38  (1.40)	12.55  (2.85)	9.84  (1.20)	9.33  (1.22)	5.93  (1.37)	−8.00  (2.36)	5.92  (1.12)	5.89  (1.29)
AP
Mean (SD)	57.37  (8.90)	−97.95  (17.69)	58.32  (9.17)	56.41  (9.84)	9.59  (1.95)	12.19  (2.25)	9.95  (1.60)	9.91  (1.45)	6.17  (1.35)	−8.48  (2.90)	5.98  (1.20)	5.83  (1.41)
p-value	0.003a	<0.001	0.185	0.002	0.441	0.241	0.673	0.033	0.199	0.042	0.739	0.736
PP
Mean (SD)	62.85  (9.38)	−96.49  (19.22)	63.44  (9.44)	62.34  (9.61)	9.81  (1.77)	12.07  (2.49)	10.34  (1.80)	10.05  (1.51)	6.64  (1.65)	−8.50  (3.07)	6.33  (1.48)	6.37  (1.48)
p-value	<0.001	0.319	<0.001	<0.001	0.105	0.074	0.074	0.004	0.001	0.029	0.052	0.031
DP
Mean (SD)	64.41  (8.75)	−94.60  (20.76)	65.92  (8.49)	64.25  (7.60)	10.72  (1.45)	13.38  (2.83)	11.00  (1.69)	10.56  (1.80)	6.11  (1.12)	−7.43  (2.45)	6.15  (1.31)	6.27  (1.37)
p-value	<0.001	0.484	<0.001	<0.001	<0.001	0.030	0.001	<0.001	0.312	0.008	0.316	0.075

AP, arterial phase; CPI, conventional polychromatic imaging; DP, delayed phase; GSI, gemstone spectral imaging; HU, Hounsfield unit; Noise, objective image noise (standard deviation of CT numbers in an ROI); PP, portal phase; ROI, region of interest; SD, standard deviation of noise; SNR, signal-to-noise ratio; UEP, unenhanced phase.

L2 Muscle, erector spinae muscle at the lumbar vertebrae level L2; L2 FAT, fat at the level of lumbar vertebrae L2; L4 Muscle, erector spinae muscle at the lumbar vertebrae level L4; T12 Muscle = erector spinae muscle at the thoracic vertebrae level T12.

^ap-values which are highlighted using italic and bold type are the significance level in the paired Student t-test between the noise in unenhanced phase and arterial phase (or portal phase, or delayed phase), and p-value less than 0.05 was considered statistically si

Table 4.

The comparison of objective image quality between unenhanced phase and enhanced phase with GSI model

Comparison	p-value of Noise	p-value of SNR
UEP-AP	0.942	0.156
UEP-PP	0.347	0.145
UEP-DP	0.239	0.056

GSI, gemstone spectral imaging; SNR, signal-to-noise ratio; UEP-AP, comparison between unenhanced phase and arterial phase, UEP-PP, comparison between unenhanced phase and portal phase; UEP-DP, comparison between unenhanced phase and delayed phase.

Clinical image quality

The agreement in terms of Kappa coefficient between the two radiologists was moderate to perfect (κ = 0.256~0.890). The percentage agreement between the two radiologists ranged from 76.7% (33/43) to 97.7% (42/43), as summarized in Table 5.

Table 5.

Percentage agreement and Kappa coefficient (κ) between clinical image quality scores of CPI and GSI determined by two radiologists

Image quality		UEP	AP	PP	DP
Noise	Agreement	93% (40/43)	81.4% (35/43)	90.7% (39/43)	76.7% (33/43)
κ	0.677	0.604	0.652	0.496
Artifacts	Agreement	95.3% (41/43)	93.0% (40/43)	90.7% (39/43)	90.7% (39/43)
κ	0.890	0.831	0.719	0.778
Anatomical structure	Agreement	90.7% (39/43)	88.4% (38/43)	93% (40/43)	90.7% (39/43)
κ	0.452	0.256	0.377	0.466
Overall image quality	Agreement	93% (40/43)	79.1% (34/43)	97.7% (42/43)	83.7% (36/43)
κ	0.860	0.625	0.879	0.707

AP, arterial phase; CPI, conventional polychromatic imaging; DP, delayed phase; GSI, gemstone spectral imaging; PP, portal phase; UEP, unenhanced phase.

