Qualitative and Quantitative MDCT Features for Differentiating Clear Cell Renal Cell Carcinoma From Other Solid Renal Cortical Masses

Genitourinary Imaging Original Research Lee-Felker et al. MDCT Differentiation of Clear Cell RCC Genitourinary Imaging Original Research Stephanie A. Lee-Felker 1 Ely R. Felker 1 Nelly Tan 1 Daniel J. A. Margolis 1 Jonathan R. Young 1 James Sayre 2 Steven S. Raman 1 Lee-Felker SA, Felker ER, Tan N, et al Keywords: clear cell renal cell carcinoma, lipid-poor angiomyolipoma, multiphasic MDCT, oncocytoma, papillary renal cell carcinoma DOI:10.2214/AJR.14.12460 Received December 22, 2013; accepted after revision March 9, 2014. 1 Department of Radiological Sciences, David Geffen School of Medicine at UCLA, Ronald Reagan UCLA Medical Center, 757 Westwood Plaza, Ste 1638, Los Angeles, CA 90095. Address correspondence to S. A. Lee-Felker (stlee@mednet.ucla.edu). 2 Department of Biostatistics, UCLA School of Public Health, Los Angeles, CA. WEB This is a web exclusive article. AJR 2014; 203:W516 W524 0361 803X/14/2035 W516 American Roentgen Ray Society Qualitative and Quantitative MDCT Features for Differentiating Clear Cell Renal Cell Carcinoma From Other Solid Renal Cortical Masses OBJECTIVE. The purpose of this study was to differentiate clear cell renal cell carcinoma (RCC) from other solid renal masses on four-phase MDCT. MATERIALS AND METHODS. Our study cohort included all pathologically proven solid renal masses that underwent pretreatment four-phase MDCT at our institution from 2001 to 2012. Both retrospective qualitative analysis (blinded dual-radiologist evaluation of morphologic features: enhancement pattern, lesion contour, neovascularity, and calcification) and quantitative analysis (mean absolute and relative attenuation and changes in attenuation across phases) were performed. ANOVA with post-hoc analysis, Pearson chi-square tests, and ROC analysis were used. RESULTS. One hundred fifty-six consecutive patients (99 men, 57 women) with a mean age of 62.7 years (range, 26 91 years) had 165 solid renal masses (median size, 3.0 cm): 86 clear cell RCCs, 36 papillary RCCs, 10 chromophobe RCCs, 23 oncocytomas, and 10 lipid-poor angiomyolipomas. Kappa for interradiologist agreement regarding morphologic features was 0.33 0.76. There were significant associations between histologic subtype and enhancement pattern (p < 0.001), lesion contour (p < 0.014), and neovascularity (p < 0.001). Clear cell RCC had the highest mean relative corticomedullary attenuation (p < 0.02). Clear cell RCC had greater deenhancement than oncocytoma (p < 0.001); deenhancement less than 50 HU or relative corticomedullary attenuation greater than 0% differentiated clear cell RCC from oncocytoma with a positive predictive value of 90%. Lipid-poor angiomyolipoma had the highest mean absolute unenhanced attenuation (p < 0.01); absolute unenhanced attenuation greater than 45 HU and relative corticomedullary attenuation less than 10% differentiated lipid-poor angiomyolipoma from clear cell RCC with a negative predictive value of 97%. CONCLUSION. Four-phase MDCT renal attenuation profiles enable differentiation of clear cell RCC from other solid renal cortical masses, most notably papillary RCC and lipidpoor angiomyolipoma. T he incidence of renal cell carcinoma (RCC) is rising by 2 4% annually, with 65,150 new cases in the United States in 2013 [1], and up to 70% of solid renal cortical masses are detected incidentally on imaging studies [2]. Generally, solid renal cortical masses without macroscopic fat detected on MDCT are presumed malignant, although up to 30% of resected solid renal masses are pathologically benign, most commonly oncocytoma and lipid-poor angiomyolipoma [3]. Malignant lesions vary significantly in biologic aggressiveness on the basis of histologic subtype. Up to 80% are clear cell RCCs, which are generally more aggressive than papillary (15%) or chromophobe (5%) subtypes [4]. Clear cell RCC accounts for 94% of metastatic RCC, with 44 69% 5-year sur- vival, whereas papillary RCC (type 1) and chromophobe RCC are more indolent, accounting for 6% of all metastatic RCC, with 78 92% 5-year survival [5 9]. Traditionally, enhancement for renal masses was defined in only one phase, either corticomedullary or nephrographic. Recently, this has been challenged, and renal mass enhancement is now viewed in terms of differences in contrast enhancement and deenhancement patterns over multiple phases on both multiphasic MDCT and dynamic contrast-enhanced MRI to distinguish among RCC subtypes [10 17]. However, distinguishing clear cell RCC from its two most common benign mimics, oncocytoma and lipid-poor angiomyolipoma, remains challenging. Many prior studies have not been able to find reliable W516 AJR:203, November 2014

