Characterization of Incidental Renal Mass With Dual-Energy CT: Diagnostic Accuracy of Effective Atomic Number Maps for Discriminating Nonenhancing Cysts From Enhancing Masses
Abstract
OBJECTIVE. The purpose of this study was to assess the diagnostic accuracy of effective atomic number maps reconstructed from dual-energy contrast-enhanced data for discriminating between nonenhancing renal cysts and enhancing masses.
MATERIALS AND METHODS. Two hundred six patients (128 men, 78 women; mean age, 64 years) underwent a CT renal mass protocol (single-energy unenhanced and dual-energy contrast-enhanced nephrographic imaging) at two different hospitals. For each set of patients, two blinded, independent observers performed measurements on effective atomic number maps from contrast-enhanced dual-energy data. Renal mass assessment on unenhanced and nephrographic images, corroborated by imaging and medical records, was the reference standard. The diagnostic accuracy of effective atomic number maps was assessed with ROC analysis.
RESULTS. Significant differences in mean effective atomic numbers (Zeff) were observed between nonenhancing and enhancing masses (set A, 8.19 vs 9.59 Zeff; set B, 8.05 vs 9.19 Zeff; sets combined, 8.13 vs 9.37 Zeff) (p < 0.0001). An effective atomic number value of 8.36 Zeff was the optimal threshold, rendering an AUC of 0.92 (95% CI, 0.89–0.94), sensitivity of 90.8% (158/174 [95% CI, 85.5–94.7%]), specificity of 85.2% (445/522 [95% CI, 81.9–88.2%]), and overall diagnostic accuracy of 86.6% (603/696 [95% CI, 83.9–89.1%]).
CONCLUSION. Nonenhancing renal cysts, including hyperattenuating cysts, can be discriminated from enhancing masses on effective atomic number maps generated from dual-energy contrast-enhanced CT data. This technique may be of clinical usefulness when a CT protocol for comprehensive assessment of renal masses is not available.
The exponential growth in use of cross-sectional imaging has resulted in increased detection of incidental renal masses, commonly appearing as small (≤ 4 cm) localized lesions at discovery [1, 2]. Because under such circumstances a CT protocol specifically designed for comprehensive assessment of a renal mass is often unavailable, inconclusive imaging features are depicted at the index CT examination [1–4]. For instance, depending on institutional policy, an unenhanced scan may not be available to ascertain whether a renal mass with attenuation higher than that of water (i.e., greater than 20 HU) is or is not enhancing. A considerable number of renal masses are, therefore, incompletely characterized, and additional tests or surveillance evaluations are performed [2–5], resulting in increased costs and increased use of medical resources, which may not necessarily be beneficial to the patient. It would be ideal if complete characterization of renal masses were to occur when they are incidentally found at the initial CT examination.
Dual-energy CT is being proposed as a useful technique for the imaging-based triage of incidentally discovered renal masses [6–11]. Thus far this approach has remained confined to the realm of research, because varying results have been obtained with different techniques [12], which are often only indirect measures of the photonic contribution from exogenous administration of iodinated contrast material.
A more recently devised dual-energy postprocessing approach is effective atomic number decomposition analysis. This technique, which is based on monochromatic data and knowledge of mass attenuation coefficients, enables elective mapping on a CT image of materials, such as iodine, that have high electron density and atomic number properties [13, 14]. Although a few studies have evaluated the characterization potential of this technique in phantoms or pilot populations enrolled for various disorders [15, 16], to our knowledge, no clinical validation studies have been conducted with a large series of patients. We hypothesized that such an approach, applied to contrast-enhanced data, may render the electronic signature of iodine uptake in renal masses directly at the atomic level.
The purpose of our study was to assess the diagnostic accuracy of effective atomic number maps reconstructed from dual-energy contrast-enhanced data for discriminating between nonenhancing renal cysts and enhancing masses.
Materials and Methods
This HIPAA-compliant, retrospective study was approved by our institutional review boards with a waiver of the requirement for informed consent.
Power Analysis of Study Sample Size
Thirty-two observations per group were required to achieve 90% power (1 minus type II error probability) with a 1% significance level (type I error probability) for testing the equality between mean effective atomic numbers of nonenhancing renal cysts and those of enhancing masses with a two-sample t test and an effect size equal to 1.
Patients
A computerized search was performed within the radiology information systems at two tertiary and quaternary academic health systems (set A, Duke University Hospital, 957 licensed beds; set B, University of Alabama at Birmingham Hospital, 1157 licensed beds) for patients undergoing dual-energy CT for renal mass evaluation from November 11, 2011, through June 24, 2015, for set A and from June 1, 2014, through November 30, 2015, for set B. Specifically, independent terms, including “renal mass,” “renal lesion,” “renal cell carcinoma,” “enhancement,” “renal cyst,” “Bosniak,” “hematuria,” “nephrographic,” and “dual-energy CT,” were adopted. This search was cross-referenced with a review of the thin-client dual-energy departmental server (Advantage Workstation 4.6, GE Healthcare) to identify available dual-energy CT studies. At the two sites, 158 (set A) and 148 (set B) subjects yielded positive matches for any of the search terms and had findings from at least one dual-energy CT nephrographic examination available. One hundred subjects (set A, n = 38; set B, n = 62) were excluded for the following reasons: no lesion was present (overall, n = 61; set A, n = 18; set B, n = 43); renal masses were smaller than 1 cm in diameter, to minimize occurrence of inaccurate quantitative measurements (overall, n = 25; set A, n = 15; set B, n = 10); renal masses displayed macroscopic fat (overall, n = 14; set A, n = 5; set B, n = 9).
Figure 1 is a flowchart of the study population accrual process, in accordance with Standards for Reporting of Diagnostic Accuracy recommendations and checklists [17–19]. The application of exclusion criteria to the initial target population resulted in the following final study cohorts: set A, 120 patients (mean age, 66 ± 12 [SD] years; range, 29–90 years; 77 men [mean age, 68.6 ± 10.7 years; range, 39–87 years]; 43 women [mean age, 63.8 ± 13.2 years; range, 29–90 years]); set B, 86 patients (mean age, 62 ± 13 years; range, 27–87 years; 51 men [mean age, 61.7 ± 13.3 years; range, 27–82 years]; 35 women [mean age, 66.9 ± 13.1 years; range, 28–87 years]).