The clinical image quality scores are summarized in Figure 2. In most comparisons, the clinical image quality between CPI and the three phases of GSI scan was not significantly different. The image quality of PP were superior to that of UEP in terms of subjective image noise and overall image quality. The overall image quality of DP were worse compared to that of UEP.

Figure 2.

Comparison of clinical image quality score between GSI and CPI mode in abdomen CT. The scores of PP were significantly higher than those of UEP in terms of subjective image noise and overall image quality, while the overall image quality of DP was significantly lower than that of UEP (all p < 0.01). CPI, conventional polychromatic imaging; GSI, gemstone spectral imaging.

Radiation Dose

The DLPs are summarized in Table 6 for the four phases. Compared with those in UEP with CPI mode (400.6 ± 213.5 mGycm), DLPs in the GSI scan were reduced: in AP (383.9 ± 190.3 mGycm), by 4.2%; in PP (399.4 ± 192.4 mGycm), by 0.3%; and in DP (320.0 ± 138.5 mGycm), by 20.1%. In AP and PP, the reduction was insignificant statistically. In DP, the reduction was significant with p < 0.001. NI of PP was the same as of UEP; nonetheless, the doses were slightly different (0.3%) because of the different tube voltages.

Table 6.

Comparisons of radiation exposure between conventional imaging and rapid kV switching dual-energy imaging in abdomen CT

Parameter	UEP	AP	PP	DP
Mean scan length/mm	213.7	213.7	213.7	213.7
DLP range/mGycm	123.3–1036.6	196.8–930.0	195.6–954.3	196.8–683.4
DLP Mean (SD)/mGycm	400.6 (213.5)	383.9 (190.3)	399.4 (192.4)	320.0 (138.5)
DLP Reduction/%	–	−4.2%	−0.3%	−20.1%
p-value	–	0.194	0.810	<0.001a

AP, arterial phase; DP, delayed phase; DLP, dose length product (in mGycm); PP, portal phase; SD, standard deviation; UEP, unenhanced phase.

Reduction_X, ( Mean_X – Mean_UEP)/ Mean_UEP ×100% (Mean_X is the mean value of phase X).

^ap-value which is highlighted using italic and bold type is the significance level in the paired Student t-test between the DLPs in unenhanced phase and arterial phase (or portal phase, or delayed phase), and p-value less than 0.05 is considered statistically significant.

Discussion

Our study has shown that GSI scan at 65 keV from fast kVp-switching spectral CT generally maintained the image quality (both objective and subjective), compared with conventional polychromatic 120-kVp images, at the expense of the same or lower levels of radiation dose in the same patients undergoing conventional abdominal CT examinations with CPI mode. These results echoed the previous findings in the literature. One previous study found that, at the same radiation dose, the image noise in monochromatic images of 70 keV were reduced, also compared to 120-kVp images performed during the same abdominal CT scan.⁶ Zhu et al also reported that the use of GSI and ASIR achieved comparable image quality with reduced radiation dose compared with those of conventional scan protocol.¹⁵ However, Zhu et al did not compare two images obtained in the same patients. To summarize, our results, by comparing the doses from the same patient group, indicated that the implementation of GSI in clinical practice is feasible with more information, slightly lower radiation and comparable image quality.

