Clinical routine use of virtual monochromatic datasets based on spectral CT in patients with hypervascularized abdominal tumors - evaluation of effectiveness and efficiency

Abstract

Background

Virtual monochromatic images (VMI) generated using spectral computed tomography (CT) are promising recently available tools to improve diagnostic performance in oncologic patients.

Purpose

To investigate if virtual monochromatic datasets are suitable for clinical routine use in patients with hypervascularized abdominal tumors.

Material and Methods

A total of 41 patients with hypervascularized hepatocellular carcinoma (HCC), renal cell carcinoma (RCC), or neuroendocrine tumors (NET) were enrolled in the study; 451 CT series were analyzed. In an intra-individual study design, virtual monochromatic datasets of the arterial phase of each scan were computed. Image quality was assessed objectively by determining signal-to-noise ratio (SNR) and contrast-to-noise ratios (CNR) and subjectively by using five-point Likert-scales. The volume CT dose index (CTDI_vol) was taken from each radiation dose report. The increase in reading time was estimated from the increase in the number of images.

Results

Intra-individual comparison of the spectral mode in the arterial phase with the portal venous phase revealed no significant increase in the applied dose. SNR, CNR_{tumor-to-liver} , and CNR_{tumor-to-muscle} were significantly increased by lowering virtual monochromatic energy. Subjective image quality scores revealed an increase of contrast in low energy datasets, resulting in significantly higher diagnostic confidence, but an increased image noise at low energies. While diagnostic confidence improved, taking all datasets into account resulted in a significantly longer estimated reading time.

Conclusion

In clinical practice, the use of low energy VMI improved diagnostic confidence without a significant increase in dose. The main disadvantage is a decrease in efficiency due to longer reading times.

Keywords

Abdomen computed tomography CT image reconstruction radiation dosage reader workflow optimization dual energy

Introduction

A typical imaging finding in patients with hepatocellular carcinoma (HCC) is hypervascularization during the arterial phase of contrast enhancement. Another tumor typically showing arterial phase hypervascularization is renal cell carcinoma (RCC). Neuroendocrine tumors (NETs) of the digestive system are a rare and heterogeneous group of neoplasms, which are also typically hypervascularized. In advanced disease, oncologic surveillance includes regular staging by conventional imaging in order not to miss progression of known tumor lesions or occurrence of new ones.

In 1973, Hounsfield demonstrated that material differentiation and quantification in computed tomography (CT) is possible by using different X-ray energies (1). Since the first dual-energy methods were investigated, several technical approaches have been developed (2,3). The known higher attenuation of iodine at lower X-ray energy levels has the potential to improve image contrast, especially in patients with highly vascularized tumors (4,5). Although CT is frequently used in the primary diagnostic work-up and follow-up of oncologic patients, all options for dose reduction should be explored. This is especially important for patients with HCC, NET, or RCC who underwent curative surgery or patients suffering from low-grade NETs or slow-growing NCCs with a relatively good prognosis and long disease course, as they may undergo frequent CT scans (6,7). In these instances, the accumulated applied dose is high and increases secondary carcinogenic risks (8,9).

The evaluation of new imaging techniques for clinical routine involves consideration of both effectiveness and efficiency. The aim of this study was therefore to analyze the image quality of virtual monochromatic datasets and the radiation dose of spectral CT to evaluate their potential effectiveness for routine clinical use in oncologic patients while also taking efficiency of the reading process into account.

Material and Methods

Patient characteristics and study design

This study was approved by the institutional ethics board. Patient data were stored anonymously. From March 2016 to March 2017, 295 patients with NET, HCC, or RCC underwent CT examinations at our institution. Inclusion criteria were histologically verified tumor and a hypervascularized primary tumor and/or metastatic lesion visible on abdominal multiphase CT scans including a spectral mode arterial phase and standard mode portal venous phase acquired on a spectral CT scanner with ultra-rapid kVp switching. Exclusion criteria were absence of visible tumor, pregnancy, and patient age < 18 years. A total of 41 patients (20 men, 21 women; 26 NET/NEC, nine HCC, six RCC; mean age = 65 ± 13 years) were enrolled in this study. In an intra-individual study design, quantitative and qualitative image quality as well as applied dose were compared, and the related increase in reading time was estimated.