A total of 206 (set A, n = 120; set B, n = 86) nonconsecutively registered adult patients (mean age, 64 ± 13 years; median, 66 years; range, 27–90 years) composed the combined final study cohort (Fig. 1). This included 128 men (mean age, 65.4 ± 12.4 years; median age, 67 years; range, 27–87 years) and 78 women (mean age, 65.3 ± 13.3 years; median, 66 years; range, 28–90 years). The mean patient effective diameter, calculated as the square root of the anteroposterior diameter multiplied by the transverse diameter, was 31 ± 4 cm (median, 31 cm; interquartile range, 29–46 cm).
Reference Standard Definition
Collection of reference standard information was performed by a board-certified abdominal radiologist (set A, 5 years of postfellowship experience in genitourinary imaging; set B, 22 years of postfellowship experience in genitourinary imaging), who was not involved in either the study design or subsequent dual-energy image analysis but had access privilege to all patients' imaging history and medical records. The reference standard was widely accepted state-of-the-art CT criteria based on a CT protocol designed for renal mass evaluation [1–3, 20–22]. These criteria were derived from comparisons of morphology, structure, attenuation, and presence of contrast enhancement (i.e., ≥ 20-HU increase in CT attenuation) between unenhanced and contrast-enhanced nephrographic images [1–3, 22]. For the nephrographic images, 70-keV monochromatic images were used because they are the standard of care adopted in clinical practice for PACS reading with the rapid kilovolt-age-switching dual-energy CT platform [23, 24] (see later, CT Technique). These image datasets have been found to yield qualitative and quantitative imaging characteristics comparable to those on traditional 120-kVp single-energy images [23, 24].
A renal mass was classified as solid enhancing if its attenuation increased by 20 HU or more between conventional nephrographic and unenhanced imaging [1–3, 22]. A diagnosis of nonenhancing cyst was assigned if a renal mass had an attenuation increase of less than 20 HU between conventional nephrographic and unenhanced imaging [1–3, 22]. In particular, nonenhancing cysts were called hypoattenuating if they had an attenuation of less than 20 HU on unenhanced images. They were called hyperattenuating if they had an attenuation of 20 HU or greater on unenhanced images [1–3, 22]. Furthermore, cystic renal lesions were categorized according to the Bosniak classification system (categories I, II, IIF, III, and IV) [1–3, 22]. In subsequent per-lesion analyses of accuracy for differentiation between nonenhancing and enhancing masses, Bosniak category III and IV cystic lesions were included in the enhancing masses subgroup, whereas Bosniak category I, II, and IIF were included in the nonenhancing masses subgroup [1–3, 22]. In detail, under strict adherence to the Bosniak classification system, category IIF lesions were considered nonenhancing because these lesions display perceptible but not measurable enhancement [25]. If a renal mass displayed unenhanced attenuation higher than that of water (i.e., > 20 HU) and had 10- to 20-HU enhancement (i.e., equivocal enhancement) and no further imaging confirmation or pathologic assessment was available, the mass was considerate indeterminate [1–3, 22].
According to the scheme of the composite reference standard [26], lesion classification was supported by information retrievable from patient medical records, including CT follow-up, confirmation with another imaging modality (i.e., ultrasound or MRI), or pathologic result whenever any or a combination of these was available. Imaging follow-up was defined as comparison between the index renal CT and subsequent or previous examinations (set A duration, 0.5–12 years; set B duration, 0.5–9 years). To minimize the confounding effect potentially deriving from multiple observations within the same subject (i.e., clustering effect), a maximum of three index renal masses per patient were selected. Mathematical accounting for potential clustering effect was also performed during subsequent data analysis (see later, Statistical Considerations). Details for identification of index renal masses in patients with multiple renal masses and lesion matching with subsequent dual-energy analyses are described in Appendix S1. (Appendixes S1 and S3, Figures S2 and S4, and Tables S5 and S6 can be viewed in the AJR electronic supplement to this article available at www.ajronline.org.)
CT Technique
Data acquisition—All scans of all patients were acquired with a 64-MDCT dual-energy CT scanner with rapid tube voltage switching and equipped with gemstone spectral imaging technology (Discovery CT 750 HD, GE Healthcare). All patients included in the final study cohort underwent a CT protocol for renal mass assessment, which incorporated unenhanced and contrast-enhanced nephrographic acquisitions from the top of both kidneys through the urinary bladder. All patients were positioned supine feet first on the scanning couch. After acquisition of anteroposterior and mediolateral digital scout radiographs, single-energy unenhanced images were acquired at 120 kV. This acquisition was followed by dual-energy (rapid switching of the tube voltage between 80 and 140 kV) nephrographic scanning through the kidneys at a fixed time delay of 90–110 seconds after the beginning of the IV contrast injection. Specific CT acquisition parameters of the dual-energy nephrographic acquisition are shown in Table 1.
| CT Parameter | Set A | Set B |
|---|---|---|
| Detector collimation (mm) | 64 × 0.625 | 64 × 0.625 |
| Tube voltage (kV) | Rapid switching 80/140 | Rapid switching 80/140 |
| FOV (cm) | 50 × 50 | 36 × 36 to 50 × 50 |
| Tube current-time product (mAs) | 640 | 370/600 |
| Gantry revolution time (s) | 0.6 | 0.7/0.8 |
| Acquisition mode | Single-source, helical | Single-source, helical |
| Pitch | 1.375 | 1.375 |
| Reconstruction section thickness (mm) | 2.50-5.00 | 2.50-5.00 |
| Reconstruction section interval (mm) | 2.50-5.00 | 2.50-5.00 |
| Matrix size | 512 × 512 | 512 × 512 |
| Reconstruction algorithm | 40% ASIR blended with 60% FBP | 40% ASIR blended with 60% FBP |
| Reconstruction kernel | Standard soft-tissue | Standard soft-tissue |
Note—ASIR = adaptive statistical iterative reconstruction, FBP = filtered back projection.
