Small (< 4 cm) Renal Masses: Differentiation of Angiomyolipoma Without Visible Fat From Renal Cell Carcinoma Using Unenhanced and Contrast-Enhanced CT

Transcription

1 Genitourinary Imaging Original Research Takahashi et al. CT of Small Renal Masses Genitourinary Imaging Original Research Naoki Takahashi 1 Shuai Leng 1 Kazuhiro Kitajima 1,2 Daniel Gomez-Cardona 1,3 Prabin Thapa 4 Rickey E. Carter 4 Bradley C. Leibovich 5 Kewalee Sasiwimonphan 1,6 Kohei Sasaguri 1,7 Akira Kawashima 1 Takahashi N, Leng S, Kitajima K, et al. Keywords: angiomyolipoma, CT, kidney, renal cell carcinoma, texture analysis DOI: /AJR Received November 26, 2014; accepted after revision January 31, Department of Radiology, Mayo Clinic, 200 First St SW, Rochester, MN Address correspondence to N. Takahashi (takahashi.naoki@mayo.edu). 2 Present address: Department of Radiology, Kobe University, Faculty of Medicine, Hyogo, Japan. 3 Present address: Department of Medical Physics, School of Medicine, University of Wisconsin, Madison, WI. 4 Department of Health Sciences Research, Mayo Clinic, Rochester, MN. 5 Department of Urology, Mayo Clinic, Rochester, MN. 6 Present address: Department of Radiology, Faculty of Medicine, King Chulalongkorn Memorial Hospital, Chulalongkorn University, Bangkok, Thailand. 7 Present address: Department of Radiology, Faculty of Medicine, Saga University, Saga, Japan. AJR 2015; 205: X/15/ American Roentgen Ray Society Small (< 4 cm) Renal Masses: Differentiation of Angiomyolipoma Without Visible Fat From Renal Cell Carcinoma Using Unenhanced and Contrast-Enhanced CT OBJECTIVE. The purpose of this study was to evaluate if small (< 4 cm) angiomyolipoma without visible fat can be differentiated from renal cell carcinoma (RCC) using contrastenhanced CT alone and using unenhanced and contrast-enhanced CT in combination. MATERIALS AND METHODS. Twenty-three patients with 24 angiomyolipomas without visible fat and 130 patients with 148 RCCs underwent unenhanced and contrast-enhanced CT. Demographic data and size, shape, CT attenuation, and heterogeneity (entropy and subjective score) of the renal mass on unenhanced CT and contrast-enhanced CT were recorded. Multivariate logistic regression models were constructed for parameters obtained by contrastenhanced CT alone and by both unenhanced and contrast-enhanced CT. Demographic data and size and shape of renal mass were used in each model. Sensitivity and specificity were calculated. RESULTS. Logistic regression model from contrast-enhanced CT data included sex, percentage of exophytic growth, entropy, and CT attenuation on contrast-enhanced CT. Model from both unenhanced and contrast-enhanced CT data included age, sex, short-axis diameter, percentage of exophytic growth, lesion-to-kidney CT attenuation difference on unenhanced CT, and CT attenuation on contrast-enhanced CT. The contrast-enhanced CT based model and combined unenhanced and contrast-enhanced CT based model differentiated angiomyolipoma from RCC with sensitivity and specificity of 42% and 98% versus 50% and 98%, respectively. CONCLUSION. Combinations of various CT and demographic findings allowed differentiation of angiomyolipoma from RCC. O ver 10% of resected solid renal tumors are benign [1 3]. Angiomyolipoma accounts for 18 59% of resected benign solid tumors [1 3]. Diagnosis of angiomyolipoma on imaging relies on identification of macroscopic fat [4, 5]. CT pixel analysis has been used to identify small amounts of fat within renal masses on unenhanced CT [6 11]. However, the utility of this technique remains controversial in differentiating angiomyolipoma without visible fat from renal cell carcinoma (RCC) [7, 11]. When angiomyolipoma does not contain visible fat, it is difficult to differentiate from RCC. Jinzaki et al. [12] showed that a hyperattenuating mass on unenhanced CT was suggestive of angiomyolipoma without visible fat. In their series, all six angiomyolipomas without visible fat were hyperattenuating on unenhanced CT, whereas only five of 100 small RCCs yielded the same finding. Kim et al. [13] showed that 10 of 24 angiomyolipomas without visible fat were hyperat- tenuating on unenhanced CT, whereas eight of 67 small RCCs yielded the same finding. Angiomyolipoma without visible fat typically shows homogeneous enhancement, whereas RCC commonly shows heterogeneous enhancement on contrast-enhanced CT [12, 13]. Kim et al. [13] showed that a prolonged enhancement pattern (i.e., difference of CT attenuation values in corticomedullary and excretory phases between 20 and 20 HU) was suggestive of angiomyolipoma. Others showed that angiomyolipoma without visible fat yielded moderately strong enhancement during the parenchymal phase [14 17]. However, these enhancement characteristics are by themselves not specific because papillary RCC is often homogeneous and clear cell RCC often shows strong enhancement. Small solid renal masses are often found incidentally on contrast-enhanced CT performed for nonurologic indications. Additional unenhanced CT examinations may not be performed in all patients before surgery or ablation treatment. Therefore, it would be 1194 AJR:205, December 2015