MDCT Differentiation of Clear Cell RCC characteristic CT features that discriminate solid renal masses. Most studies had small sample sizes and were insufficiently powered to detect differences, specifically between clear cell RCC and oncocytoma [12, 18, 19]. The purpose of our study was therefore to investigate the performance of four-phase MDCT in differentiating clear cell RCC from other solid renal cortical masses on the basis of both qualitative and quantitative imaging features. The study was designed to differentiate clear cell RCC from other solid renal cortical masses because in routine clinical practice, clear cell RCC is the most commonly encountered aggressive solid renal cortical mass and has the worst prognosis. Materials and Methods Patients and Lesions Our institutional review board approved this study and waived the requirement for informed consent. This study complied with the HIPAA. Inclusion criteria included any RCC, oncocytoma, or angiomyolipoma with both pretreatment fourphase MDCT between September 2001 and February 2012 and institutional pathology diagnosis. Exclusion criteria included detectable macroscopic fat on MDCT for lipid-poor angiomyolipoma. We compiled a cohort of 156 consecutive patients with 165 solid renal masses for retrospective analysis. Of these patients, 83 (53.2%) were from our previous study [11], which had a total of 274 patients. The other 191 patients were not included in our current study because they did not undergo four-phase renal mass protocol MDCT. Rather, data regarding renal mass attenuation across the four imaging phases for these 191 patients were collected from various unenhanced, portal venous phase, and urographic MDCT studies performed on different dates. CT Examination All patients underwent pretreatment imaging evaluation with contrast-enhanced four-phase MDCT. CT studies were performed on 4-MDCT (seven patients), 16-MDCT (15 patients), and 64- MDCT (134 patients) helical scanners over the course of the study period (2001 2003, LightSpeed Qx/I, GE Healthcare; 2003 2006, Sensation 16, Siemens Healthcare; 2006 2012, Sensation 64 or Definition 64, Siemens Healthcare). CT images were acquired during patient breath-hold using the following parameters: 120 kvp, 200 400 ma depending on patient size, and 3-mm scan collimation with 3-mm reconstruction intervals. The four-phase MDCT renal mass protocol consisted of imaging in the unenhanced, corticomedullary, nephrographic, and excretory phases after power injection of nonionic IV contrast material (iodixanol, Omnipaque 350, GE Healthcare), 35 45 g iodine dosed to weight at a rate of 3 ml/s. Patients who weighed less than 45 kg, those who weighed 45 90 kg, and those who weighed more than 90 kg received ml (35 g iodine), 125 ml (45 g iodine), and 150 ml (54 g iodine) of contrast material, respectively. A bolus tracking program (Care Bolus VB10, Siemens Healthcare or Smart- Prep, GE Healthcare) was used to determine the onset of imaging in the corticomedullary, nephrographic, and excretory phases. For bolus tracking, an ROI was placed in the thoracoabdominal aorta junction, with a trigger set to begin at 150 HU. Images in the corticomedullary, nephrographic, and excretory phases were acquired 55 seconds, 120 seconds, and 8 minutes after the threshold level of 150 HU was reached. For the patients imaged with earlier generation (4- and 16-MDCT) helical scanners, the corticomedullary and nephrographic phases were performed slightly earlier (40 and 50 seconds and 90 and 110 seconds after the threshold level of 150 HU was reached, respectively). Overall, there was very little variation in the timing of image acquisition; all included patients were imaged within 5 seconds of the stated times. CT Image Analysis Qualitative analysis Two abdominal fellowship-trained genitourinary radiologists with 15 and 7 years of experience reviewed CT images independently. The radiologists were aware that imaging was performed to evaluate renal lesions but were blinded to follow-up imaging and pathology results. Before image interpretation, both radiologists met to define the qualitative features to be used for renal mass characterization. All previously studied cases were reevaluated for this series. Qualitative analysis was completed in four independent sessions by each radiologist. Each solid renal mass was evaluated for several previously described morphologic features: enhancement pattern, lesion contour, neovascularity, and calcification. For patients with multiple renal lesions, the radiologists were told how many lesions were present before interpretation. Enhancement pattern was determined qualitatively as either homogeneous or heterogeneous. Heterogeneous lesions contained a mixture of solid enhancing areas and cystic or necrotic nonenhancing areas. Lesion contour was smooth or irregular. Neovascularity was defined as the presence of increased, irregular, and unnamed vessels in the Gerota fascia adjacent to the involved kidney. Calcification was documented when present [12]. On the basis of assessment of morphologic features of each lesion, the radiologists predicted a specific histologic diagnosis for each lesion with the following 4-point confidence score: 1, less than 25%; 2, between 26% and 50%; 3, between 51% and 75%; and 4, greater than 75%. The predicted histologic subtype was compared with the pathologic diagnosis, which was obtained by reviewing existing pathology records, with tissue obtained from partial nephrectomy in 82 patients, radical nephrectomy in 43 patients, and percutaneous