Patients in set A received 150 mL of an IV nonionic contrast medium with a concentration of 300 mg I/mL (iopamidol, Isovue 300, Bracco Diagnostics). The bolus of contrast medium was injected through an 18- to 20-gauge IV angiocatheter in the patient's antecubital fossa or forearm through a dual-chamber mechanical power injector (Empower CTA, Bracco Diagnostics) at a flow rate of 3 mL/s. For patients in set B, weight-based IV administration of nonionic contrast medium was performed with a concentration of 350 mg I/mL (iohexol, Omnipaque 350, GE Healthcare). For example, a typical 82-kg individual received 42 g I as 120-mL Omnipaque 350. The bolus of contrast medium was injected through an 18- to 20-gauge IV angiocatheter in the patient's antecubital fossa or forearm at 2.8–4.0 mL/s to achieve a 30-second fixed injection duration.
Data reconstruction—For each patient at both sites, gemstone spectral imaging data were reconstructed at 2.5-mm and 5.00-mm section collimation with a standard soft-tissue kernel by means of dual-energy decomposition algorithms operating in the projection-space domain (Table 1). All image datasets were networked to the departmental dual-energy CT web-based thin client (Advantage Workstation 4.6, GE Healthcare) and made available for effective atomic number (Zeff) decomposition analysis. With selection of the Zeff option, effective atomic number decomposition analysis allows encoding and display of tissue elemental composition based on their electron density and atomic number properties [13, 14, 27, 28]. According to a gray-scale scheme, tissues taking up the iodine contrast agent are selectively displayed, whereas non–iodine-containing structures are visualized as devoid of signal (Fig. S2). Quantitative values obtained from ROI measurements on dual-energy effective atomic number maps represent the meanatomic number for mixtures of materials variously assorted in vivo, rather than representing the absolute values one would expect from pure single chemical species [13, 14, 27, 28]. Technical details on synthesis of dual-energy effective atomic number maps are provided in Appendix S3.
Image Analysis of Effective Atomic Number Maps
Per each patient set, two observers (set A, 3 and 5 years of experience in genitourinary imaging; set B, 3 and 4 years of experience in genitourinary imaging) independently performed effective atomic number measurements on the gemstone spectral imaging viewer available with the dual-energy departmental thin-client system (Advantage Workstation 4.6, GE Healthcare). The readers were blinded to final lesion diagnosis, reference standard, and patient's imaging and medical history. For each lesion, manually defined oval or circular ROIs were drawn on axial effective atomic number dual-energy maps. For homogeneous lesions, the ROIs had to include as much of the lesion as possible and avoid the boundary of masses, where partial volume effects can occur. In cases of lesions with complex cystic texture (i.e., septa, wall thickening, or soft-tissue nodules in the context of a lesion with predominantly fluid-attenuation background) or heterogeneous architecture (i.e., coexistence of solid enhancing components and extensive areas of necrosis), the largest possible ROI was placed on the portion of the lesion that subjectively had the highest attenuation. If a lesion contained definite calcification or artifacts, these were excluded from the ROI measurements. Dual-energy data were recorded on electronic spreadsheets, made available to the observers only at the time of dual-energy data collection. The electronic spreadsheets had been populated previously with template headers, which depicted patient and lesion numbering, lesion size, and lesion location information.
Statistical Considerations
Numeric values of continuous variables were expressed as mean ± SD, and categoric variables are expressed as frequencies or percentages. Pearson product moment correlation coefficients (r) were used to assess interobserver agreement. The dual-energy effective atomic number sample means associated with nonenhancing cysts and enhancing renal masses were compared by two-sample t test. In cases in which the normality assumption was questionable, the Wilcoxon-Mann-Whitney test was also used to compare both groups. The Tukey honest significant difference method was used to compare mean values of dual-energy effective atomic numbers based on the Bosniak classification system. The combined measurements of dual-energy atomic numbers were analyzed by use of a linear model of the form Yijk = μ + β1si + β2dj + β3sdij + Eijk, where Yijk is the k-th observation with j-th diagnosis level (j is nonenhancing cyst or enhancing renal mass) from the i-th set (i = A or B), μ is an intercept term, si is the set effect from different set of patients, dj is the diagnosis effect, sdij is the site-by-diagnosis interaction effect, βi for i = 1, 2, 3 are the unknown regression coefficients, and Eij is the random measurement error, assumed to be normally distributed with zero mean and constant variance, denoted as Eij ~ N(0,σ2). The model was fit by the least-squares method.
The ROC curve was constructed by use of the Youden index (J = sensitivity + specificity – 1), which was calculated at all points of the ROC curve. The maximum value of this index was used as a criterion for selecting the optimal threshold. Accuracy, sensitivity, specificity, positive predictive value, and negative predictive value were calculated on the basis of the optimal threshold.
To assess the reproducibility of the optimal threshold calculated, the summary statistics accuracy, sensitivity, specificity, positive predictive value, and negative predictive value for detection of enhancing renal masses were calculated in one set by use of the optimal threshold obtained from the other set.
The accuracy for lesion diagnosis, which is a binary outcome, was modeled by use of a mixed effects multivariate logistic regression approach. The log odds of accuracy were expressed in terms of a linear combination of predictors as follows:
where pijkl is the probability given the covariates that the j-th lesion from i-th set made by the k-th reader on the l-th patient is correctly diagnosed. The model contained an intercept term (μ) along with effect from different sets of patients (seti), patient size (sizeijkl), lesion size (lesionijkl), size of ROI (ROIijkl), unenhanced attenuation (pre ijkl), contrast-enhanced nephrographic attenuation (postijkl), reader effect (readerk), and a patient random effect to account for the clustering effect of lesions within patients (patienti ~ N(0,σ2p); βi for i = 1,. 7 are the regression coefficients. The model was fit via the maximum likelihood method.
The threshold for assessing statistical significance was set at p < 0.05 in all cases. The statistical packages used were R version 3.2.1 (R Development Core Team, R Foundation) and SAS version 9.4 (SAS Institute).