2 CT of Small Renal Masses helpful to be able to predict the probability of angiomyolipoma on the basis of the contrast-enhanced CT alone. The purpose of our study was to determine if small (< 4 cm) angiomyolipoma without visible fat can be differentiated from RCC by use of contrastenhanced CT alone and by combination of unenhanced and contrast-enhanced CT. Materials and Methods Patients This HIPAA-compliant retrospective study was approved by our institutional review board. All patients had previously consented to the use of their medical records for research purposes. Between January 2003 and January 2011, 1214 patients underwent surgical removal of primary renal tumors less than 4 cm in diameter at our institution (Fig. 1). These included 962 patients with RCC, 79 with angiomyolipoma, 168 with oncocytoma, and five with other benign tumors. Of these, 50 patients with angiomyolipoma without visible fat (excluding 29 patients with angiomyolipoma with visible fat) and 320 randomly selected patients with RCC (i.e., approximately one third of the entire RCC cohort) were evaluated. Both unenhanced and contrast-enhanced CT scans were performed preoperatively in 27 of 50 patients with angiomyolipoma without visible fat and 139 of 320 patients with RCC. Patients with either a predominantly cystic mass without a measurable solid component or a small mural nodule (eight RCCs, four angiomyolipomas) and patients with a mass undetectable on contrast-enhanced CT (one RCC) were excluded. When a patient had angiomyolipomas both with and without visible fat, only the angiomyolipomas without visible fat were included in the study. One patient had two angiomyolipomas without visible fat, and 11 patients had multiple (2 25) RCCs. In these cases, up to three of the largest lesions in each kidney were included. The final cohort included 23 patients with 24 angiomyolipomas without visible fat (mean age, 53.0 years; three men, 20 women) and 130 patients with 148 RCC (mean age, 60.5 years; 78 men, 52 women) (Fig. 1). The 148 RCCs included 98 clear cell RCCs, 36 papillary RCCs, and 14 other subtypes of RCC. CT Protocol All patients underwent both unenhanced and contrast-enhanced CT, but the CT protocols were variable. One hundred sixteen CT examinations (of 122 masses) were performed at our institution, and 46 CT examinations (of 50 masses) were performed at outside institutions. At least one phase was obtained in mid- to late corticomedullary differentiation or generalized nephrographic phase in all patients (single phase in 128 masses Fig. 1 Flowchart shows patient inclusion and exclusion criteria and final diagnosis. RCC = renal cell carcinoma. [17 angiomyolipomas, 111 RCCs] and dual or triple phase in 44 masses [seven angiomyolipomas, 37 RCCs]). Slice thickness for unenhanced CT ranged from 2 to 7 mm, with a median of 5 mm; slice thickness for contrast-enhanced CT ranged from 1 to 5 mm, with a median of 5 mm. The dose and injection rate of contrast material were variable. The routine abdominal and kidney CT protocols at our institution were applied using 140 ml of IV contrast material (iohexol, 300 mg I/mL [Omnipaque 300, GE Healthcare]) with an injection rate of 3 ml/s. CT Image Analysis CT images were reviewed on an Advantage Windows 4.6 workstation (GE Healthcare). CT attenuation values of the renal mass and renal parenchyma were measured on transverse unenhanced and contrast-enhanced CT images. When there were multiple phases of contrast-enhanced CT, the measurements were done in each phase, and the highest attenuation value was used as the representative attenuation value of the lesion. The largest possible circular ROI was placed over the renal mass, but avoiding volume averaging from the adjacent renal parenchyma and perinephric fat. When calcification was present in the mass, the area of calcification was not included in the ROI. An ROI was also placed over the renal parenchyma (cortex) on the same or an adjacent slice. Presence of calcification within the renal mass was recorded. Long- and short-axis diameters of the tumor were measured on a transverse plane. Long-to-short-axis diameter ratio was calculated. For these parameters, a single 1214 Consecutive patients with resected renal masses < 4 cm 962 Patients with RCC 320 Patients randomly selected 130 Patients with 148 RCCs 201 RCCs, 23 angiomyolipomas excluded: either unenhanced or contrast-enhanced CT unavailable 