biopsy in 40 patients. A correct predicted histology with a confidence score of 4 was considered accurate for the purpose of calculating radiologist performance. A correct predicted histology with a confidence score of less than 4 was not considered accurate for these calculations. Quantitative analysis Two authors, who did not perform morphologic analysis, obtained lesion attenuation data by manually selecting an ROI in the center of homogeneous lesions and in the maximally attenuating portion of heterogeneous lesions on the imaging phase of maximal attenuation. Using the coregistration tool on PACS and anatomic landmarks for guidance, three round or elliptical ROIs not smaller than 0.1 cm 2 were placed in the same location on each of the four imaging phases, such that three ROIs were present within the maximally attenuating portion of each lesion at the same time. The average value of the three ROIs in each phase was recorded. One ROI was also placed in the adjacent uninvolved renal cortex in each phase to normalize for variations in attenuation due to individual patient and technical factors. For example, in patients with mild renal dysfunction, renal artery stenosis, or diminished cardiac output, the timing and magnitude of peak renal cortical attenuation differ, and normalizing renal lesion attenuation to that of renal cortex helps to control for these differences [18]. Areas with calcification or artifact were avoided. The mean absolute attenuation was calculated as the average of the three ROI measurements in each phase. The relative attenuation was calculated using the formula [(lesion ROI cortex ROI) / cortex ROI] %. The mean relative attenuation was calculated as the average of the three relative attenuation measurements for each phase. Absolute enhancement from the unenhanced to corticomedullary phases was calculated as lesion ROI corticomedullary lesion ROI unenhanced, and absolute deenhancement from the corticomedullary to nephrographic phases was calculated as lesion ROI corticomedullary lesion ROI nephrographic. Statistical Analysis For qualitative analysis, an unweighted kappa statistic was used to determine interradiologist agreement regarding morphologic features, with a kappa value of 0.20 0.40 indicating fair agreement; 0.41 0.60, moderate agreement; 0.61 0.80, good agreement; and greater than 0.80, excel- AJR:203, November 2014 W517

Lee-Felker et al. TABLE 1: Patient and Lesion Characteristics Characteristic All Lesions Clear Cell RCC Papillary RCC a Chromophobe RCC Lipid-Poor Angiomyolipoma Sex Male 107 54 28 5 16 4 Female 58 32 8 5 7 6 Mean age (y) 62.7 61.9 63.4 63.7 67.5 53.5 Range (y) 26 91 26 91 40 84 47 80 37 84 27 70 Pathologic source of lesions Partial nephrectomy 82 33 23 7 14 5 Radical nephrectomy 43 32 4 2 4 1 Biopsy 40 21 9 1 5 4 Median lesion size (cm) 3.0 3.1 3.0 3.2 3.7 2.7 Range (cm) 0.7 20.8 0.7 16.1 0.8 9.5 1.2 6.0 1.7 20.8 0.7 4.2 No. of lesions 165 86 36 10 23 10 Pathologic tumor stage b T1a 55 34 16 5 NA NA T1b 19 11 5 3 NA NA T2 10 7 3 0 NA NA T3 17 13 3 1 NA NA T4 0 0 0 0 NA NA Note Study consisted of 156 consecutive patients with 165 solid renal masses. RCC = renal cell carcinoma, NA = not applicable. a There were 29 type 1 papillary RCCs and seven type 2 papillary RCCs. b Stage was recorded when available. lent agreement. Cohen simple unweighted kappa was used because the categories for analysis were nominal. The Pearson chi-square test was used to assess the distribution of morphologic features across histologic subtypes. For qualitative features, a p value less than or equal to 0.05 was significant. Accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. For quantitative analysis, 95% CIs of mean attenuation were calculated. Significant differences among histologic subtypes were determined using ANOVA with post-hoc analysis (Dunnett T3 tests for attenuation on unenhanced phase and Bonferroni tests for attenuation across the unenhanced, corticomedullary and nephrographic phases). For quantitative features, an α less than or equal to 0.01 was significant. Significant differences among histologic subtypes regarding enhancement and deenhancement were determined using generalized estimating equations and ANOVA with posthoc analysis (Bonferroni tests). ROC analysis was used to determine optimal thresholds for differentiating histologic subtypes. Results Patient and Lesion Characteristics Our study cohort included 156 consecutive patients (mean age, 62.7 years) with 165 solid renal masses (median size, 3.0 cm). Pathologic specimens were acquired most commonly after partial or radical nephrectomy and less commonly after percutaneous biopsy. Of the 165 lesions, 132 (80%) were malignant. Of these, 86 (65%) were clear cell RCC, 36 (27%) were papillary RCC, and 10 (8%) were chromophobe RCC. Of 33 (20%) benign lesions, 23 (14%) were oncocytoma, and 10 (6%) were lipid-poor angiomyolipoma. Three patients had multiple lesions: one had five clear cell RCCs on a total nephrectomy specimen, one had two clear cell RCCs on partial nephrectomy specimen, and one had two oncocytomas on a total nephrectomy specimen. Of patients with available pathologic staging, most had early