CT technique

Contrast-enhanced multiphase scans were acquired on a 128-multislice CT scanner (Revolution HD, GE Healthcare, Milwaukee, WI, USA). Following acquisition of a posterior–anterior scout and intravenous contrast medium injection, scans were acquired during arterial, portal venous, and venous phases. Phases were defined using automated scan-triggering software (SmartPrep®, GE Healthcare). Delay times after reaching the attenuation threshold (150 HU) in the suprarenal aorta were set to 9 s (arterial phase), 55 s (portal venous phase), and 120 s (venous phase) according to Gordic et al. for HCC (10). Patients received non-ionic contrast medium (Xenetix 350®, Guerbet, Villepinte, France) at a dose of 1.5 mL/kg body weight (max. 120 mL) and a flow rate of 4 mL/s followed by a saline solution flush using a mechanical injector (Medtron CT2, MEDTRON AG, Saarbrücken, Germany).

Gemstone spectral imaging (GSI) CT allows the creation of several datasets, such as VMI, based on the acquisition of polychromatic raw data by rapid tube potential switching between 80 and 140 peak kilovoltage (kVp). Except for use of the dual-energy mode, automated dose modulation (not provided for spectral mode by vendor), and tube current (dual-energy mode: average = 260–640 mAs; single-energy mode: range = 100–500 mAs), all further CT parameters (noise index = 21, pitch = 1.375, collimation = 64 × 0.625 mm, rotation time = 0.7 s, recon. ASIR-level = 70%) were kept constant for acquisition of the different contrast phases in this study and images were reconstructed using a soft tissue kernel. VMI of the spectral arterial phase were computed using GE’s GSI Volume Viewer® application in 10-keV increments from 40 keV to 140 keV at 0.625-mm slice thickness.

Quantitative image analysis

All analyses were performed on a commercially available workstation (Advantage Workstation®, GE Healthcare) using preset window settings (width = 400 HU, center = 50 HU). In one slice from a 40-keV dataset, circular regions of interest (ROIs) were manually placed in the aorta, liver, pancreas, paraspinal muscle, fat of the anterior abdominal wall, and pre-abdominal air and then cloned to all other series, ensuring an identical positioning of the ROIs in all analyzed datasets. The image level showing the strongest hypervascularization of tumor in the arterial phase was used to place a polygonal ROI in the lesion and a circular ROI in adjacent normal liver parenchyma. The area of the lesion was measured. To calculate signal-to-noise ratio (SNR) and contrast-to-noise ratios (CNR), standard deviation of HU in pre-abdominal air was defined as a denominator. Numerators were defined as the mean HU of all ROIs (SNR) or of specific ROIs (CNR) (CNR = mean HU of ROI_{tissue 1} − mean HU of ROI_tissue 2/standard deviation of HU of ROI_fat). CNRs were calculated as CNR_{tumor-to-liver} , CNR_{tumor-to-muscle}, and CNR_{liver-to-muscle.}

Qualitative image analysis

Measurements were performed by two independent, experienced, and blinded readers, who could change zoom and window settings. Subjective image quality was scored in six categories (image noise, contrast, visibility of small structures, suspicious lesion conspicuity, artifacts, diagnostic confidence) using a 5-point Likert scale (1 = insufficient to 5 = excellent).

Radiation dose

The volume CT dose index (CTDI_vol) as defined by the U.S. Department of Health and Human Services is a standard parameter for estimating the CT radiation dose based on phantom measurements. It is independent of patient size as well as scan length; it was therefore used in this study and taken from the dose report of each scan.

Potential reading time

The potential reading time for evaluation of virtual monochromatic datasets as a surrogate parameter for efficiency of workflow compared with the time required for the standard process was estimated from the number of images which had to be taken into consideration. This procedure assumes the same reading intensity for all datasets, which leads to a direct correlation between the number of images and reading time. As the standard reading process includes at least thin axial images, their number was assumed to define the standard reading time. The relative increase in reading time was then estimated by comparing the number of images in the thin axial standard series with the number of images in the virtual monochromatic series.