8 RCCs, 4 angiomyolipomas excluded: predominantly cystic mass 173 Patients excluded: oncocytoma or other benign tumor 29 Patients excluded: angiomyolipoma with visible fat 50 Patients with angiomyolipoma without visible fat 23 Patients with 24 angiomyolipomas without visible fat radiologist (with 13 years of experience in abdominal imaging) who was blinded to the final diagnosis analyzed the CT images. The degree of exophytic growth of the tumor was graded by percentage of tumor outside the expected contour of renal parenchyma (0 100% at 10% increments). When the tumor had an angular interface sign (i.e., exophytic tumor having an angular-shaped intraparenchymal component and a round exophytic component) [18], this was recorded. Tumor heterogeneity on contrastenhanced CT was subjectively rated using a 3-point grading scale (1, homogeneous; 2, mildly heterogeneous; and 3, markedly heterogeneous). For these subjective parameters, two radiologists (with 13 and 7 years of experience in abdominal imaging) independently analyzed CT images. The mean of the two reviewers scores (for percentage of exophytic tumor growth and tumor heterogeneity) or the consensus of two readers (for angular interface sign) was used for statistical analysis. Objective Tumor Heterogeneity Analysis (Entropy) Contrast-enhanced CT images (in DICOM format) were exported to a PC, and the entropy of a CT attenuation histogram was calculated using in-house software based on Matlab (MathWorks). Image-based denoising was performed to reduce the effect of image noise [19] before calculating entropy. When there were multiple contrast phases, the phase in which the renal mass enhanced maximally was used. An ROI was placed on each renal mass by a single radiologist. The entropy, E, of the renal mass was defined as follows: AJR:205, December

3 Takahashi et al. TABLE 1: Univariate Analysis Demographic Data and Lesion Characteristics E = imax p(i)*log 2 p(i), i = i min where p(i) is the probability of image pixels having CT attenuation of i, with i ranging from minimal CT attenuation i min to maximal CT attenuation i max ; smaller value of E represents more homogeneous texture of the mass. Reference Standard and Statistical Analysis Standard of reference was pathologic examination in all tumors. Univariate analysis was performed to compare the various demographic and radiologic features between angiomyolipoma without visible fat and RCC. Parameters considered significant were entered in the multivariate logistic regression models. Three multivariate models were constructed for a combination of parameters obtained by unenhanced CT alone, by contrast-enhanced CT alone, and by both unenhanced and contrast-enhanced CT. Demographic data and size and shape of the renal mass were used in each model. Multivariate model selection was performed with backward elimination. Higher-order logistic regression was performed for selected variables after examining the scatterplot. Linear prediction function was calculated as a product of linear combination of the explanatory variables and a set of regression coefficients. On the basis of the logistic regression models, simplified scoring systems were constructed that could be used at a clinical reading session [20]. Each Angiomyolipomas Without Visible Fat explanatory variable used in the logistic regression models was divided into 2 4 categories, and score was assigned for each category. Total score was calculated by adding all assigned scores for the variables. AUC and sensitivities with 95% CIs were calculated for a few high specificity values (e.g., 90%, 95%, and 98% specificity) for each model. The 95% CIs were calculated by assuming statistical independence of the 172 lesions. Whereas the degree of clustering within patients was small (172 tumors in 153 patients), the potential for variance inflation because of the clustering of lesions within patients was assessed by using generalized estimating equations and empirical variance. No appreciable differences were observed in terms of width of 95% CIs; therefore, the clustering effect was considered negligible, and the 95% CIs of the assigned scores were reported on the basis of their optimal statistical performance [21]. The positive predictive values were adjusted for the disease prevalence (962 RCCs, 50 angiomyolipomas without visible fat). To assess the interobserver agreement for the subjective variables, we calculated weighted kappa value or intraclass correlation coefficient for those parameters found to be statistically significant. All statistical comparisons were performed by using statistical software (SPSS, version , SPSS; SAS, version 9.3, SAS Institute; or JMP, version 9, SAS Institute); p values less than 0.05 was considered to be statistically significant. Renal