Results
Renal Masses
There were a total of 350 renal masses (set A, 188; set B, 162) in 206 patients. The mean number of index renal lesions per patient was 1.6 (set A, 1.6; set B, 1.8). One hundred fourteen patients had one lesion (set A, 74; set B, 40), 40 had two lesions (set A, 24; set B, 16), and 52 had three lesions (set A, 22; set B, 30). Table 2 summarizes lesion classification and imaging features according to the reference standard. Two hundred sixty-one lesions were classified as nonenhancing cysts (Bosniak category I, 186 [set A, 105; set B, 81]; Bosniak category II, 66 [set A, 37; set B, 29]; Bosniak category IIF, nine [set A, five; set B, four), 87 as enhancing masses (Bosniak category III, seven [set A, seven; set B, zero]; Bosniak category IV, one [set A, one; set B, zero]; solid, 79 [set A, 31; set B, 48]), and two as indeterminate. Among the 57 renal masses with histopathologic assessment (nephrectomy, 45; percutaneous core biopsy, 12) at either institution, the following diagnostic outcomes were obtained: clear cell renal cell carcinoma (RCC), 23 (set A, nine; set B, 14); papillary RCC, 15 (set A, one; set B, 14); chromophobe RCC, two (set A, one; set B, one); oncocytoma, 11 (set A, three; set B, eight); lipid-poor angiomyolipoma, one (set A, one; set B, zero); juxtaglomerular cell tumor, one (set A, zero; set B, one); chronic infected cyst, one (set A, one; set B, zero). In addition, in set A, two renal lesions in two patients, who underwent radical nephrectomy for removal of another large solid renal mass, proved to be simple cysts. In the same set, percutaneous biopsy of a solid enhancing lesion yielded inconclusive results.
| Diagnosis | No. of Lesions | Size (cm) | Attenuation (HU) | Lost to Follow-Up | CT Follow-Up | Confirmation With Ultrasound or MRI | Pathologic Confirmation | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Unenhanced | Nephrographic | Total | Set A | Set B | Total | Set A | Set B | Total | Set A | Set B | Total | Set A | Set B | |||
| Nonenhancing | ||||||||||||||||
| Bosniak I | 186 | 2.8 ± 1.8 | 9.2 ± 9 | 17.4 ± 14.9 | 121 | 61 | 60 | 18 | 3 | 15 | 45 | 39 | 6 | 2 | 2 | 0 |
| Bosniak II | 66 | 2.5 ± 1.8 | 36.1 ± 25.3 | 42.3 ± 25.5 | 36 | 16 | 20 | 10 | 4 | 6 | 20 | 17 | 3 | — | — | — |
| Bosniak IIF | 9 | 2.2 ± 1.2 | 31 ± 34 | 45.9 ± 44.7 | 1 | 1 | 0 | 6 | 3 | 3 | 2 | 1 | 1 | — | — | — |
| Enhancing | ||||||||||||||||
| Bosniak III | 7 | 2.6 ± 0.9 | 28.7 ± 10.3 | 98.5 ± 38.6 | 3 | 3 | 0 | — | — | — | 1 | 1 | 0 | 3 | 3 | 0 |
| Bosniak IV | 1 | 5 ± 0 | 34 ± 0 | 76 ± 0 | 1 | 1 | 0 | — | — | — | ||||||
| Solid | 79 | 3.4 ± 2.5 | 32.8 ± 13.2 | 101.7 ± 32.8 | 7 | 6 | 1 | 12 | 4 | 8 | 8 | 7 | 1 | 52 | 14 | 38 |
Note—Dash (—) indicates lack of lesion for that given category. Values are number of lesions or mean ± SD. Two of seven (28.6%) Bosniak III complex cystic renal masses were pathologically proven to be clear cell renal cell carcinoma. In set A, two patients with two lesions (not shown), which had unenhanced attenuation greater than that of water (i.e., > 20 HU) and equivocal enhancement (lesion 1 size, 1.8 cm; unenhanced attenuation, 58 HU; nephrographic attenuation, 73 HU; lesion 2 size, 1.9 cm; unenhanced attenuation, 52 HU; nephrographic attenuation, 69 HU). These lesions were called indeterminate because assessment with another imaging modality or pathologic analysis was unavailable. According to Israel and Silverman as well as Israel and Bosniak [2, 22, 25], only renal lesions with measurable contrast enhancement (increase in attenuation by or more than 20 HU between nephrographic and unenhanced images) were classified as enhancing. Imaging follow-up was defined as comparison between the index renal CT examination and subsequent or previous examinations (set A duration, 0.5–12 years; set B duration, 0.5–9 years).
The two indeterminate lesions were encountered in two patients in set A. Both had unenhanced attenuation greater than that of water and equivocal enhancement (lesion 1 size, 1.8 cm; unenhanced attenuation, 58 HU; nephrographic attenuation, 73 HU; lesion 2 size, 1.9 cm; unenhanced attenuation, 52 HU; nephrographic attenuation, 69 HU) and were classified as indeterminate because no imaging follow-up, assessment with another imaging modality, or pathologic analysis was available. To avoid inaccurate classification or virtually improper clinical management, these two lesions were excluded from diagnostic accuracy analyses.
Effective Atomic Number Values
There were significant differences in mean dual-energy effective atomic numbers between nonenhancing renal cysts and enhancing masses (set A, 8.19 ± 0.46 Zeff for cysts vs 9.59 ± 0.75 Zeff for masses; set B, 8.05 ± 0.34 Zeff vs 9.19 ± 0.68 Zeff; combined, 8.13 ± 0.42 Zeff vs 9.37 ± 0.74 Zeff) (p < 0.0001 for all comparisons) (Table 3 and Figs. 2–4). There was some degree of overlap in distribution among the different diagnostic subgroups (Figs. 5 and 6). The two indeterminate renal masses had effective atomic numbers of 8.41 and 8.97 Zeff (Fig. S4), falling in the lower end of the range of effective atomic numbers identified in the subset of enhancing renal masses.
