Histogram Analysis of Whole-Lesion Enhancement in Differentiating Clear Cell from Papillary Subtype of Renal Cell Cancer 1

Transcription

1 Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at Original Research n Genitourinary Imaging Hersh Chandarana, MD Andrew B. Rosenkrantz, MD Thais C. Mussi, MD Sooah Kim, MD Afshan A. Ahmad, MD Sean D. Raj, BA John McMenamy, MD Jonathan Melamed, MD James S. Babb, PhD Berthold Kiefer, PhD Atilla P. Kiraly, PhD 1 From the Departments of Radiology (H.C., A.B.R., T.C.M., S.K., A.A.A., S.D.R., J. McMenamy., J.S.B.) and Pathology (J. Melamed), New York University Langone Medical Center, 550 First Ave, HW201, New York, NY 10016; MR Application and Workflow Development, Siemens Healthcare Sector, Erlangen, Germany (B.K.); and Siemens Corporate Research, Princeton, NJ (A.P.K.). Received June 17, 2011; revision requested August 5; revision received March 30, 2012; accepted May 4; final version accepted June 1. Address correspondence to H.C. ( hersh. chandarana@nyumc.org). q RSNA, 2012 Histogram Analysis of Whole-Lesion Enhancement in Differentiating Clear Cell from Papillary Subtype of Renal Cell Cancer 1 Purpose: Materials and Methods: Results: Conclusion: To compare histogram analysis of voxel-based wholelesion (WL) enhancement to qualitative assessment and region-of-interest (ROI)-based enhancement analysis in discriminating the renal cell cancer (RCC) subtype clear cell RCC (ccrcc) from papillary RCC (prcc). In this institutional review board approved, HIPAA-compliant retrospective study, 73 patients underwent magnetic resonance (MR) imaging prior to surgery for RCC between January 2007 and January Three-dimensional fatsuppressed T1-weighted gradient-echo corticomedullary phase acquisitions, obtained before and after contrast agent administration, were transferred to a workstation at which automated registration followed by semiautomated segmentation of the RCC was performed. Percent enhancement was computed on a per-voxel basis: (SI post SI pre )/SI pre 100, where SI pre and SI post indicate signal intensity before and after contrast enhancement, respectively. The WL quantitative parameters of mean, median, and third quartile enhancement and histogram distribution parameters kurtosis and skewness were computed for each lesion. WL enhancement parameters were compared with ROI-based analysis and qualitative assessment with regards to diagnostic accuracy and interreader agreement in differentiating ccrcc from prcc. There were 19 prccs and 55 ccrccs at pathologic examination. ccrcc had significantly higher WL mean, median, and third quartile enhancement compared with prcc and hade significantly lower kurtosis and skewness (all P,.001). Third quartile enhancement had the highest accuracy (94.6%; area under the curve, 0.980) in discriminating ccrcc from prcc, which was significantly higher than the accuracy of qualitative assessment (86.0%; P =.04) but not significantly higher than that of ROI enhancement (89.2%; P =.52). WL enhancement parameters had higher interreader agreement (k = ) compared with ROI enhancement or qualitative assessment (k = 0.83 and 0.7, respectively) in discriminating ccrcc from prcc. WL enhancement histogram analysis is feasible and can potentially be used to differentiate ccrcc from prcc with high accuracy. q RSNA, 2012 Supplemental material: /suppl/doi: /radiol /-/dc1 790 radiology.rsna.org n Radiology: Volume 265: Number 3 December 2012

2 Renal masses are increasingly being discovered incidentally as the use of medical imaging rises (1 4). This incidental discovery of renal neoplasms can lead to a management dilemma because these tumors have variable natural history, growth rate, histopathologic features, and clinical behavior (5 8). The two common types of renal cell cancer (RCC) are clear cell RCC (ccrcc) and papillary RCC (prcc), which account for approximately 70% and 20% of all RCCs, respectively. These two subtypes differ in biologic behavior and prognosis, with ccrcc known to have a relatively more aggressive behavior and worse prognosis compared with prcc (9,10). Furthermore, these two subtypes respond differently to available targeted therapies in advanced disease; tyrosine kinase inhibitors are more effective in patients with advanced ccrcc, whereas mammalian target of rapamycin inhibitor temsirolimus has been recently shown to be more effective against advanced prcc (11 13). These differences in biologic behavior and response to targeted therapies underscore the need to accurately differentiate between these two common subtypes of RCC. Although Advances in Knowledge nn Analysis of voxel-based wholelesion