stage (T1a or T1b) lesions at the time of diagnosis. No patients with lipid-poor angiomyolipoma had tuberous sclerosis (Table 1). Qualitative Analysis Kappa values for interradiologist agreement for morphologic features ranged from 0.33 to 0.76, indicating fair to good agreement depending on the morphologic feature. Agreement was fair for enhancement pattern (0.33), moderate for lesion contour (0.53) and neovascularity (0.57), and good for calcification (0.76) (Table 2). There was a significant association between histologic subtype and enhancement pattern (p < 0.001 for both radiologists), lesion contour (radiologist 1, p < 0.004; radiologist 2, p < 0.014), and neovascularity (p < 0.001 for both radiologists). There was no association between histologic subtype and the presence of calcifications (radiologist 1, p = 0.15; radiologist 2, p = 0.21). For both radiologists, a heterogeneous enhancement pattern was most commonly noted in clear cell RCC (radiologist 1, 88%; radiologist 2, 76%) and also in oncocytoma (radiologist 1, 87%; radiologist 2, %). A homogeneous enhancement pattern was most commonly noted in papillary RCC (radiologist 1, 75%; radiologist 2, 72%), and lipidpoor angiomyolipoma (radiologist 1, 80%; radiologist 2, 70%). A smooth contour was most commonly seen in papillary RCC (radiologist 1, 78%; radiologist 2, 72%) and lipid-poor angiomyolipoma (radiologist 1, 80%; radiologist 2, 70%). Finally, neovascularity was most commonly noted in clear cell RCC, oncocytoma, and lipid-poor angiomyolipoma ( 30% of cases for each radiologist) and least commonly noted in papillary RCC (one [3%] of 36 papillary RCCs for each radiologist). There were also significant associations between several qualitative imaging features W518 AJR:203, November 2014

MDCT Differentiation of Clear Cell RCC TABLE 2: Qualitative Analysis Morphologic Feature Clear Cell RCC (n = 86) and lesion size. Lesions smaller than 3 cm were more likely to be homogeneous in enhancement pattern, smooth in contour, and without neovascularity, regardless of histologic subtype (p < 0.005 for both radiologists). Based on qualitative analysis of morphologic features, radiologist 1 predicted clear cell RCC with accuracy of 79% (130/165), sensitivity of 90% (77/86), specificity of Papillary RCC (n = 36) Chromophobe RCC (n = 10) (n = 23) 67% (53/79), PPV of 75% (77/103), and NPV of 85% (53/62). Radiologist 2 predicted clear cell RCC with accuracy of 69% (114/165), sensitivity of 90% (77/86), specificity of 47% (37/79), PPV of 60% (71/119), and NPV of 80% (37/46). Radiologist performance for predicting histologic subtypes based on qualitative analysis of morphologic features is summarized in Table 2. Lipid-Poor Angiomyolipoma (n = 10) Homogeneous enhancement Radiologist 1 10 (12) 27 (75) 5 (50) 3 (13) 8 (80) a p < 0.001 Radiologist 2 21 (24) 26 (72) 3 (30) 0 (0) 7 (70) Heterogeneous enhancement Radiologist 1 76 (88) a 9 (25) 5 (50) 20 (87) 2 (20) p < 0.001 Radiologist 2 65 (76) 10 (28) 7 (70) 23 () 3 (30) Smooth contour Radiologist 1 39 (45) 28 (78) 5 (50) 9 (39) 8 (80) a p < 0.004 Radiologist 2 37 (32) 26 (72) 5 (50) 8 (35) 7 (70) Lobular contour Radiologist 1 47 (55) 8 (22) 5 (50) 14 (61) a 2 (20) p < 0.004 Radiologist 2 49 (57) 10 (28) 5 (50) 15 (65) 3 (30) Neovascularity Radiologist 1 34 (40) a 1 (3) 2 (20) 7 (30) 4 (40) a p < 0.001 Radiologist 2 42 (49) 1 (3) 2 (20) 10 (43) 3 (30) Calcification Radiologist 1 19 (22) 2 (6) 3 (30) 3 (13) 1 (10) p = 0.151 Radiologist 2 22 (26) 3 (9) 3 (30) 5 (22) 1 (10) Accuracy Radiologist 1 130/165 (79) 146/165 (88) 146/165 (88) 144/165 (87) 160/165 (97) Radiologist 2 114/165 (69) 149/165 (90) 141/165 (85) 146/165 (88) 157/165 (95) Sensitivity Radiologist 1 77/86 (90) 25/36 (69) 3/10 (30) 5/23 (22) 6/10 (60) Radiologist 2 77/86 (90) 21/36 (58) 3/10 (30) 7/23 (30) 4/10 (40) Specificity Radiologist 1 53/79 (67) 121/129 (94) 143/155 (92) 139/142 (98) 154/155 (99) Radiologist 2 37/79 (47) 128/129 (99) 138/155 (89) 139/142 (98) 153/155 (99) PPV Radiologist 1 77/103 (75) 25/33 (76) 3/15 (20) 5/8 (63) 6/7 (86) Radiologist 2 71/119 (60) 21/22 (95) 3/20 (15) 7/10 (70) 4/6 (67) NPV Radiologist 1 53/62 (86) 121/132 (92) 143/150 (95) 139/157 (89) 154/158 (97) Radiologist 2 37/46 (80) 128/143 (90) 138/145 (95) 139/155 (90) 153/159 (96) Note Data are in numbers of lesions satisfying each category per radiologist. Data in parentheses are percentages. RCC = renal cell carcinoma. a Denotes histologic subtype most strongly associated with morphologic feature for radiologist 1; kappa values are as follows: 0.33 for enhancement pattern, 0.53 for lesion contour, 0.57 for neovascularity, and 0.76 for calcifications. Quantitative Analysis Unenhanced absolute attenuation Lipid-poor angiomyolipoma had a significantly higher mean absolute unenhanced attenuation compared with all other histologic subtypes (p < 0.01 for all comparisons) (Fig. 1 and Table 3). Of all lipid-poor angiomyolipomas, 70% (7/10) had an absolute unenhanced attenuation greater than 45 HU (range, 41.4 p AJR:203, November 2014 W519