| Diagnosis | Effective Atomic Number Values (Zeff) | Size of ROI (mm2) | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Set A | Set B | Combined | Set A | Set B | Combined | |||||||
| Mean ± SD | Median | IQR | Mean | Median | IQR | Mean | Median | IQR | ||||
| Nonenhancing | ||||||||||||
| Bosniak I | 8.09 ± 0.3 | 8.04 | 7.94-8.19 | 8.00 ± 0.3 | 8.01 | 7.80-8.14 | 8.05 ± 0.3 | 8.03 | 7.85-8.17 | 200±307 | 386 ± 788 | 282±578 |
| Bosniak II | 8.3 ± 0.4 | 8.2 | 8.02-8.57 | 8.12 ± 0.4 | 8.05 | 7.94-8.3 | 8.24 ± 0.4 | 8.12 | 7.99-8.4 | 204±697 | 187±305 | 199±562 |
| Bosniak IIF | 9.33 ± 0.9 | 9.04 | 8.31-9.91 | 8.44 ± 0.8 | 8.16 | 7.82-8.78 | 8.93 ± 0.9 | 8.9 | 8.15-9.68 | 27 ± 40 | 108 ± 76 | 60 ± 69 |
| Enhancing | ||||||||||||
| Bosniak III | 9.09 ± 0.8 | 9.3 | 8.39-9.47 | NA | NA | NA | 9.09 ± 0.8 | 9.28 | 8.39-9.47 | 20 ± 23 | NA | 20 ± 23 |
| Bosniak IV | 9.36 ± 0.1 | 9.36 | 9.31-9.41 | NA | NA | NA | 9.36 ± 0.1 | 9.36 | 9.31-9.41 | 8 ± 0 | NA | 8 ± 0 |
| Solid | 9.71 ± 0.7 | 9.84 | 9.33-10.2 | 9.19 ± 0.7 | 9.07 | 8.71-9.8 | 9.39 ± 0.7 | 9.43 | 8.8-10.07 | 100±288 | 265±226 | 200±264 |
Note—Cystic lesions categorized by Bosniak classification (category I, II, IIF, III, and IV lesions); there were no Bosniak category III or IV cystic lesions in set B. The two indeterminate renal lesions (not shown), which had unenhanced attenuation greater than that of water and equivocal enhancement, yielded effective atomic number values of 8.41 and 8.97 Zeff. IQR = interquartile range, NA = not applicable.
We observed significant differences in effective atomic numbers when comparing Bosniak I (8.05 ± 0.3 Zeff) and Bosniak II (8.24 ± 0.4 Zeff) nonenhancing cysts to enhancing renal masses proven to represent clear cell (9.64 ± 0.7 Zeff) or papillary (8.63 ± 0.4 Zeff) RCC at pathologic analysis (p < 0.0001 for all comparisons) (Fig. 7 and Table S5). Among cases of enhancing renal masses with a final histopathologic diagnosis, there were significant differences in dual-energy effective atomic numbers between RCC (9.24 ± 0.76 Zeff) and other types of solid enhancing masses, such as oncocytomas (9.71 ± 0.73 Zeff) (p = 0.006). Importantly, there were significant differences in dual-energy effective atomic numbers in the comparison of clear cell (9.9 ± 0.68 Zeff) and non–clear cell (8.64 ± 0.48 Zeff) RCCs (p < 0.0001). No significant difference in dual-energy effective atomic numbers was observed between clear cell RCC (9.64 ± 0.68 Zeff) and solid enhancing tumors other than RCC (9.71 ± 0.5 Zeff) (p < 0.0001), although solid renal tumors other than RCC (e.g., oncocytomas) (8.7 ± 0.48 Zeff) were significantly different from nonclear cell subtypes of RCC (9.71 ± 0.5 Zeff) (p < 0.0001).
Interobserver Agreement
There was excellent agreement between the two investigators who independently performed dual-energy effective atomic number analyses (set A, r = 0.955 [95% CI, 0.941–0.966]; set B, r = 0.941 [95% CI, 0.921–0.957]).
Independent and Combined Diagnostic Accuracy
Set A—ROC analysis of data from set A showed that 8.91 Zeff was the optimal threshold for discerning nonenhancing renal cysts from enhancing masses. A lesion effective atomic number threshold of 8.92 Zeff yielded an AUC of 0.922 (Fig. 8) with sensitivity of 85% (66/78), specificity of 93% (274/294), and overall diagnostic accuracy of 91% (340/372) (Table 4).
| Location | Sensitivity (%) | Specificity (%) | Positive Predictive Value (%) | Negative Predictive Value (%) | Overall Accuracy (%) | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Value | 95% CI | Value | 95% CI | Value | 95% CI | Value | 95% CI | Value | 95% CI | |
| Set A | 85 (66/78) | 75-92 | 93 (274/294) | 90-96 | 77 (66/86) | 66-85 | 96 (274/286) | 93-98 | 91 (340/372) | 88.1-94 |
| Set B | 92 (88/96) | 84-96 | 90 (205/228) | 85-93 | 79 (88/111) | 70-87 | 96 (205/213) | 93-98 | 90 (293/324) | 86.7-93.4 |
| Combined (set A plus B) | 91 (158/174) | 85.5-94.7 | 85 (445/522) | 81.9-88.2 | 67 (158/235) | 60.8-73.2 | 97 (445/461) | 94.4-98 | 87 (603/696) | 83.9-89.1 |
| Threshold from set A applied to set B | 61 (59/96) | 51-71.2 | 97 (222/228) | 51-71.2 | 91 (59/65) | 81-96.5 | 86 (222/259) | 81-89.7 | 87 (281/324) | 82.5-90.2 |
| Threshold from set B applied to set A | 90 (70/78) | 80.8-95.5 | 82 (240/294) | 76.7-85.9 | 57 (70/124) | 47.3-65.3 | 97 (240/248) | 93.7-98.6 | 83 (310/372) | 79.1-87 |
Note—Numbers used to calculate the percentages are in parentheses and refer to the number of observations from two readers in each set.
Set B—ROC analysis of data from site B indicated that 8.36 Zeff was the optimal diagnostic threshold, which yielded an AUC of 0.935 (Fig. 8) with sensitivity of 92% (88/96), specificity of 90% (205/228), and overall diagnostic accuracy of 90% (293/324) (Table 4).
Combined data from sets A and B—When the two datasets from the two sites were blended, the optimal diagnostic threshold was 8.36 Zeff which yielded an AUC of 0.921 (Fig. 8) with sensitivity of 91% (158/174), specificity of 85.2% (445/522), and overall diagnostic accuracy of 87% (603/696) (Table 4). Among cases with a final histopathologic diagnosis, overall diagnostic accuracy values for discrimination among different solid renal mass types were as follows: RCC versus other solid renal tumors, 59.4% (63/106); solid renal tumors other than RCC versus non–clear cell subtypes of RCC, 88.3% (53/60); clear cell RCC versus non–clear cell subtypes of RCC, 82.5% (66/80); clear cell RCC versus solid renal tumors other than RCC, 58.3% (42/72).