Cell Carcinomas All Clear Cell Papillary Other Subtype Age (y) 53 ± ± ± ± ± 15 < 0.05 Sex (% female) 88 (21/24) 37 (55/148) 43 (42/98) 22 (8/36) 36 (5/14) < Lesion size and shape Short-axis diameter (mm) 15 ± 7 21 ± 8 22 ± 8 18 ± 7 23 ± 8 < 0.01 Long-to-short-axis diameter ratio 1.3 ± ± ± ± ± Angular interface sign 8 (2/24) 2 (3/148) 1 (1/98) 3 (1/36) 7 (1/14) 0.1 % Exophytic growth 56 ± ± ± ± ± 28 < 0.05 Unenhanced CT data Lesion CT attenuation (HU) 40 ± ± ± ± ± 13 < Lesion-to-kidney CT attenuation difference (HU) 10 ± 12 0 ± 10 2 ± 9 3 ± 10 4 ± 11 < Calcification 0 (0/24) 7 (11/148) 8 (8/98) 3 (1/36) 14 (2/14) 0.2 Contrast-enhanced CT data Subjective heterogeneity score 1.6 ± ± ± ± ± 0.7 < 0.01 Entropy 4.6 ± ± ± ± ± 1.0 < 0.01 Lesion CT attenuation (HU) 113 ± ± ± ± ± 38 < 0.05 Note Data are given as either mean ± SD or percentages with proportion of cases in parentheses. Results Univariate Analyses Results of univariate analyses are summarized in Table 1 (Figs. 2 4). Among the demographic data, female sex (p < 0.001) and younger age (p < 0.05) were associated with angiomyolipoma without visible fat. Among the shape and size of the tumor, smaller long- and short-axis diameters (p < 0.01) and greater percentage of exophytic tumor growth (p < 0.05) were associated with angiomyolipoma without visible fat. Among variables obtained on unenhanced CT, higher lesion CT attenuation (p < 0.001) and higher lesion-to-kidney CT attenuation difference (p < ) were associated with angiomyolipoma without visible fat. Among variables obtained on contrast-enhanced CT, lower entropy (homogeneous) (p < 0.01), lower subjective heterogeneity score (homogeneous) (p < 0.01), and higher lesion CT attenuation (p < 0.05) were associated with angiomyolipoma without visible fat. Multivariate Logistic Regression Analyses Results of multivariate logistic regression analyses are summarized in Table 2. For the unenhanced CT prediction model, younger age, female sex, smaller short-axis diameter, greater percentage of exophytic tumor growth, and higher lesion-to-kidney p 1196 AJR:205, December 2015

4 CT of Small Renal Masses TABLE 2: Multivariate Logistic Regression Models and Simplified Scoring Models Unenhanced CT based Model Coefficients for Logistic Regression Models Intercept 3.3 Scores for Simplified Scoring Models Age (y) < Sex a 3.1 Male Female Lesion-to-kidney CT attenuation difference on unenhanced CT (HU) < 10 HU 10 to Short-axis diameter (mm) 0.16 < % Exophytic growth < Enhanced CT based Intercept Sex a 2.1 Male Female 0 2 % Exophytic growth < Entropy or subjective heterogeneity b 1.3 Homogeneous Mildly heterogeneous Markedly heterogeneous Lesion CT attenuation on contrast-enhanced CT HU 0.22 < HU*HU Unenhanced and enhanced CT based Intercept 7.4 Age (y) < Sex a 2.6 Male Female Lesion-to-kidney CT attenuation difference on unenhanced CT (HU) < to Short-axis diameter (mm) < % Exophytic growth < Lesion CT attenuation on contrast-enhanced CT HU 0.17 < HU*HU Note Scores are shown in boldface below parametric data. Larger coefficients and score are associated with angiomyolipoma without visible fat. The logit can be calculated by linear combination of the explanatory variables and a set of regression coefficient. The total score for the simplified scoring models can be calculated by adding assigned scores for the explanatory variables. a For logistic regression models, female sex is considered 0, and male sex is considered 1. b Entropy is used for logistic regression models, and subjective heterogeneity is used for simplified scoring models. AJR:205, December

5 Takahashi et al. CT attenuation difference on unenhanced CT were considered as significant. For the contrast-enhanced CT prediction model, female sex, greater percentage of exophytic tumor growth, lower entropy (homogeneous), and higher lesion CT attenuation on contrast-enhanced CT were considered as significant. For the combined unenhanced CT and contrast-enhanced CT prediction model, younger age, female sex, smaller short-axis diameter, greater percentage of exophytic tumor growth, higher lesion-to-kidney CT attenuation difference on unenhanced CT, and higher lesion CT attenuation on contrast-enhanced CT were considered as significant. Simplified Scoring System Based on Multivariate Logistic Regression Analyses Simplified scoring systems are shown in Table 2. For the contrast-enhanced CT based model, the entropy was replaced with the subjective heterogeneity score (by one reviewer) in the logistic regression model before constructing the simplified scoring system. For example, the patient shown in Figure 2 (female sex, 2 points; exophytic growth, 1; mildly heterogeneous mass, 1; enhancement of 113 HU, 5) would have a total score of 9 in the contrast-enhanced CT based model, and thus has a 