Statistical analysis

All statistical analyses were performed using commercially available software (SPSS v. 23, IBM., Armonk, NY, USA). Values of P ≤ 0.05 were considered statistically significant while P ≤ 0.1 was considered a trend. Graphics were created with GraphPad Prism v.5® (GraphPad Software, San Diego, CA, USA). Normal distribution of data was analyzed with the Kolmogorov–Smirnov (KS) test. Based on the results of the KS test, a paired t-test was performed for SNR and Wilcoxon signed-rank test for applied dose, CNRs, and subjective scores. Kendall tau-b was used to evaluate interrater reliability (−1 < τ < 1, − 1 = perfect disagreement, 1 = perfect agreement). Entity-dependent attenuation at 40 keV was analyzed using the Kruskal–Wallis test.

Results

Patient/tumor characteristics

We included 26 NET/NEC patients, nine HCC patients, and six RCC patients (20 men, 21 women; mean age = 65 years). The most common primary tumor sites of the NET/NEC were pancreas (n = 9) and ileum (n = 8); the tumors were mostly graded as G2 (n = 18); and eight patients suffered from carcinoid syndrome. AJCC stage was 4 in all RCCs and most NETs (n = 23) and the BCLC stages of HCC were 2 (n = 5) or 3 (n = 4). Patients’ maximum abdominal diameters in sagittal (256 ± 39 mm) and coronal (342 ± 38 mm) orientation matched with previous observations in oncologic patients in our department. The mean tumor area of the strongest hypervascularized lesion was 2029 ± 2941 mm².

Image quality analysis

Image quality analysis results showed an added diagnostic value for virtual monochromatic datasets at low energy levels in the detection of hypervascularized abdominal tumors. As theoretically predicted, attenuation measurements showed increased levels in both tissues and tumors at lower energies (Figs. 1 and 2). We also observed a trend (P = 0.061) toward entity-dependent differences in mean tumor attenuation, which reached the highest levels at 40 keV. We found higher attenuation levels at 40 keV especially in the RCC group compared to HCC and NET (RCC = 421 ± 167 HU, HCC = 245 ± 70 HU, NET =266 ± 140 HU, Fig. 2). We observed significant (P < 0.001) increases of SNR, CNR_{tumor-to-liver} , and CNR_{tumor-to-muscle} at lower energy levels (Fig. 3). The higher scores for low energy datasets indicate that the increase in signal through greater iodine attenuation is relatively higher than the simultaneous rise in image noise.

Fig. 1.

ROI-based spectral curves – attenuation of tissues in arterial phase. Quantitative image analysis showed more marked attenuation of tissues at lower virtual monochromatic energy levels due to higher attenuation of the iodine contrast agent. The largest attenuation differences between the energy levels were observed in the aorta (not shown due to axis limitations by other tissues) followed by tissues with strong vascularization such as hypervascularized tumors and pancreas. In tissues with less vascularization such as liver and muscle, we found smaller differences for different keV levels.

Fig. 2.

ROI-based spectral curves – attenuation of arterial hypervascularized tumors. Quantitative image analysis showed a trend (P = 0.061) toward entity-dependent differences in mean tumor attenuation, which were greatest at 40 keV. We found more marked attenuation at 40 keV, especially in the RCC group compared to HCC and NET, but the differences were not statistically significant, possibly because of the small sample size.

Fig. 3.

Objective image quality. We found significant (P < 0.001) increases in SNR, CNR_{tumor-to-liver} , and CNR_{tumor-to-muscle} at lower energy levels. The higher scores for low energy datasets indicate that the increase in signal through greater iodine attenuation is relatively higher than the concomitant rise in image noise, which may contribute to tumor detection.

Subjective image quality and interrater reliability were very good for all different keV datasets (Figs. 4 and 5, Table 1). We found a typical increase of contrast and image noise at low energy levels. Furthermore, diagnostic confidence was significantly higher at lower energies, confirming the objective measurements, which showed increased SNR/CNRs due to a relatively higher increase of useful signal compared to the increase of noise.

Fig. 4.

Subjective image quality. Higher iodine contrast in virtual monochromatic datasets with lower energy leads to better visualization of hypervascularized tumors despite a concomitant increase in image noise. Row a: a 61-year-old woman with HCC in liver segment II, applied dose (CTDI_vol) = 6.36 mGy; row b: a 73-year-old woman with large RCC of the kidney, applied dose (CTDI_vol) = 7.52 mGy; row c: a 69-year-old man with NET of the body of pancreas and hepatic metastasis, applied dose (CTDI_vol) = 6.36 mGy. Arrowheads show the tumor.