enhancement histogram parameters can be used to discriminate between clear cell and papillary subtypes of renal cell cancers; the whole-lesion parameter of third quartile enhancement had the highest accuracy (area under the curve [AUC], 0.98) with sensitivity of 96% and specificity of 90%. nn Whole-lesion enhancement histogram distribution parameters kurtosis and skewness can be used to discriminate clear cell from papillary renal cell cancer with high accuracy (AUC, and 0.938, respectively), without reliance on the absolute measure of signal intensity change. percutaneous biopsy of the renal tumor can enable accurate histologic diagnosis in many patients (14 16), potential for sampling error and associated procedural complications has hindered widescale clinical acceptance. Furthermore, in patients in whom watchful waiting is being considered as a management option, biopsy remains an invasive method for making a diagnosis. Dynamic contrast material enhanced computed tomography (CT) or magnetic resonance (MR) imaging provides a useful method of differentiating ccrcc from prcc, as studies have shown that prcc is a less-vascularized lesion compared with ccrcc (17 22). Although it may be possible to discriminate between these tumors on the basis of visual or qualitative inspection of contrast enhancement, quantitative methods of measuring enhancement may provide a higher degree of accuracy with less subjective variability. Recently, a study found that a region-of-interest (ROI)-based method of assessing enhancement demonstrated high accuracy in differentiating ccrcc from prcc (19). However, some of the limitations of such ROI-based methods include interobserver variability in ROI placement, technical problems such as misregistration between pre- and postcontrast acquisitions, and difficulty with appropriate ROI placement on an unenhanced image for T1-isointense intrarenal tumors. Hence, a tool that can perform automatic registration, lesion segmentation, and whole-lesion (WL) enhancement analysis can overcome many of the above-mentioned limitations of ROI-based assessment. Furthermore, analysis of WL enhancement histogram Implication for Patient Care nn Ability to noninvasively discriminate clear cell from papillary renal cell cancer with analysis of whole-lesion enhancement histogram parameters can potentially affect patient management, as these two subtypes have different prognoses and respond differently to available chemotherapy in advanced disease. distribution may be able to be used to discriminate ccrcc from prcc without reliance on the absolute measure of signal intensity change. Therefore, the purpose of our study was to compare histogram analysis of voxel-based WL enhancement to qualitative assessment and ROI-based enhancement analysis in discriminating ccrcc from prcc. Materials and Methods Patients Siemens Medical Solution and Siemens Corporate Research (Princeton, NJ) provided research support in the form of the software (MROncoTreat) used to perform automated registration and semiautomated WL segmentation. The nine authors who are not employees of Siemens had full control of the data. This Health Insurance Portability and Accountability Act compliant retrospective single-center study was performed after obtaining a waiver of informed patient consent from our institutional review board. A retrospective review of the pathology database was performed for patients who underwent surgery for renal neoplasm (either partial or total Published online /radiol Radiology 2012; 265: Content code: Abbreviations: AUC = area under the curve ccrcc = clear cell RCC prcc = papillary RCC RCC = renal cell cancer ROI = region of interest VIBE = volumetric interpolated breath-hold examination WL = whole lesion Author contributions: Guarantor of integrity of entire study, H.C.; study concepts/ study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; literature research, H.C., S.D.R., A.P.K.; clinical studies, H.C., T.C.M., S.K., A.A.A., S.D.R., J. McMenamy, J. Melamed, A.P.K.; statistical analysis, H.C., J.S.B., A.P.K.; and manuscript editing, H.C., A.B.R., S.K., A.A.A., S.D.R., J. McMenamy, J.S.B., B.K., A.P.K. Conflicts of interest are listed at the end of this article. Radiology: Volume 265: Number 3 December 2012 n radiology.rsna.org 791