51% chance of having an angiomyolipoma (Table 3). For comparison, clear cell and papillary RCCs are shown in Figures 3 and 4. Diagnostic Performance of the Logistic Regression and Simplified Scoring Models AUC, sensitivities and specificities with 95% CIs, and prevalence-adjusted positive predictive values are summarized in Table 3. The contrast-enhanced CT based model and combined unenhanced and contrast-enhanced CT based model differentiated angiomyolipoma from RCC with sensitivity, specificity, and AUC of 42% (10/24), 98% (145/148), and and 50% (12/24), 98% (145/148), and 0.939, respectively. Agreement for Subjective Data Agreement between two reviewers was 0.68 for subjective heterogeneity score (weighted κ) and 0.77 for percentage of exophytic growth (intraclass correlation coefficient). Discussion The present study showed that the logistic regression model constructed from the findings of contrast-enhanced CT can diagnose angiomyolipoma without visible fat with moderate accuracy. Addition of unenhanced CT further improved the diagnostic accuracy. Several studies have described CT findings associated with angiomyolipoma without visible fat. The two most useful imaging findings suggestive of angiomyolipoma without visible fat are hyperattenuating mass on unenhanced CT and homogeneous enhancing mass on contrast-enhanced CT, both of which were first described by Jinzaki et al. [12]. Hyperattenuating mass on unenhanced CT had a sensitivity of % and a specificity of TABLE 3: Sensitivity, Specificity, Positive Predictive Value (PPV), and AUC for Logistic Regression and Simplified Scoring Models Model Sensitivity (%) a Specificity (%) a PPV (%) b Threshold AUC Logistic regression Unenhanced CT based 46 (11/24) [26 67] 98 (145/148) [94 100] (17/24) [49 87] 95 (141/148) [91 98] (17/24) [49 87] 90 (133/148) [84 94] Contrast-enhanced CT based 42 (10/24) [22 63] 98 (145/148) [94 100] (16/24) [45 86] 95 (141/148) [91 98] (19/24) [58 93] 90 (133/148) [84 94] (20/24) [63 95] 85 (126/148) [78 90] Unenhanced and contrast-enhanced CT based 50 (12/24) [33 74] 98 (145/148) [94 100] (17/24) [49 87] 95 (141/148) [91 98] (20/24) [63 95] 90 (133/148) [84 94] Simplified scoring Unenhanced CT based 46 (11/24) [26 67] 97 (144/148) [93 99] (13/24) [34 75] 95 (141/148) [91 98] (18/24) [53 90] 85 (126/148) [78 90] 21 9 Contrast-enhanced CT based 38 (9/24) [19 59] 97 (144/148) [93 99] (16/24) [45 84] 97 (143/148) [92 99] (18/24) [53 90] 87 (129/148) [81 92] (20/24) [63 95] 80 (118/148) [72 86] 18 7 Unenhanced and contrast-enhanced CT based 50 (12/24) [33 74] 97 (144/148) [93 99] (16/24) [45 84] 96 (142/148) [91 99] (20/24) [63 95] 86 (127/148) [79 91] Note For each model, 3 4 sets of sensitivity, specificity, positive predictive values are shown for different threshold levels with high specificity values (e.g., 90%, 95%, and 98% specificity). a Proportion of cases are given in parentheses and 95% CIs in brackets. b Prevalence adjusted AJR:205, December 2015

6 CT of Small Renal Masses A B Fig year-old woman with angiomyolipoma without visible fat ( cm) in left kidney. A, Transverse unenhanced CT scan shows exophytic hyperattenuating (22 HU above renal parenchyma) left renal mass. B, On transverse contrast-enhanced CT scan, mass is mildly heterogeneous (entropy, 4.9) and shows moderate degree of enhancement (113 HU). Calculated probability of angiomyolipoma was approximately 33% using contrast-enhanced CT data alone but increased to approximately 98% when unenhanced CT data were added % for the diagnosis of angiomyolipoma without visible fat [12, 13]. In our study, the respective sensitivities and specificities of the hyperattenuating mass were 63% (15/24) and 87% (111/128) for a threshold of 10 HU or higher CT attenuation difference between renal mass and renal parenchyma and 42% (10/24) and 88% (113/128) for a threshold of 45 HU or higher CT attenuation of the renal mass. A recent study by Schieda et al. [22] also found overlap in the CT attenuation values between angiomyolipoma without visible fat and RCC on unenhanced CT. Our study confirmed that angiomyolipoma without visible fat is homogeneous on contrast-enhanced CT as compared with RCC. In this study, the entropy calculated from CT attenuation histogram was used as an objective index of the heterogeneity. Both entropy and subjective heterogeneity score (mean of two readers scores) were significantly different between angiomyolipoma without visible fat and RCC on univariate analysis; however, only entropy remained significant in the multivariate logistic regression analysis. Entropy has the advantage of better reproducibility. However, entropy cannot be calculated on a regular workstation, and it is influenced by the image noise. Therefore, a process to reduce the effect of image noise is