Fig. 5.

Subjective image quality – scoring. Subjective scores were significantly (P < 0.001) different between 40 keV and 140 keV. Higher iodine contrast in low energy datasets leads to better contrast, resulting in an increased visibility of hypervascularized tumors and small structures. As a consequence, diagnostic confidence was also higher at lower energies. Nevertheless, scores for visibility of suspected lesions and diagnostic confidence were highest at the 60 keV level due to significantly higher image noise in low energy datasets.

Table 1.

Subjective image quality – interrater reliability.

Virtual monochromatic energy (keV)	Kendall tau-b (τ)
Virtual monochromatic energy (keV)	40	50	60	70	80	90	100	110	120	130	140
Noise	0.76	0.81	0.85	0.80	0.79	0.75	0.72	0.85	0.89	0.82	0.87
Contrast	1.00	0.88	0.88	0.81	0.90	0.86	0.93	0.77	0.86	0.87	0.80
Visibility of small structures	0.93	0.93	0.91	0.92	0.92	0.90	0.88	0.84	0.91	0.87	0.84
Visibility of suspect lesions	0.87	0.83	0.92	0.89	0.89	0.89	0.85	0.90	0.89	0.94	0.88
Artifacts	0.91	0.93	0.92	0.91	0.90	0.92	0.91	0.91	0.94	0.91	0.89
Diagnostic confidence	0.90	0.84	0.94	0.91	0.85	0.86	0.87	0.91	0.88	0.91	0.89

Kendall tau-b was used to evaluate interrater reliability (−1 < τ < 1, −1 = perfect disagreement, 1 = perfect agreement) and we found good to excellent interrater reliability for all datasets. Agreement was highest for contrast at the 40-keV level and lowest for noise at the 100-keV level.

Radiation dose

The use of the dual-energy-based spectral mode for acquisition of the arterial phase of the scans did not significantly increase the applied dose intra-individually compared to the standard single-energy portal venous phase (Fig. 6; dual-energy arterial phase = 9.6 ± 4.7 mGy, single-energy portal venous phase = 10.5 ± 5 mGy). Nevertheless, we observed differences in the distribution of the applied doses (Fig. 6).

Fig. 6.

Radiation doses of multiphase CT scans combining spectral and conventional scans. The use of the spectral mode in the arterial phase of multiphase CT scans did not significantly increase the applied dose as revealed by intra-individual comparison with the dose of the standard single-energy portal venous phase acquisition (dual-energy GSI spectral mode arterial phase = 9.6 ± 4.7 mGy, single-energy portal venous phase = 10.5 ± 5 mGy). Nevertheless, we observed differences in the distribution of the applied doses, which is probably attributable to the lack of automated tube current modulation in the spectral mode (not available from vendor). While the use of ATCM led to a normal distribution of dose values in standard CT, the vendor-set minimal dose for spectral scans defines a lower border resulting in a narrower and skewed dose distribution in these series.

Potential reading time

Under assumption of the same reading intensity for all datasets, the potential total reading time including virtual monochromatic datasets, which we used as a surrogate parameter for workload and efficiency in clinical routine use, was estimated to be 11 times higher compared to the standard procedure based on an increase of number of images which had to be taken into consideration.

Discussion

CT images acquired at low energy levels have been shown to increase image contrast and conspicuity of suspicious lesions, resulting in higher diagnostic confidence, especially with regard to the detection of hypervascularized tumors (4,11,12).

The results of our study show that contrasts are significantly increased in objective analysis at low VMI energy levels. We also observed the typical increase in image noise in subjective analysis, but higher scores for diagnostic confidence indicate that the increase in signal is relatively higher than the rise of image noise, which makes these datasets valuable for tumor detection in the clinical setting. These findings are confirmed by our objective data, which show higher SNR and CNRs at lower energies. Similar observations have been reported in previous studies (4,13 –16). Other studies found that differences in attenuation could be useful for differentiation between HCC and benign liver lesions (14 –16). Our results also show a trend toward tumor-entity-dependent differences in attenuation, especially at very low energies. The fact that these differences did not reach significance may be due to the small sample size.