3 nephrectomy) at our institution from January 2007 to January 2010 with diagnosis of either ccrcc or prcc at pathologic evaluation. Patients who underwent preoperative contrast material enhanced MR examination performed within 180 days prior to surgery were included. A total of 254 patients underwent surgery and had diagnosis of either ccrcc (n = 202) or prcc (n = 52). In 101 of these patients, MR imaging was performed prior to the surgery at our institution. Patients were excluded for following reasons: In 21 patients MR imaging was performed with a protocol other than a dedicated renal mass protocol or without intravenous gadolinium-based contrast material, in three patients the renal lesion was incompletely imaged, and four patients had a lesion smaller than 1 cm. Our study cohort consisted of 73 patients (48 male, with mean age of 60 years and range of years; 25 female, with mean age of 63 years and range of years), with a mean age of 61 years (age range, years). In one patient there were two lesions, one in either kidney, and hence 74 renal lesions were evaluated. For this study, histopathologic slides were re-reviewed by a pathologist (J. Melamed, with 19 years of experience in genitourinary histopathology) who rendered the diagnosis of ccrcc or prcc. Immunohistochemical staining was performed when deemed necessary at the discretion of the pathologist. MR Imaging MR imaging was performed with a 1.5- T clinical system (Magnetom Sonata, Symphony, or Avanto; Siemens Healthcare, Erlangen, Germany) by using torso phased-array coils. The routine renal protocol included following sequences: transverse breathhold T1-weighted in-phase and opposedphase sequence, transverse and coronal breath-hold T2-weighted sequence with half-fourier single-shot turbo spin-echo, and transverse three-dimensional fatsuppressed T1-weighted interpolated spoiled gradient echo (volumetric interpolated breath-hold examination [VIBE]) sequence. Imaging parameters were as follows: repetition time msec/ echo time msec, / ; flip angle, 12 ; matrix interpolated to ; field of view, mm, with rectangular field of view; and interpolated section thickness, 2 3 mm. The VIBE sequence was performed prior to and three times after administration of an intravenously administered gadolinium chelate (gadopentetate dimeglumine, Magnevist; Bayer Healthcare Pharmaceuticals, Wayne, NJ). Acquisition delays were time-to-peak (TTP) for the corticomedullary phase, TTP plus 60 seconds for the nephrographic phase, and TTP plus 180 seconds for the excretory phase acquisition. Dynamic injection of 0.1 mmol gadopentetate dimeglumine per kilogram body weight was administered with a power injector (Spectris; Medrad, Pittsburgh, Pa) at a rate of 2 ml/ sec followed by a 20-mL saline flush (also administered at a rate of 2 ml/sec). A 1-mL test bolus was administered to determine TTP arterial enhancement of the aorta at the level of the renal artery. Precontrast and postcontrast images were obtained at end expiration. Acquisition time for the VIBE sequence images averaged 17 seconds. MR Image Evaluation Qualitative assessment. Three radiologists (A.A.A., J. McMenamy, and S.K., with 2 11 years of experience in the interpretation of abdominal MR images), who were blinded to the pathologic findings, independently evaluated pre- and postcontrast acquisitions including subtraction data sets (23). A training session was performed in which readers were shown five examples of ccrcc and prcc subtypes. These example cases were not included in the study cohort. Specifically, all readers were made aware that ccrcc enhances heterogeneously and avidly (Fig 1), such that a lesion with a portion of the tumor that demonstrates contrast enhancement similar to or higher than the renal cortex in the corticomedullary phase is diagnostic of ccrcc. A prcc, on the other hand, demonstrates low-level homogeneous enhancement that may be progressive on images obtained at later phases (Fig 2) (17,19). Quantitative evaluation of WL enhancement. Two readers (T.C.M. and H.C., with 3 and 4 years of experience, respectively, in interpretation of abdominal MR images), who were blinded to histopathologic diagnosis, independently performed WL enhancement evaluation. Pre- and postcontrast corticomedullary phase acquisitions were transferred to a research application (MROncoTreat, Siemens Corporate Research). The software was developed utilizing elements of the Open Inventor (24) for graphics display. A deformable registration was performed first by using flows of diffeomorphisms with a local cross-correlation cost criterion and Gaussian-based regularization (25). This technique iteratively modifies a deformation field on the basis of the results of the cost function comparing the two volumes. The advantage of the local cross-correlation cost function is that it allows for correlations to be established over images of different MR modalities or varying contrast enhancement. The registration was applied