Diagnostic Accuracy of Thresholds From One Set for Diagnosis in the Other Set
Threshold from set A applied to set B—When the optimal diagnostic threshold from set A (8.92 Zeff) was applied to data from set B, we observed sensitivity of 61% (59/96), specificity of 97% (222/228), and overall diagnostic accuracy of 87% (281/324) (Table 4).
Threshold from set B applied to set A—When the optimal diagnostic threshold from set B (8.36 Zeff) was applied to data from set A, we found sensitivity of 90% (70/78), specificity of 82% (240/294), and overall diagnostic accuracy of 83% (310/372).
Diagnostic Accuracy Comparisons
There was no significant difference in diagnostic accuracy between the two sets (set A, 91%; set B, 90%) with use of their own thresholds (set A, 8.92 Zeff; set B, 8.36 Zeff) (p = 0.657). There were 4.1% and 8.8% decreases in diagnostic accuracy when the optimal diagnostic threshold from set A was applied to set B (from 90.4% to 86.7%) and vice versa (from 91.3% to 83.3%).
Mixed-Effects Multivariate Logistic Regression Analysis of Factors Affecting Diagnostic Accuracy
When variables potentially affecting the diagnostic accuracy of dual-energy effective atomic number maps were evaluated, patient size (p = 0.0153), size of ROI (p = 0.0037), and unenhanced lesion attenuation (p = 0.0172) were found to have a significant effect (Table S6). By comparison, different patient set, lesion size, nephrographic lesion attenuation, and influence of different readers did not exert significant effects on diagnostic accuracy.
Discussion
Our study results show that effective atomic number maps reconstructed from dual-energy contrast-enhanced data can be used to discriminate nonenhancing renal cysts, including hyperattenuating cysts, from enhancing masses. They also hold potential for noninvasive differentiation of types of solid renal tumors. Our data indicate, however, that dual-energy effective atomic number analysis is not immune to misclassification, which occurred in 14.6% (51/348) of renal masses in our series. Notably, 16.1% (42/261) of nonenhancing cystic lesions were over-classified (Bosniak I, 8.6% [16/186]; Bosniak II, 30.3% [20/66]; Bosniak IIF, 66.7% [6/9]), and 10.3% (9/87) of enhancing masses were underclassified (Bosniak III, 28.6% [2/7]; solid masses, 8.9% [7/79]). Among solid masses that turned out to be RCC at pathologic analysis, 8.7% (2/23) of cases were false-negative for clear cell RCC and 13.3% (2/15) for papillary RCC.
Our false-positive and false-negative cases may be explained by the material misclassification that can occur in the attempt to discriminate cationic metal salts with high electron density and atomic number, such as iron- and calcium-based salts—which are consistently observed within complex cystic renal lesions [29]—from low concentrations of iodine in minimally enhancing renal lesions [30, 31]. For example, dual-energy effective atomic number maps from contrast-enhanced data may have limited utility for differentiating faint linear hyperdensities commonly observed in complex cystic renal masses [32, 33], particularly when such features are seen within the dependent portions of the lesion, where they can represent either colloidal suspension of precipitated calcium salts (milk of calcium) or small quantities of iodine traveling along the neovascularized hair-thin septa [32–34]. This may explain the overclassification of Bosniak IIF complex cysts, lesions that traditionally exhibit perceptible but not measurable enhancement of hair-thin smooth septa or walls at conventional CT assessment [25].
On the other hand, because effective atomic number analysis may be a more direct measure of iodine presence within a lesion (i.e., depiction of electron changes at an atomic level) than attenuation measurements (i.e., normalized scale from linear attenuation coefficients), the electronic contribution from subtle amounts of iodine within a minimally vascularized renal mass may be amplified on dual-energy effective atomic number maps. This hypothesis is substantiated by the significant differences in effective atomic number values we observed in the direct comparison between simple or hyperattenuating renal cysts and pathologically proven cases of clear cell or papillary RCC. The differentiation between a renal cyst with greater than 20-HU attenuation on contrast-enhanced images (i.e., due to the pseudoenhancement phenomenon or proteinaceous or hemorrhagic content) and a hypovascular renal neoplasm (e.g., papillary RCC) is a common conundrum in clinical practice when CT protocols not designed for renal mass evaluation are used (e.g., no corresponding unenhanced series has been obtained), as in the emergency department or for routine assessment of abdominal pain [1, 2]. In the evaluation of a patient with a renal mass serendipitously found in a dual-energy examination, effective atomic number maps may be of diagnostic aid, potentially obviating further imaging evaluation and hence reducing costs, alleviating patient anxiety, and possibly avoiding additional radiation exposure.
Our study results also indicate that, to some extent, dual-energy effective atomic number maps potentially bear clinical relevance for the noninvasive differentiation of types of solid renal tumors. Our data suggest that indolent solid renal tumors (e.g., renal oncocytomas) can be identified with greater than 90% sensitivity. The specificity and overall accuracy rates for differentiating them from clear cell RCC were substantially lower (46.2% and 58.3%). However, the high sensitivity, specificity, and accuracy observed in discriminating indolent solid renal tumors (e.g., renal oncocytomas) from the non–clear cell RCC subset (e.g., papillary histotype) (92.3%, 85.3%, and 88.3%) and in differentiating clear cell and non–clear cell RCCs (80.4%, 85.3%, and 82.5%) hold promise for guiding the CT-based treatment of patients with solid renal tumors and may be of incremental diagnostic value, especially in patients whose condition prevents them from being surgical candidates or in patients for whom percutaneous biopsy is contraindicated.
We observed 4.1% and 8.8% decreases in diagnostic accuracy when the optimal diagnostic threshold from set A was applied to set B and vice versa. It is conceivable that the diverse proportions in which different renal mass subtypes were distributed in each set could have produced such an effect. For example, there were more complex cystic renal masses in set A (60.7% [51/84]) than in set B (39.3% [33/84]). Apart from Bosniak II hyperattenuating cysts (i.e., lesions that are avascular [2, 22]), there were only four Bosniak IIF cysts and no Bosniak III or IV cystic lesions in set B as opposed to five Bosniak IIF, seven Bosniak III, and one Bosniak IV lesions in set A. A larger number of pathologically proven cases of RCC were seen in set B (60.4% [29/48]) than in set A (28.2% [11/39]). Among these cases were 14 pathologically proven cases of papillary RCC in set B versus only one of the same histotype in set A. These cases together constitute a variety of renal masses that exhibit a wide spectrum of degrees of vascularization [33].