Because the X-ray absorption of tissues is higher at lower energy levels, the use of dual-energy spectral CT could increase the applied dose compared to standard mode (9,17). The use of spectral mode in the arterial phase did not significantly increase the applied dose when intra-individually compared with standard mode portal venous phase in our data. Nevertheless, we observed differences in the distribution of applied doses, which is probably attributable to the lack of automated tube current modulation (ATCM). While the use of ATCM led to a normal distribution of dose values in standard CT, the vendor-set minimal dose for spectral scans defines a lower border, resulting in a narrower and skewed dose distribution. Our results are in agreement with previous observations, which reported even higher applied doses for abdominal DECT scans (14 –16). These higher doses are probably due to the use of different CT protocols for answering different clinical questions. The protocol used in this study has been optimized for routine clinical examinations of patients with hypervascularized abdominal tumors and can therefore be recommended for this indication to GE scanner users. Other vendors may have further options to improve monochromatic imaging as reported, e.g. for Siemens by Husarik et al. (13).

Some additional aspects need to be considered to evaluate whether VMI is suitable for clinical routine use. Against the background of pressure to cut costs, workflows must be optimized. In our opinion, the most important parameters are the effort to generate the additional data and the time a reader needs to take them into account. With regard to VMI, the effort to generate the data is minimal due to a few-click solution. On the other hand, it should be noted that VMI, as used here, means an increase in the number of images to read for the arterial phase by a factor of 11. In our experience, the extra time required to read all images is too time-consuming for an efficient routine clinical workflow. Therefore, a simplified solution is needed to use the additional information in clinical routine.

One approach involves analysis of ROI-based spectral curves instead of the full set of images for lesion characterization (15). Consistent with the reports from Yu et al., the results of this study show similarities in the shape of the spectral curves of tumors in arterial phase (Fig. 2) (15). However, given the high interindividual variation, the shape of spectral curves appears not to be suitable for routine use. Another approach has been proposed by Yang et al. which uses the spectral curve slope between 40 keV and 100 keV (18). According to our results, this could be a possible method for the characterization of abdominal tumors once lesions have already been identified, but interindividual variation is also high, and the slope is dependent on the distance between the energy levels selected for calculation.

Our first clinical experiences with spectral CT in patients with abdominal tumors suggest that there is a need for a “screening” series for the identification of suspected tumor lesions taking spectral information into account. The results of our study further indicate that a kind of combination/blending of high and low energy VMI would be desirable to take advantage of the high contrast from low and the low noise from high energy datasets. This approach should be evaluated in further studies.

This study has some limitations. First, we used an intra-individual patient analysis to minimize methodological bias by avoiding interpersonal differences in the morphologic appearance of suspicious lesions, which might have influenced evaluation of both objective and subjective image quality as well as dose measurement. Second, measurement errors are possible due to manually placement of ROIs despite care in avoiding potentially interfering structures. As just one ROI in one series was placed, a three-dimensional tumor-segmentation may deliver other results due to heterogeneous iodine uptake of lesions. Third, the contrast-phase timing was chosen following the literature reports for HCC as there were no specific recommendations for RCC or even the rare NET. Furthermore, in clinical routine the tumor entity is not always known before the scan and the observed differences in the standardized protocol may be helpful whether they are based on entity-dependent variances in peak times or entity-dependent differences in maximum enhancement. Fourth, we did not analyze this in detail, but Mileto et al. utilized an adjusted multivariate regression model to calculate an optimal monochromatic energy level depending on patients’ diameters (19). Finally, we did not analyze further options of spectral imaging such as iodine mapping which was, for example, suggested by Thaiss et al. to deliver surrogate parameters for perfusion in HCC and would be of interest for clinical routine use (20).

In conclusion, our results indicate that VMI, based on spectral CT scans acquired in the arterial contrast phase, significantly improve diagnostic confidence in staging hypervascularized abdominal tumors, especially at low energy levels, without a significant increase in dose. Furthermore, we found a trend toward entity-dependent differences in tumor attenuation at low energy levels, which might contribute to the differential diagnosis. On the other hand, the time required for reading the complete VMI is considerable and precludes efficient routine use of this method. Alternative, less time-consuming approaches for exploiting the information provided by VMI such as using a kind of screening series, e.g. image fusion of different keV images, should therefore be evaluated and established in further studies.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Hounsfield

. Computerized transverse axial scanning (tomography). 1. Description of system. Br J Radiol 1973; 46: 1016–1022.