necessary, and in our study, image-based noise reduction was applied before calculating the entropy. Quantitative histogram analysis of tumor texture using CT or MR images has recently been shown to be useful for prediction of patient prognosis [23 26] and for differentiation between various renal masses [27]. Previous reports of enhancement pattern of angiomyolipoma without visible fat have A Fig year-old woman with clear cell renal cell carcinoma ( cm) in right kidney. A, Transverse unenhanced CT scan shows isoattenuating right renal mass (arrow). B, On transverse contrast-enhanced CT scan, mass (arrow) is homogeneous (entropy, 3.5) and shows strong enhancement (159 HU). Calculated probability of angiomyolipoma was approximately 38% using contrast-enhanced CT data alone but decreased to approximately 5% when unenhanced CT data were added. B AJR:205, December

7 Takahashi et al. A Fig year-old man with papillary renal cell carcinoma ( cm) in left kidney. A, Transverse unenhanced CT scan shows exophytic hyperattenuating (13 HU above renal parenchyma) mass (arrow) in left kidney laterally. B, On transverse contrast-enhanced CT scan, mass (arrow) is homogeneous (entropy, 4.0) and shows low-grade enhancement (57 HU). Calculated probability of angiomyolipoma was approximately 8% using unenhanced CT data alone but decreased to approximately 2% when contrast-enhanced CT data were added. Another small papillary renal cell carcinoma is partially seen posteriorly in left kidney. been somewhat contradictory. Kim et al. [13] found that the gradual enhancement pattern on biphasic CT was a useful predictor distinguishing angiomyolipoma without visible fat from RCC. Hafron et al. [14] found that angiomyolipoma without visible fat had an increase in attenuation of at least 90 HU on contrast-enhanced CT. Zhang et al. [15] found that the mean attenuation of angiomyolipoma without visible fat was 128 HU on the parenchymal phase of contrast-enhanced CT. In our study, angiomyolipoma without fat had CT attenuation similar to that of clear cell RCC on contrast-enhanced CT, with a mean attenuation of 113 HU. Recent studies showed that angiomyolipoma without visible fat had strong enhancement with washout pattern on MRI [16, 17]. Washout pattern could not be assessed in the present study because many of the CT examinations were single-phase studies after administration of contrast material. Jinzaki et al. [12] showed that angiomyolipoma was more often exophytic than RCC. In their study, 83% (5/6) of angiomyolipomas and 3% (3/100) of RCCs were exophytic without the beak sign of renal parenchyma at the tumor-parenchyma interface. In our study, a greater percentage of exophytic growth of the tumor was associated with angiomyolipoma without visible fat. The angular interface sign of tumor was previously reported to be specific for the presence of a benign tumor such as angiomyolipoma or a benign cystic lesion [18]. In our study, only two of 24 angiomyolipomas without visible fat had such findings, although when seen these findings were considered to be specific. Calcification was found in 7% (11/148) of RCCs, but this was not seen in any of the angiomyolipomas without visible fat. Although our proposed logistic regression models did not include calcification because it was not considered statistically significant, the possibility of angiomyolipoma approaches zero when calcification is present in a small solid renal mass [13]. Smaller tumor size, female sex and younger age were also associated with angiomyolipoma without visible fat, in agreement with previous reports [1, 2]. Although several radiologic and demographic features of renal masses were confirmed to be associated with angiomyolipoma without visible fat in our study, none of the findings were sufficiently accurate because the prevalence of angiomyolipoma without visible fat in our study cohort was only 5% (50/1012). Therefore, we constructed a logistic regression model by combining several features to improve the diagnostic accuracy. This is similar to the approaches used by Woo et al. [28] for differentiating angiomyolipoma without visible fat from non clear cell RCC and by Lane et al. [29] for differentiating benign from malignant renal masses. We also constructed a simplified scoring model that could be calculated at any workstation. As expected, two strong predictors of angiomyolipoma without visible fat in our model were hyperattenuation on unenhanced CT and homogeneous enhancement on contrastenhanced CT. Smaller size and female sex were similarly strong predictors of angiomyolipoma. Low-grade enhancement (< 75 HU) on contrast-enhanced CT was also a strong negative predictor of angiomyolipoma. Small solid renal masses are often found incidentally on contrast-enhanced CT performed for