to the precontrast and postcontrast acquisition. The results of the registration were visible to the operator to evaluate. Registration was robust and no manual interactions were performed at this stage. Subsequently, lesions were segmented by using an interactive random walk approach on the registered data set (26). This method consists of placing internal and external seeds in and around the lesion. Seeds were placed on two axial sections; one through the superior portion of the lesion and the other through the inferior portion of the lesion. The segmentation algorithm first computes a probability map of the points on the image either belonging to the inside or outside seeds. A threshold of this probability map taken at 50% determines the segmentation. Result of the segmentation for each lesion was reviewed by the radiologist and if necessary the segmentation contour was manually modified. Lesion contour was manually modified in 12 cases by reader 1 and seven cases by reader 2. Time required from initial seed placement to manual correction of contours was recorded for each lesion. 792 radiology.rsna.org n Radiology: Volume 265: Number 3 December 2012

4 Figure 1 Figure 1: Findings in a 38-year-old man with renal lesion (arrow) in the right kidney shown to be ccrcc at pathologic evaluation. (a) On precontrast T1-weighted fat-saturated gradient-echo VIBE MR image the tumor is hypointense, and (b) on postcontrast corticomedulary phase MR image it demonstrates avid enhancement. (c) Histogram of WL enhancement demonstrates a flat peak consistent with low kurtosis value of 20.5 and low skewness value of Next, voxel-based contrast enhancement values were collected from the lesion segmentation. Percent enhancement was computed by means of the following formula: (SI post SI pre )/SI pre 100, where SI pre and SI post indicate signal intensity of each voxel obtained before and after contrast enhancement, respectively. A histogram of the voxel-based enhancement values was displayed and the mean, median, and third quartile enhancement of the sorted values was computed. Additionally, a spreadsheet of all of the values was created, which was used to compute histogram distribution parameters kurtosis and skewness offline (Appendix E1 [online]). Quantitative ROI enhancement eval uation. Pre- and postcontrast corticomedullary acquisitions were transferred to a stand-alone workstation (Leonardo; Siemens Health Care). Two readers (T.C.M. and S.K., with 3 and 11 years of experience, respectively, in interpretation of abdominal MR images), who were blinded to the histopathologic diagnosis, independently place a small ROI in the tumor. The intratumoral ROI was placed within the region of the tumor that demonstrated most avid contrast enhancement during the corticomedullary phase. This ROI was copied to the precontrast acquisition. Manual correction was performed owing to respiratory or patient motion such that the same region of the tumor was sampled on the precontrast acquisition. Mean ROI size was 70 mm 2, with a range of mm 2. Signal intensity was measured on precontrast and corticomedullary phase acquisitions. ROI enhancement was calculated as (SI post SI pre )/SI pre 100, as above. Statistical Analysis For qualitative assessment, sensitivity, specificity, and accuracy for differentiating ccrcc from prcc were calculated for each reader and for the three readers summed together. The k coefficient was calculated to assess interreader agreement. An exact Mann-Whitney test was used to compare lesion subtypes (ccrcc and prcc) with respect to quantitative measures. The diagnostic test criterion of each measure observed to be optimal in maximizing the average value of sensitivity and specificity for the detection of the ccrcc subtype was calculated. Receiver operating characteristic and logistic regression analyses were conducted to assess the ability of the imaging measures, alone or in combination, to be used to discriminate between papillary and clear-cell cancer subtypes. The lower and upper limits of a 95% confidence interval for the sensitivity and specificity of each measure were also computed from generalized estimating equation. Logistic regression for correlated data was used to characterize and Radiology: Volume 265: Number 3 December 2012 n radiology.rsna.org 793