McCollough

Leng

et al.

Dual- and multi-energy CT: principles, technical approaches, and clinical applications. Radiology 2015; 276: 637–653.

De Cecco

Boll

Bolus

et al.

White paper of the Society of Computed Body Tomography and Magnetic Resonance on Dual-Energy CT, Part 4: Abdominal and Pelvic Applications. J Comput Assist Tomogr 2017; 41: 8–14.

Boning

Kahn

Kaul

et al.

CT follow-up in patients with neuroendocrine tumors (NETs): combined radiation and contrast dose reduction. Acta Radiol 2018; 59: 517–526.

De Cecco

Darnell

Rengo

et al.

Dual-energy CT: oncologic applications. Am J Roentgenol 2012; 199: 98–105.

Boning

Schafer

Grupp

et al.

Comparison of applied dose and image quality in staging CT of neuroendocrine tumor patients using standard filtered back projection and adaptive statistical iterative reconstruction. Eur J Radiol 2015; 84: 1601–1607.

Pavel

O’Toole

Costa

et al.

ENETS consensus guidelines update for the management of distant metastatic disease of intestinal, pancreatic, bronchial neuroendocrine neoplasms (NEN) and NEN of unknown primary site. Neuroendocrinology 2016; 103: 172–185.

Brenner

Elliston

. Estimated radiation risks potentially associated with full-body CT screening. Radiology 2004; 232: 735–738.

Foley

Shuman

Siegel

et al.

White paper of the Society of Computed Body Tomography and Magnetic Resonance on Dual-Energy CT, Part 2: Radiation Dose and Iodine Sensitivity. J Comput Assist Tomogr 2016; 40: 846–850.

10.

Gordic

Puippe

Krauss

et al.

Correlation between dual-energy and perfusion CT in patients with hepatocellular carcinoma. Radiology 2016; 280: 78–87.

11.

Altenbernd

Heusner

Ringelstein

et al.

Dual-energy-CT of hypervascular liver lesions in patients with HCC: investigation of image quality and sensitivity. Eur Radiol 2011; 21: 738–743.

12.

Lin

Chen

et al.

Spectral CT in patients with small HCC: investigation of image quality and diagnostic accuracy. Eur Radiol 2012; 22: 2117–2124.

13.

Husarik

Gordic

Desbiolles

et al.

Advanced virtual monoenergetic computed tomography of hyperattenuating and hypoattenuating liver lesions: ex-vivo and patient experience in various body sizes. Invest Radiol 2015; 50: 695–702.

14.

Sun

et al.

Differentiating hepatocellular carcinoma from angiomyolipoma of the liver with CT spectral imaging: a preliminary study. Clin Radiol 2013; 68: 491–497.

15.

Lin

Chen

et al.

Hepatocellular carcinoma and focal nodular hyperplasia of the liver: differentiation with CT spectral imaging. Eur Radiol 2013; 23: 1660–1668.

16.

Lin

et al.

Differentiation of small hepatic hemangioma from small hepatocellular carcinoma: recently introduced spectral CT method. Radiology 2011; 259: 720–729.

17.

Henzler

Fink

Schoenberg

et al.

Dual-energy CT: radiation dose aspects. Am J Roentgenol 2012; 199: 16–25.

18.

Yang

Dong

Wang

et al.

Non-small cell lung cancer: Spectral computed tomography quantitative parameters for preoperative diagnosis of metastatic lymph nodes. Eur J Radiol 2017; 89: 129–135.

19.

Mileto

Nelson

Samei

et al.

Dual-energy MDCT in hypervascular liver tumors: effect of body size on selection of the optimal monochromatic energy level. Am J Roentgenol 2014; 203: 1257–1264.

20.

Thaiss

Haberland

Kaufmann

et al.

Iodine concentration as a perfusion surrogate marker in oncology: Further elucidation of the underlying mechanisms using volume perfusion CT with 80 kVp. Eur Radiol 2015; 26: 2929–2936.