nonurologic indications. At our institution, additional unenhanced CT is not performed in all patients before surgery or ablation treatment of small renal masses incidentally found on contrast-enhanced CT; only 43% (139/320) of patients with small RCCs underwent both unenhanced and contrast-enhanced CT before surgery in our study population. One could argue that all patients with a small solid mass should undergo unenhanced CT before surgery or ablation treatment because unenhanced CT may detect a hyperattenuating mass [12] or a small amount of macroscopic fat undetectable on contrastenhanced CT [4, 10]. However, it may be reasonable to perform additional examination with unenhanced CT only in selected patients with a higher pretest probability of angiomyolipoma given that the prevalence of angiomyolipoma without visible fat was only 5% in surgically resected renal masses smaller than 4 cm at our institution. The present study showed that the contrast-enhanced CT prediction model can improve the posttest B 1200 AJR:205, December 2015

8 CT of Small Renal Masses probability. For example, the posttest probability of angiomyolipoma can be improved to 23% from 5% while identifying 83% of angiomyolipomas (logit threshold, 1.9). Addition of unenhanced CT further improved the diagnostic accuracy. The diagnostic performance of the proposed model is not sufficient to make the definitive diagnosis of angiomyolipoma by itself. The prevalence-adjusted positive predictive values are in the range of 50 60% at best. Therefore, the model should be used in conjunction with other diagnostic tests. Treatment options for small solid renal masses include surgical resection, percutaneous ablation, percutaneous needle biopsy, or watchful waiting on the basis of both the likelihood of malignancy at imaging and other clinical factors [30]. Imageguided percutaneous biopsy can accurately diagnose angiomyolipomas without visible fat [31]. Silverman et al. [32] have recommended that if a small renal mass is hyperattenuating on unenhanced CT and homogeneously enhancing on contrast-enhanced CT and shows low T2 signal on MR images, then percutaneous biopsy should be considered. Our proposed CT diagnostic model could be incorporated in the diagnostic algorithm to determine who should undergo additional dedicated CT, MRI, or percutaneous biopsy before a decision is made regarding surgery, ablation, or watchful waiting. Our study had a number of limitations. Oncocytoma was not included, and different subtypes of RCC were considered as a single entity. Having more than two outcome categories would require construction of polynomial logistic regression models; however, with such models, it becomes more difficult to intuit the contribution of each parameter, and construction of the simplified scoring system would not be possible. Higher-order logistic regression was used for the lesion enhancement value to overcome the issue of combining different subtypes of RCC. The present study was retrospective in nature, and CT technique was variable with inclusion of CT examinations performed at outside institutions; moreover, amount and injection rate of iodinated contrast material and scan delay were not recorded. Only a single phase of contrast-enhanced CT was evaluated. A relatively small number of angiomyolipoma cases was included. A large number of patients were excluded because unenhanced CT had not been performed before surgery. Only tumors smaller than 4 cm were included. Objective measurements (CT attenuation, size) were performed by a single radiologist. Finally, a larger-scale, multiinstitutional study would be necessary to validate our study results. In conclusion, contrast-enhanced CT can help to differentiate angiomyolipoma without visible fat from RCC. Addition of unenhanced CT can further improve the diagnostic accuracy. References 1. Frank I, Blute ML, Cheville JC, Lohse CM, Weaver AL, Zincke H. Solid renal tumors: an analysis of pathological features related to tumor size. J Urol 2003; 170: Fujii Y, Komai Y, Saito K, et al. Incidence of benign pathologic lesions at partial nephrectomy for presumed RCC renal masses: Japanese dual-center experience with 176 consecutive patients. Urology 2008; 72: Kutikov A, Fossett LK, Ramchandani P, et al. Incidence of benign pathologic findings at partial nephrectomy for solitary renal mass presumed to be renal cell carcinoma on preoperative imaging. Urology 2006; 68: Bosniak MA, Megibow AJ, Hulnick DH, Horii S, Raghavendra BN. CT diagnosis of renal angiomyolipoma: the importance of detecting small amounts of fat. AJR 1988; 151: Israel GM, Hindman N, Hecht E, Krinsky G. The use of opposed-phase chemical shift MRI in the diagnosis of renal angiomyolipomas. AJR 2005; 184: Simpson E, Patel U. Diagnosis of angiomyolipoma