Lee-Felker et al. TABLE 3: Multiphasic Absolute Attenuation: Distinguishing Clear Cell Renal Cell Carcinoma (RCC) From Other Renal Lesions Imaging Phase Clear Cell RCC (n = 86) 65.3 HU; 95% CI, 46.3 59.4 HU), whereas only 6% (5/86) of clear cell RCCs had unenhanced attenuation greater than 45 HU. An absolute unenhanced attenuation less than 45 HU differentiated clear cell RCC from lipid-poor angiomyolipoma with accuracy of 91.2%, sensitivity of 93.2%, specificity of 80.0%, PPV of 97.6%, and NPV of 57.1%. Absolute attenuation After IV contrast injection, 91.9% (79/86) of clear cell RCCs, 90.0% (9/10) of chromophobe RCCs, 82.6% (19/23) of oncocytomas, and 80.0% (8/10) of lipid-poor angiomyolipomas had maximal attenuation in the corticomedullary phase. Clear cell RCC had the highest mean absolute attenuation compared with all other histologic subtypes and was significantly higher in attenuation than papillary RCC (p < 0.0001). All lesions except papillary RCC decreased in attenuation on subsequent phases. Papillary RCC was the only lesion with an absolute attenuation less than 70 HU in the corticomedullary phase, and 85.0% (28/33) of papillary RCC had maximal attenuation in the nephrographic phase at a lower absolute mean attenuation than all other subtypes (Figs. 2 4). Of note, the difference in mean absolute corticomedullary attenuation between clear cell RCC and papillary RCC was more than HU. Relative attenuation When lesion attenuation was normalized to uninvolved renal cortex, clear cell RCC had significantly higher mean corticomedullary relative attenuation (20.2; 95% CI, 10.5 29.9) compared with oncocytoma ( 16.2, 24.7 to 7.8, p < 0.001), papillary RCC ( 60.1, 65.3 to 55.0, p < 0.001), and chromophobe RCC ( 21.6, 38.5 to 4.7, p < 0.005) (Fig. 5). Clear cell Papillary RCC (n = 36) Chromophobe RCC (n = 10) A (n = 23) Lipid-Poor Angiomyolipoma (n = 10) Unenhanced (HU) 29.5 (27.4 31.6) 37.4 (31.6 43.2) 30.0 (19.5 40.5) 35.1 (31.1 39.0) 52.9 (46.3 59.4) p vs clear cell RCC 0.1 1 0.1 0.001 a p vs lipid-poor angiomyolipoma 0.001 a 0.01 a 0.0005 a 0.0001 a Corticomedullary (HU) 174.4 (163.5 185.3) 62.2 (56.6 67.5) 140.6 (106.4 174.8) 151.9 (132.2 171.6) 146.0 (117.0 175.0) p vs clear cell RCC 0.001 a 0.05 0.06 0.1 Nephrographic (HU) 113.2 (107.2 119.2) 81.8 (74.2 89.5) 91.4 (71.6 111.1) 120.0 (108.6 131.4) 107.5 (90.2 124.8) p vs clear cell RCC 0.001 a 0.02 a 0.3 0.5 Excretory (HU) 87.9 (83.4 92.3) 64.5 (57.9 71.1) 71.3 (60.3 82.4) 89.8 (80.5 99.0) 83.6 (67.8 99.4) p vs clear cell RCC 0.001 a 0.02 a 0.7 0.5 Note Data in parentheses are 95% CI. a p value less than or equal to 0.01 was significant. C D Fig. 1 Multiphasic absolute attenuation of lipid-poor angiomyolipoma in 42-year-old woman. Circles show ROIs. A D, Axial contrast-enhanced MDCT images of left kidney show multiphasic attenuation during unenhanced (A), corticomedullary (B), nephrographic (C), and excretory (D) phases. B W520 AJR:203, November 2014

MDCT Differentiation of Clear Cell RCC RCC could be differentiated from all other histologic subtypes on the basis of relative attenuation greater than 0% in the corticomedullary phase with accuracy of 75% (124/165), sensitivity of 64% (55/86), specificity of 87% (69/79), PPV of 85% (55/65), and NPV of 69% (69/). Absolute enhancement and deenhancement Clear cell RCC had a greater absolute enhancement between unenhanced and corticomedullary phases (lesion ROI corticomedullary lesion ROI unenhanced ) compared with all other subtypes (p < 0.05) and a significantly greater absolute deenhancement between corticomedullary and nephrographic phases (lesion ROI corticomedullary lesion ROI nephrographic ) compared with oncocytoma (p < 0.006) and papillary RCC (p < 0.001). Deenhancement greater than 50 HU differentiated clear cell RCC from oncocytoma with accuracy of 58.7%, sensitivity of 54.7%, specificity of 73.9%, PPV of 88.7%, and NPV of 30.4%. Quantitative features in combination Because of the importance of distinguishing between clear cell RCC and benign renal masses, the predictive power of a combination of quantitative features was evaluated for lipid-poor angiomyolipoma and oncocytoma. The combination of absolute unenhanced attenuation greater than 45 HU and relative corticomedullary attenuation less than 10% differentiated lipidpoor angiomyolipoma from clear cell RCC with accuracy, sensitivity, specificity, PPV, and NPV of 95%, 70%, 98%, 78%, and 97%, respectively. The combination of absolute deenhancement greater than 50 HU or relative attenuation greater than 0% in the corticomedullary phase differentiated clear cell RCC from oncocytoma with accuracy, sensitivity, specificity, PPV and NPV of 74%, 76%, 70%, 90%, and 43%, respectively (Table 4 and Figs. 6 and 7). Discussion To date, renal lesions have been evaluated by comparing unenhanced scans with singlephase enhanced scans, without regard for optimal timing of enhancement. Furthermore, the concept of lesion deenhancement, which is important for characterizing liver and adrenal masses, has only begun to undergo more widespread investigation for renal lesions. In the largest series of its kind to our knowledge, we previously evaluated 298 benign and malignant solid renal masses, find- TABLE 4: Multiphasic Relative Attenuation: Distinguishing Clear Cell Renal Cell Carcinoma (RCC) From