We therefore postulate that the unbalanced allotment of such lesion subtypes may explain differences in optimal diagnostic threshold, resulting in the considerable decline in sensitivity observed when diagnostic rules from site A were applied to set B. Furthermore, adoption of different contrast injection strategies (i.e., fixed amount of contrast material in set A vs weight-based formula in set B) might have contributed to observed differences in diagnostic accuracy between the two populations.
Our multivariate linear regression analysis showed that only patient size, size of ROI, and unenhanced lesion attenuation had significant effects on diagnostic accuracy of the newly proposed dual-energy technique. The effect of patient size can be explained by sustained beam-hardening and photon starvation phenomena, a complex interplay that affects precision and accuracy of dual-energy quantification with increasing patient body sizes [12, 30, 31]. Long recognized among key determinants in the quantitative assessment of renal masses with CT [3, 4], the impact of ROI size on diagnostic accuracy was not surprising. By comparison, the significant influence of unenhanced lesion attenuation was an unexpected effect, especially if one considers that it is derived from a CT dataset independent from contrast-enhanced data, from which dual-energy effective atomic number maps are reconstructed. Because CT attenuation information was part of our reference standard, we could not perform a direct assessment of whether and to what extent it correlates with effective atomic number values. Future studies are needed to elucidate this observation.
The major limitation of our study was that, owing to its retrospective nature, there was no prospective, standardized strategy in place for consistent follow-up; 54.8% (192/350) of renal masses in our series did not have further clinical or pathologic confirmation. For instance, we could not determine the outcome of the two lesions in our series displaying unenhanced attenuation greater than that of water and having equivocal levels of enhancement. On the other hand, the other 99.0% (190/192) of lesions without further imaging assessment or pathologic confirmation had a conclusive diagnosis of nonenhancing cyst or enhancing mass based on typical imaging features on unenhanced and nephrographic CT images [1–3, 22, 25]. This subset of renal masses (e.g., Bosniak I or Bosniak II cystic lesions) requires no further imaging evaluation [1–3, 22, 25, 35].
Our primary endpoint was to discriminate between nonenhancing and enhancing renal masses—an exclusively imaging-based concept for which there is no reference standard, to our knowledge, other than a CT protocol designed for renal mass evaluation based on unenhanced and nephrographic acquisitions [1–3, 22]. Because characterization of renal masses as benign or malignant was beyond the scope of this work, it could be said that the unavailability of histopathologic assessment for 37% (32/87) of enhancing lesions did not affect our evaluation of diagnostic accuracy for differentiating nonenhancing cysts and enhancing masses. We acknowledge that the lack of final pathologic result in the same cases limited our analyses of accuracy for discriminating among different subtypes of solid renal tumors, further compounded by small sample size. Future investigation in a larger series of pathologically proven cases is warranted. Another important limitation of our study was that we did not entirely reproduce the typical clinical scenario of a renal mass incidentally appreciated on a portal venous scan of the abdomen. To simulate this condition, it would be necessary to prospectively incorporate an additional, earlier contrast-enhanced acquisition (i.e., 60–70 seconds) into the unenhanced and contrast-enhanced nephrographic series, which in most cases already yields complete renal mass characterization [1–3, 22].
We used a monochromatic series at 70 keV instead of a separate single-energy series acquired at 120 kVp. The former has, however, been found to have both qualitative and quantitative image characteristics comparable to the latter [23, 24]. For these reasons, the prospective addition of another scan to the clinical acquisition protocol in this study would be technically impractical and ethically unjustified. Finally, we excluded renal masses smaller than 1 cm in diameter; therefore, our data may not be applicable to this renal mass subset.
Our data suggest that effective atomic number maps reconstructed from dual-energy contrast-enhanced data can be used to discriminate nonenhancing renal cysts, including hyperattenuating cysts, from enhancing masses. This method holds promise for noninvasive differentiation of subtypes of enhancing renal tumors, albeit with less accuracy. These findings may be of clinical relevance for practicing radiologists to facilitate the imaging workup of renal masses incidentally found at CT examinations not prospectively designed for renal mass evaluation.
Footnotes
At the time this investigation was conducted, A. Mileto had a research fellowship sponsored by Bracco Diagnostics. D. E. Morgan has served as a consultant to create educational materials for GE Healthcare.
WEB
This is a web exclusive article.
Supplemental Content
File (10_16_17325_suppdata_s02_thumb.jpg)
- Download
- 578.10 KB
File (10_16_17325_suppdata_s04_thumb.jpg)
- Download
- 34.60 KB
File (adobe_pdf_file_icon_32x32.png)
- Download
- .96 KB
File (10_16_17325_suppdata_s01.pdf)
- Download
- 435.20 KB
File (10_16_17325_suppdata_s02.jpg)
- Download
- 319.47 KB
File (10_16_17325_suppdata_s03.pdf)
- Download
- 501.49 KB
File (10_16_17325_suppdata_s04.jpg)
- Download
- 224.82 KB
File (10_16_17325_suppdata_s05.pdf)
- Download
- 407.89 KB
File (10_16_17325_suppdata_s06.pdf)
- Download
- 407.66 KB
References
1.
Silverman SG, Israel GM, Trinh QD. Incompletely characterized incidental renal masses: emerging data support conservative management. Radiology 2015; 275:28–42
2.
Israel GM, Silverman SG. The incidental renal mass. Radiol Clin North Am 2011; 49:369–383
3.
Israel GM, Bosniak MA. Pitfalls in renal mass evaluation and how to avoid them. RadioGraphics 2008; 28:1325–1338
4.
Birnbaum BA, Jacobs JE, Ramchandani P. Multi-phasic renal CT: comparison of renal mass enhancement during the corticomedullary and nephrographic phases. Radiology 1996; 200:753–758
5.
Yuh BI, Cohan RH. Different phases of renal enhancement: role in detecting and characterizing renal masses during helical CT. AJR 1999; 173:747–755
6.
Graser A, Becker CR, Staehler M, et al. Single-phase dual-energy CT allows for characterisation of renal masses as benign or malignant. Invest Radiol 2010; 45:399–405
7.