using computed tomography: region of interest -10 HU or 4 adjacent pixels -10 HU are recommended as the diagnostic thresholds. Clin Radiol 2006; 61: Catalano OA, Samir AE, Sahani DV, Hahn PF. Pixel distribution analysis: can it be used to distinguish clear cell carcinomas from angiomyolipomas with minimal fat? Radiology 2008; 247: Kim JY, Kim JK, Kim N, Cho KS. CT histogram analysis: differentiation of angiomyolipoma without visible fat from renal cell carcinoma at CT imaging. Radiology 2008; 246: Simpfendorfer C, Herts BR, Motta-Ramirez GA, et al. Angiomyolipoma with minimal fat on MDCT: can counts of negative-attenuation pixels aid diagnosis? AJR 2009; 192: Davenport MS, Neville AM, Ellis JH, Cohan RH, Chaudhry HS, Leder RA. Diagnosis of renal angiomyolipoma with Hounsfield unit thresholds: effect of size of region of interest and nephrographic phase imaging. Radiology 2011; 260: Chaudhry HS, Davenport MS, Nieman CM, Ho LM, Neville AM. Histogram analysis of small solid renal masses: differentiating minimal fat angiomyolipoma from renal cell carcinoma. AJR 2012; 198: Jinzaki M, Tanimoto A, Narimatsu Y, et al. Angiomyolipoma: imaging findings in lesions with minimal fat. Radiology 1997; 205: Kim JK, Park SY, Shon JH, Cho KS. Angiomyolipoma with minimal fat: differentiation from renal cell carcinoma at biphasic helical CT. Radiology 2004; 230: Hafron J, Fogarty JD, Hoenig DM, Li M, Berkenblit R, Ghavamian R. Imaging characteristics of minimal fat renal angiomyolipoma with histologic correlations. Urology 2005; 66: Zhang J, Lefkowitz RA, Ishill NM, et al. Solid renal cortical tumors: differentiation with CT. Radiology 2007; 244: Sasiwimonphan K, Takahashi N, Leibovich BC, Carter RE, Atwell TD, Kawashima A. Small (<4 cm) renal mass: differentiation of angiomyolipoma without visible fat from renal cell carcinoma utilizing MR imaging. Radiology 2012; 263: Vargas HA, Chaim J, Lefkowitz RA, et al. Renal cortical tumors: use of multiphasic contrast-enhanced MR imaging to differentiate benign and malignant histologic subtypes. Radiology 2012; 264: Verma SK, Mitchell DG, Yang R, et al. Exophytic renal masses: angular interface with renal parenchyma for distinguishing benign from malignant lesions at MR imaging. Radiology 2010; 255: Manduca A, Yu L, Trzasko JD, et al. Projection space denoising with bilateral filtering and CT noise modeling for dose reduction in CT. Med Phys 2009; 36: Sullivan LM, Massaro JM, D Agostino RB Sr. Presentation of multivariate data for clinical use: the Framingham Study risk score functions. Stat Med 2004; 23: Newcombe RG. Two-sided confidence intervals for the single proportion: comparison of seven methods. Stat Med 1998; 17: Schieda N, Hodgdon T, El-Khodary M, Flood TA, McInnes MD. Unenhanced CT for the diagnosis of minimal-fat renal angiomyolipoma. AJR 2014; 203: Goh V, Ganeshan B, Nathan P, Juttla JK, Vinayan A, Miles KA. Assessment of response to tyrosine kinase inhibitors in metastatic renal cell cancer: CT texture as a predictive biomarker. Radiology 2011; 261: Ng F, Ganeshan B, Kozarski R, Miles KA, Goh V. Assessment of primary colorectal cancer heterogeneity by using whole-tumor texture analysis: contrast-enhanced CT texture as a biomarker of AJR:205, December

9 Takahashi et al. 5-year survival. Radiology 2013; 266: Raman SP, Chen Y, Schroeder JL, Huang P, renal tumors 7 cm or less amenable to partial ne- 25. Downey K, Riches SF, Morgan VA, et al. Relation- Fishman EK. CT texture analysis of renal masses: phrectomy. J Urol 2007; 178: ship between imaging biomarkers of stage I cervi- pilot study using random forest classification for 30. Silverman SG, Israel GM, Herts BR, Richie JP. cal cancer and poor-prognosis histologic features: prediction of pathology. Acad Radiol 2014; Management of the incidental renal mass. quantitative histogram analysis of diffusion- 21: Radiology 2008; 249:16 31 weighted MR images. AJR 2013; 200: Kyriazi S, Collins DJ, Messiou C, et al. Metastatic ovarian and primary peritoneal cancer: assessing chemotherapy response with diffusion-weighted MR imaging value of histogram analysis of apparent diffusion coefficients. Radiology 2011; 261: Woo S, Cho JY, Kim SH, Kim SY. Angiomyolipoma with minimal fat and non-clear cell renal cell carcinoma: differentiation on MDCT using classification and regression tree analysis-based algorithm. Acta Radiol 2014; 55: Lane BR, Babineau D, Kattan MW, et al. A preoperative prognostic nomogram for solid enhancing 31. Silverman SG, Gan YU, Mortele KJ, Tuncali K, Cibas ES. Renal masses in the adult patient: the role of percutaneous biopsy. Radiology 2006; 240: Silverman SG, Mortele KJ, Tuncali K, Jinzaki M, Cibas ES. Hyperattenuating renal masses: etiologies, pathogenesis, and imaging evaluation. RadioGraphics 2007; 27: AJR:205, December 2015