Other Renal Lesions Imaging Phase Clear Cell RCC (n = 86) Papillary RCC (n = 36) Chromophobe RCC (n = 10) (n = 23) Lipid-Poor Angiomyolipoma (n = 10) Unenhanced (%) 4.7 ( 4.5 to 13.6) 32.7 (10.4 55.0) 6.1 ( 38.3 to 26.1) 24.5 (5.8 43.3) 67.9 (32.1 103.7) p vs clear cell RCC 0.2 1.0 0.4 0.03 a Corticomedullary (%) 20.2 (10.5 29.9) 60.1 ( 65.3 to 55.0) 21.6 ( 38.5 to 4.7) 16.2 ( 24.7 to 7.8) 15.6 ( 30 to 1.3) p vs clear cell RCC 0.001 a 0.005 a 0.001 a 0.02 Nephrographic (%) 24.3 ( 28.6 to 19.9) 48.4 ( 52.8 to 43.9) 48.7 ( 59.4 to 37.9) 23.2 ( 32.1 to 14.3) 36.9 ( 45.2 to 28.5) p vs clear cell RCC 0.001 a 0.01 a 0.8 0.05 Excretory (%) 5.8 ( 45.1 to 56.6) 41.8 ( 47.8 to 35.9) 41.7 ( 51.7 to 31.7) 12.5 ( 24.0 to 0.9) 27 ( 42.9 to 11.0) p vs clear cell RCC 0.3 0.5 0.7 0.7 Note Relative attenuation = [(lesion ROI cortex ROI) / cortex ROI] %. Data in parentheses are 95% CI. a p less than or equal to 0.01 was significant. A C D Fig. 2 Multiphasic absolute attenuation of clear cell renal cell carcinoma (RCC) in 74-year-old man. Circles show ROIs. A D, Axial contrast-enhanced MDCT images of right kidney show multiphasic attenuation during unenhanced (A), corticomedullary (B), nephrographic (C), and excretory (D) phases. There is also peripheral simple cyst. B AJR:203, November 2014 W521

Lee-Felker et al. ing that most lesions had maximal attenuation in the corticomedullary phase, with significant differences in mean attenuation between clear cell RCC and oncocytoma as well as between clear cell RCC and other RCC subtypes [11]. This study, as well as MRI studies by Sun et al. [20] and Vargas et al. [2], advanced the concept of multiphasic renal mass attenuation profiles for individual histologic subtypes, proposed initially but not validated in a small series by Jinzaki et al. [21]. We expanded on our previous study by including lipid-poor angiomyolipomas, adjusting lesion attenuation to normal renal cortex attenuation and evaluating solid renal lesions for significant differences on the basis of enhancement and deenhancement. We also improved on the previous study design by using a uniform four-phase MDCT protocol, which may explain why we observed chromophobe RCCs to have maximal attenuation in the corticomedullary phase in the current study, whereas we previously found maximal attenuation in the nephrographic phase. Finally, we further improved on our prior study design by using statistical analysis that adjusted for multiple comparisons. Overall, the results of our current study validate our previous work in that individual histologic subtypes have characteristic attenuation profiles across the unenhanced, corticomedullary, nephrographic, and excretory phases. Although there were significant relationships among histologic subtypes and qualitative imaging features, small lesions were more likely to be homogeneous in enhancement and smooth in contour and less likely to show neovascularity, regardless of histologic subtype. Radiologist performance for predicting histologic subtype on the basis of morphologic features alone was suboptimal, with PPVs for most subtypes less than 75%. These results suggest that qualitative analysis alone is not a reliable means for differentiating histologic subtypes, especially for lesions smaller than 3 cm. Quantitative analysis was a critical adjunct for differentiating the most common solid renal mass subtypes. On the basis of the significantly higher relative corticomedullary attenuation of clear cell RCC, we were able to differentiate clear cell RCC from all other histologic subtypes. On the basis of the significantly higher absolute unenhanced attenuation of lipid-poor angiomyolipoma, we were able to differentiate lipid-poor angiomyolipoma from all other histologic subtypes. In fact, all lipidpoor angiomyolipomas had absolute attenuation greater than 40 HU on the unenhanced phase, lending % NPV to this feature. The hyperdense appearance of lipid-poor angiomyolipoma has been described in several studies [22, 23], most recently by Yang et al. [24], who concluded that unenhanced high attenuation was the only parameter that consistently and significantly differentiated lipid-poor angiomyolipoma and clear cell RCC. We expanded on the study by Yang et al. by evaluating the predictive power of selected features in combination. The combination of A C unenhanced attenuation greater than 45 HU and relative attenuation less than 10% in the corticomedullary phase enabled differentiation of lipid-poor angiomyolipoma from clear cell RCC with 98% specificity and 97% NPV. In a lesion with these imaging features, the likelihood of clear cell RCC is only 3%. We were also able to distinguish clear cell RCC from oncocytoma with high PPV on the basis of two significant features: absolute deenhancement from the corticomedullary to nephrographic phases and relative attenuation in the corticomedullary phase. Clear cell RCC Fig. 3 Multiphasic absolute attenuation of oncocytoma in 79-year-old man. A D, Axial contrast-enhanced MDCT images of left kidney show multiphasic attenuation during unenhanced (A), corticomedullary (B), nephrographic (C), and excretory (D) phases. B D W522 AJR:203, November 2014