Chandarana H, Megibow AJ, Cohen BA, et al. Iodine quantification with dual-energy CT: phantom study and preliminary experience with renal masses. AJR 2011; 196:[web]W693–W700
8.
Mileto A, Marin D, Ramirez-Giraldo JC, et al. Accuracy of contrast-enhanced dual-energy MDCT for the assessment of iodine uptake in renal lesions. AJR 2014; 202:[web]W466–W474
9.
Kaza RK, Caoili EM, Cohan RH, Platt JF. Distinguishing enhancing from nonenhancing renal lesions with fast kilovoltage-switching dual-energy CT. AJR 2011; 197:1375–1381
10.
Mileto A, Sofue K, Marin D. Imaging the renal lesion with dual-energy multidetector CT and multi-energy applications in clinical practice: what can it truly do for you? Eur Radiol 2016; 26:3677–3690
11.
Kaza RK, Platt JF. Renal applications of dual-energy CT. Abdom Radiol 2016; 41:1122–1132
12.
Mileto A, Barina A, Marin D, et al. Virtual monochromatic images from dual-energy multidetector CT: variance in CT numbers from the same lesion between single-source projection-based and dual-source image-based implementations. Radiology 2016; 279:269–277
13.
Marshall WH Jr, Alvarez R, Macovski A, Healy J, Zatz LM. Dual kilovoltage at computed tomography: a prereconstruction method for estimation of effective atomic number and electron density. Neuroradiology 1978; 16:605–606
14.
McCollough CH, Leng S, Yu L, Fletcher JG. Dual- and multi-energy CT: principles, technical approaches, and clinical applications. Radiology 2015; 276:637–653
15.
Tatsugami F, Higaki T, Kiguchi M, et al. Measurement of electron density and effective atomic number by dual-energy scan using a 320-detector computed tomography scanner with raw databased analysis: a phantom study. J Comput Assist Tomogr 2014; 38:824–827
16.
Ju Y, Liu A, Dong Y, et al. The value of nonenhanced single-source dual-energy CT for differentiating metastases from adenoma in adrenal glands. Acad Radiol 2015; 22:834–839
17.
Bossuyt PM, Reitsma JB, Bruns DE, et al. STARD 2015: An updated list of essential items for reporting diagnostic accuracy studies. Radiology 2015; 277:826–832
18.
Korevaar DA, Wang J, van Enst WA, et al. Reporting diagnostic accuracy studies: some improvements after 10 years of STARD. Radiology 2015; 274:781–789
19.
Levine D, Kressel HY. Radiology 2016: The care and scientific rigor used to process and evaluate original research manuscripts for publication. Radiology 2016; 278:6–10
20.
Taouli B, Thakur RK, Mannelli L, et al. Renal lesions: characterization with diffusion-weighted imaging versus contrast-enhanced MR imaging. Radiology 2009; 251:398–407
21.
Kim S, Jain M, Harris AB, et al. T1 hyperintense renal lesions: characterization with diffusion-weighted MR imaging versus contrast-enhanced MR imaging. Radiology 2009; 251:796–807
22.
Israel GM, Bosniak MA. How I do it: evaluating renal masses. Radiology 2005; 236:441–450
23.
Matsumoto K, Jinzaki M, Tanami Y, Ueno A, Yamada M, Kuribayashi S. Virtual monochromatic spectral imaging with fast kilovoltage switching: improved image quality as compared with that obtained with conventional 120-kVp CT. Radiology 2011; 259:257–262
24.
Yu L, Christner JA, Leng S, Wang J, Fletcher JG, McCollough CH. Virtual monochromatic imaging in dual-source dual-energy CT: radiation dose and image quality. Med Phys 2011; 38:6371–6379
25.
Israel GM, Bosniak MA. Follow-up CT of moderately complex cystic lesions of the kidney (Bosniak category IIF). AJR 2003; 181:627–633
26.
Arrivé L, Renard R, Carrat F, et al. A scale of methodological quality for clinical studies of radiologic examinations. Radiology 2000; 217:69–74
27.
Landry G, Seco J, Gaudreault M, Verhaegen F. Deriving effective atomic numbers from DECT based on a parameterization of the ratio of high and low linear attenuation coefficients. Phys Med Biol 2013; 58:6851–6866
28.
Goodsitt MM, Christodoulou EG, Larson SC. Accuracies of the synthesized monochromatic CT numbers and effective atomic numbers obtained with a rapid kVp switching dual energy CT scanner. Med Phys 2011; 38:2222–2232
29.
Boll DT, Patil NA, Paulson EK, et al. Focal cystic high-attenuation lesions: characterization in renal phantom by using photon-counting spectral CT—improved differentiation of lesion composition. Radiology 2010; 254:270–276
30.
Primak AN, Ramirez Giraldo JC, Liu X, Yu L, McCollough CH. Improved dual-energy material discrimination for dual-source CT by means of additional spectral filtration. Med Phys 2009; 36:1359–1369
31.
Krauss B, Grant KL, Schmidt BT, Flohr TG. The importance of spectral separation: an assessment of dual-energy spectral separation for quantitative ability and dose efficiency. Invest Radiol 2015; 50:114–118
32.
Hindman NM, Hecht EM, Bosniak MA. Follow-up for Bosniak category 2F cystic renal lesions. Radiology 2014; 272:757–766
33.
Hartman DS, Choyke PL, Hartman MS. From the RSNA refresher courses: a practical approach to the cystic renal mass. RadioGraphics 2004; 24(suppl 1):S101–S115
34.
Ascenti G, Mazziotti S, Mazziotti S, et al. Complex cystic renal masses: characterization with contrast-enhanced US. Radiology 2007; 243:158–165
35.
Smith AD, Remer EM, Cox KL, et al. Bosniak category IIF and III cystic renal lesions: outcomes and associations. Radiology 2012; 262:152–160
FOR YOUR INFORMATION
The data supplement accompanying this web exclusive article can be viewed by clicking “Supplemental” at the top of the article.
Information & Authors
Information
Published In
Presented at
Based on a presentation at the Society of Abdominal Radiology 2017 annual meeting, Hollywood, FL.
Copyright
© American Roentgen Ray Society.
History
Submitted: September 3, 2016
Accepted: February 21, 2017
First published: July 13, 2017
Keywords
Authors
Metrics & Citations
Metrics
Citations
Export Citations
To download the citation to this article, select your reference manager software.