The image quality vs dose performance is known to be a function of both the patient size and the X-ray spectrum.²¹ In our study, the effect of patient size has been eliminated by performing intra-individual comparison. Generally, the X-ray spectrum depends on the tube current-time product and tube voltage. Traditional CT uses polychromatic X-ray, and the low-energy X-ray generates beam hardening artifacts and affects the image quality.²² In GSI mode, by fast kVp-switching technique, projections are acquired with two different polychromatic spectra of 80 kVp and 140 kVp alternately every 0.2 milliseconds.² The exposures at 80 kVp are generally longer than those at 140 kVp.²³ Hence, in present study, the average kVp in GSI mode was less than 110, which was lower than 120 kVp in CPI mode. The average tube current in GSI was higher in AP, and lower in DP than that in UEP scan (Table 1). In PP scan, the NI was chosen to be 10 for setting tube current product empirically according to prior studies,^{24, 25} the same as in UEP. Lowering the tube current-time product or tube voltage would generate lower radiation dose but is accompanied by noisier images, and vice versa. This fact may explain the significantly lower dose in DP to a certain extent. However, compared to UEP, the image quality of the enhanced scans in GSI mode was better or comparable. It was reported that monochromatic images reconstructed from GSI scan diminish beam hardening artifacts.²⁶ Therefore, by using GSI mode, although the radiation exposure is comparable or lower, the image quality in the monochromatic images can be maintained.

To compare the image quality, we performed the enhanced CT scan with GSI technique and the unenhanced imaging with CPI mode. The comparisons of image quality between enhanced images and unenhanced images may not be reasonable. Thus, we drew ROIs in fat and muscle, in which the enhancement is not significant in the CT images (Figure 1). Also we did an experiment to explore how SNR was affected by the presence of the contrast agent and the results (Table 4) indicated that with GSI mode SNR is not significantly affected by contrast. Indeed, the enhancement in fat and muscle, as shown in the different CT numbers in Table 3, albeit small, may induce error in the comparison of the noise. However, for the purpose of radiation protection, it is usually unacceptable to do two scans with GSI mode and CPI mode respectively in each patient especially when these two scans were not necessary for diagnosis. One alternative choice is to compare two groups of patients: one with GSI mode and the other with CPI mode. However, since the patients are not the same in the two groups, the comparison of radiation dose and image quality is still unfair.

We did not evaluate CNR values in this study, but just the CT values, image noise and SNRs in terms of the objective image quality. This is because, an accurate CNR comparison is possible only if the absolute CT values in both GSI and CPI scans are comparable. However, in our study, this cannot be achieved in the ROIs of liver and abdominal aorta because in GSI mode the images were enhanced while in CPI scans are unenhanced, and the “contrast” in CNR is not the same.

In our study, for GSI mode, the monochromatic images of 65 keV were used in the image quality analysis. As stated above, we had some reasons for this choice; however, different keV images may have some effect on the measures of image quality. In this regard, Cui et al reported that 70 keV monochromatic images had significantly better image quality than CPI images in abdominal CT,²⁷ and Lv et al reported that the noise levels in 50 to 70 keV monochromatic images were similar or significantly lower compared to CPI images.¹⁶ To some extent, different choices of monochromatic energy level may lead to different results of quantitative assessment of image quality; however, such difference of image quality between 60/70 keV and 65 keV would not be so significant to ultimately affect our conclusion.

Our study has several limitations. First, only a small cohort of 43 patients was included in this study. Thus the statistical power may not be strong enough. Further investigations with larger samples are necessary in order to confirm these findings. Secondly, the calculation of quantitative parameters in ROIs may be dependent on the locations of these ROIs, thus each ROI was drawn three times. We calculated the variance of CT numbers and noise among these three measurements, and the results were less than 5 and 10% of mean values respectively for CT number and noise. We concluded that, such errors cannot be completely eliminated, however may be reduced to an acceptable level by placing multiple ROIs and calculating the mean values. Thirdly, our study did not have assessments with a specific diagnostic task. A more thorough design with specific diagnostic tasks would provide more practical information for this kind of studies. Finally, all images were reconstructed with FBP but not ASIR which may lower the image noise further. This was because in our clinic, most of the abdomen CT scans were reconstructed with combined ASIR-FBP method; however, in the unenhanced scan and enhanced scan the ratio of ASIR were different. Thus, for a fair comparison between GSI and CPI modes, we reconstructed the images with only the FBP method.

Conclusions

With tailored protocols in GSI scan, the monochromatic spectral images in clinically abdomen CT of 65 keV maintain the image quality compared to the conventional polychromatic images, while the radiation exposure is comparable or slightly reduced. Our results may add to a growing evidence base to warrant the application of GSI in patients referred to clinically abdomen CT.