MDCT Differentiation of Clear Cell RCC Attenuation (HU) 200 180 160 140 120 80 60 40 20 0 Unenhanced Clear cell RCC Papillary RCC Chromophobe RCC Lipid-poor angiomyolipoma Corticomedullary Nephrographic Excretory Imaging Phase Fig. 4 Chart shows multiphasic absolute attenuation of renal mass subtypes. Data represent mean absolute attenuation values. Error bars represent 95% CIs. RCC = renal cell carcinoma. Absolute Unenhanced Attenuation 80 60 40 20 0 0 Relative Corticomedullary Attenuation (%) Lipid-poor angiomyolipoma Clear cell RCC Attenuation (%) 120 40 60 80 had greater absolute deenhancement than did oncocytoma and significantly higher relative attenuation in the corticomedullary phase. A recent study by Kopp et al. [25] evaluated a CT washout formula for differentiating clear cell RCC from other renal masses. Those authors found rapid washout in clear cell RCC; however, they did not observe this to be significantly different from that seen in oncocytoma, likely because the MDCT protocol in their study consisted only of unenhanced, nephrographic, and excretory phases, excluding the corticomedullary phase, which we and others have found critical for differentiating solid renal masses [2, 11, 20]. A quantitative means for distinguishing oncocytoma from clear cell RCC is advantageous because oncocytoma is notoriously variable in its qualitative imaging features. Although our data suggest that a lesion with positive relative corticomedullary attenuation or absolute deenhancement greater than 50 HU is more likely to be clear cell RCC than an oncocytoma, there is still considerable overlap among these two groups with regard to their quantitative enhancement features. In most cases biopsy or further imaging, such as recently promising arterial spin-labeling MRI [26], will be necessary to differentiate them. Our study has several limitations. Its retrospective design may introduce selection bias. Our institution is a tertiary referral center in which patients may have a higher prevalence of malignancy compared with the general 200 80 60 40 20 0 20 Clear cell RCC Papillary RCC Chromophobe RCC Lipid-poor angiomyolipoma Unenhanced Corticomedullary Nephrographic Excretory Imaging Phase Fig. 5 Chart shows multiphasic relative attenuation of renal mass subtypes. Relative enhancement = [(lesion ROI cortex ROI) / cortex ROI] %]. Error bars represent 95% CIs. RCC = renal cell carcinoma. Deenhancement 200 150 50 0 50 population. We included renal masses that were pathologically proven with percutaneous 18-gauge needle biopsy, which may introduce sampling error. Another limitation is the relatively small number of lipid-poor angiomyolipomas because of the low incidence of this lesion. Although our data suggest that lipidpoor angiomyolipomas may be noninvasively diagnosed on the basis of the combination of unenhanced attenuation and corticomedullary enhancement pattern with a high degree of confidence, these results should be confirmed in larger studies. In addition, the fourphase MDCT renal mass protocol results in a high patient radiation dose and should be reserved only for cases in which lesion discrimination is required before surgery or ablation. 0 Relative Corticomedullary Attenuation (%) Clear cell RCC 200 Fig. 6 Scatterplot shows attenuation profile for lipid-poor angiomyolipoma versus clear cell renal cell carcinoma (RCC). Blue shaded area represents those renal masses with unenhanced absolute attenuation greater than 45 HU and relative corticomedullary attenuation less than 10%. Relative enhancement = [(lesion ROI cortex ROI) / cortex ROI] %. Fig. 7 Scatterplot shows attenuation profiles for clear cell RCC versus oncocytoma (RCC). Blue shaded area represents those masses with absolute deenhancement greater than 50 HU or relative corticomedullary attenuation greater than 0%. Absolute deenhancement = lesion ROI corticomedullary lesion ROI nephrographic. AJR:203, November 2014 W523

Lee-Felker et al. Of our patients, 53.2% were enrolled in our previous study, which may have introduced recall bias. Finally, interradiologist agreement was only fair to good for qualitative analysis, although it is similar to that reported in prior studies [12]. This inherent subjectivity in classifying renal masses on the basis of their qualitative features alone underscores the importance of establishing quantitative means to strengthen the discriminatory power. In this study, we have shown that fourphase MDCT may uncover distinct and reproducible patterns of attenuation among the most common solid renal cortical masses before and after contrast injection. Rather than merely evaluate attenuation of renal lesions in a single phase, a multiphasic renal mass attenuation profile is developed for each subtype, encompassing the unenhanced, corticomedullary, nephrographic, and excretory phases. A smoothly contoured, hyperdense (unenhanced attenuation greater than 45 HU) renal mass that shows homogeneous peak attenuation in the corticomedullary phase is most consistent with lipid-poor angiomyolipoma. Clear cell RCC enhances heterogeneously but avidly with the highest magnitude of absolute attenuation in the corticomedullary phase and with significantly higher relative corticomedullary attenuation than all other subtypes. Furthermore, clear cell RCC deenhances to a greater degree than does oncocytoma, with a decrease in attenuation greater than 50 HU highly predictive of clear cell RCC. Chromophobe RCC and oncocytoma show peak attenuation in the corticomedullary phase at an intermediate magnitude and show smaller magnitude deenhancement thereafter. Finally, papillary RCC is mildly hyperdense on the unenhanced phase and enhances homogeneously but weakly with absolute attenuation less than 70 HU in the corticomedullary phase and has peak attenuation in the nephrographic phase at lower absolute and relative attenuations than all other subtypes. On the basis of our findings, we believe that four-phase MDCT enables development of multiphasic renal mass attenuation profiles that help distinguish clear cell RCC from other solid renal cortical masses, most notably lipid-poor angiomyolipoma. References 1. American Cancer Society. Cancer facts & figures 2013. Atlanta, GA: American Cancer Society, 2013:13 2. 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; 264:779 788 3. Kovacs G, Akhtar M, Beckwith BJ, et al. The Heidelberg classification of renal cell tumours. J Pathol 1997; 183:131 133 4. Snyder ME, Bach A, Kattan TT, et al. Incidence of benign lesions for clinically localized renal masses smaller than 7 cm in radiological diameter: influence of sex. J Urol 2006; 176:2391 2395 5. Truong LD, Shen SS. Immunohistochemical diagnosis of renal neoplasms. Arch Pathol Lab Med 2011; 135:92 109 6. Cheville JC, Lohse CM, Zincke H, et al. Comparisons of outcome and prognostic features among histologic subtypes of renal cell carcinoma. Am J Surg Pathol 2003; 27:612 624 7. Moch H, Gasser T, Amin MB, et al. Prognostic utility of the recently recommended histologic classification and revised TNM staging system of renal cell carcinoma: a Swiss experience with 588 tumors. Cancer 2000; 89:604 614 8. Hoffmann NE, Gillett MD, Cheville JC, et al. Differences in organ system of distant metastasis by renal cell carcinoma subtype. J Urol 2008; 179:474 477 9. Weiss LM, Gelb AB, Medeiros LJ. Adult renal epithelial neoplasms. Am J Clin Pathol 1995; 103:624 635 10. Tse G, Chow DS, Hsu M, et al. Multidetector computed tomographic features of oncocytic papillary renal cell carcinoma, a new subtype. J Comput Assist Tomogr 2010; 34:380 384 11. Young JR, Margolis D, Sauk S, Pantuck AJ, Sayre J, Raman SS. Clear cell renal cell carcinoma: discrimination from other renal cell carcinoma subtypes and oncocytoma at multiphasic multidetector CT. Radiology 2013; 267:444 453 12. Zhang J, Lefkowitz RA, Ishill NM, et al. Solid renal cortical tumors: differentiation with CT. Radiology 2007; 244:494 504 13. Sheir KZ, El-Azab M, Mosbah A, et al. Differentiation of renal cell carcinoma subtypes by multislice computerized tomography. J Urol 2005; 174:451 455 14. Kim JK, Kim TK, Ahn HJ, et al. Differentiation of subtypes of renal cell carcinoma on helical CT scans. AJR 2002; 178:1499 1506 15. Ruppert-Kohlmayr AJ, Uggowitzer M, Meissnitzer T, et al. Differentiation of renal clear cell carcinoma and renal papillary carcinoma using quantitative CT enhancement parameters. AJR 2004; 183:1387 1391 16. Bird VG, Kanagarajah P, Morillo G, et al. Differentiation of oncocytoma and renal cell carcinoma in small renal masses (<4cm): the role of 4-phase computerized tomography. World J Urol 2011; 29:787 792 17. Yuh BI, Cohn RH. Different phases of renal enhancement: role in detecting and characterizing renal masses during helical CT. AJR 1999; 173:747 755 18. Wildberger JE, Adam G, Boeckmann W, et al. Computed tomography characterization of renal cell tumors in correlation with histopathology. Invest Radiol 1997; 32:596 601 19. Davidson AJ, Hayes WS, Hartman DS, et al. Renal oncocytoma and carcinoma: failure of differentiation with CT. Radiology 1993; 186:693 696 20. Sun MR, Ngo L, Genega EM, et al. Renal cell carcinoma: dynamic contrast-enhanced MR imaging for differentiation of tumor subtypes: correlation with pathologic findings. Radiology 2009; 250:793 802 21. Jinzaki M, Tanimoto A, Mukai M, et al. Doublephase helical CT of small renal parenchymal neoplasms: correlation with pathologic findings and tumor angiogenesis. J Comput Assist Tomogr 2000; 24:835 842 22. Jinzaki M, Tanimoto A, Narimatsu Y, et al. Angiomyolipoma: imaging findings in lesions with minimal fat. Radiology 1997; 205:497 502 23. Kim JK, Park SY, Shon JE, et al. Angiomyolipoma with minimal fat: differentiation from renal cell carcinoma at biphasic helical CT. Radiology 2004; 230:677 684 24. Yang CW, Shen SH, Chang YH, et al. Are there useful CT features to differentiate renal cell carcinoma from lipid-poor renal angiomyolipoma? AJR 2013; 201:1017 1028 25. Kopp RP, Aganovic L, Palazzi KL, Cassidy FH, Sakamoto K, Derweesh IH. Differentiation of clear from non-clear cell renal cell carcinoma using CT washout formula. Can J Urol 2013; 20:6790 6797 26. Lanzman RS, Robson PM, Sun MR, et al. Arterial spin-labeling MR imaging of renal masses: correlation with histopathologic findings. Radiology 2012; 265:799 808 W524 AJR:203, November 2014