5 Figure 2 Figure 2: Findings in a 60-year-old man with a renal lesion (arrow) in the left kidney shown to be prcc at pathologic evaluation. (a) On precontrast T1-weighted fat-saturated gradient-echo VIBE image this tumor is T1-hyperintense. (b) On postcontrast corticomedulary phase image it is difficult to discern enhancement visually, but (c) subtracted data set demonstrates low-level enhancement. (d) Histogram of WL enhancement demonstrates peaked distribution with high kurtosis value of 3.73; this distribution was positively skewed (skewness, 1.94). accuracy for discriminating ccrcc from prcc. All reported P values were two sided, and a statistically significant difference was defined at P,.05. Statistical software (SAS 9.0; SAS Institute, Cary, NC) was used for all computations. compare the qualitative and quantitative assessments in terms of diagnostic accuracy for the detection of ccrcc. Generalized estimating equation based on a binary logistic regression model was used to model the likelihood of a correct diagnosis as a function of the type of information on which the diagnosis was based. The dependent variable was the indicator of concordance between the reference standard (histopathologic evaluation) diagnosis and the diagnosis derived for the same patient by using each qualitative and quantitative assessment; diagnostic test results for the quantitative assessments were based on the threshold levels from receiver operating characteristic analysis. The correlation structure was modeled by assuming results to be correlated only when derived for the same patient. ROI enhancement and WL enhancement histogram parameters were each rendered binary by classifying the result as test-positive or test-negative depending if the results were above or below a specific threshold. The threshold for each metric was chosen by a receiver operating characteristic analysis which identified the threshold that maximized Results Patients Seventy-four renal masses, ranging in diameter from 1.1 to 12.5 cm (mean, 4.7 cm) were evaluated in 73 patients. Histologic diagnosis of ccrcc was made in 55 cases, and prcc was diagnosed in 19. On precontrast T1-weighted images, 37 tumors were isointense; 12 were hyperintense; 14 lesions were heterogeneous with some hyperintense foci; and 11 were hypointense to the renal cortex. 794 radiology.rsna.org n Radiology: Volume 265: Number 3 December 2012

6 Qualitative Assessment Sensitivity, specificity, and accuracy summed for three readers was 84.2% (139 of 165), 91.2% (52 of 57), and 86.0% (191 of 222), respectively (Table 1). The mean k coefficient over the three reader pairs was 0.70 (range: ). There were discordant readings in nine (12%) cases between readers 1 and 2, in eight (11%) cases between readers 1 and 3, and in 12 (16%) cases between readers 2 and 3. ROI Enhancement Analysis ccrcc had significantly higher ROI enhancement compared with prcc for reader 1 (mean 6 standard deviation, vs ; P,.001) and reader 2 ( vs ; P,.001). Data summed across the readers demonstrated sensitivity of 90.9% (100 of 110), specificity of 84.2% (32 of 38), and accuracy of 89.2% (132 of 148) for diagnosis of ccrcc with an ROI enhancement cutoff value greater than 95. There was good agreement between the two readers for discriminating ccrcc from prcc, with a k coefficient of WL Enhancement Histogram Analysis Segmentation was semiautomated, requiring on average 80 seconds (range, seconds) of user interaction per lesion. ccrcc had significantly higher (P,.001) mean, median, and third quartile enhancement on corticomedullary phase images when compared with prcc (Table 2). Third quartile enhancement had the highest accuracy for diagnosing ccrcc, with an area under the curve (AUC) of 0.980, followed by mean enhancement, which had an AUC of By using a third quartile enhancement cutoff value of greater than 100, diagnosis of ccrcc could be made with sensitivity and specificity of 96.4% and 89.5%, respectively (Table 3). The WL enhancement histogram distribution parameter kurtosis was significantly lower in ccrcc compared with prcc ( vs ; P,.0001) (Table 2, Figs 1c and 2d). WL enhancement kurtosis enabled discrimination of ccrcc from prcc with high accuracy (AUC of 0.954) Table 1 Discrimination of ccrcc from prcc at Qualitative Assessment of Enhancement by Three Readers Reader Sensitivity (%) Specificity (%) Accuracy (%) (47/55) 89.4 (17/19) 86.4 (64/74) (48/55) 89.4 (17/19) 87.8 (65/74) (44/55) 94.7 (18/19) 83.8 (62/74) Note. Numbers in parentheses are raw data used to calculate percentages. Table 2 Quantitative WL Enhancement Histogram Parameters in ccrcc and prcc WL Enhancement Measure ccrcc prcc P Value Mean ,.001 Median ,.001 Third quartile ,.001 Kurtosis ,.001 Skewness ,.001 Note. Data are mean 6 standard deviation, unless otherwise indicated. (Fig 3). With a cutoff value of 2.1 or less, ccrcc could be diagnosed with a sensitivity and specificity of 90.9% and 89.5%, respectively (Table 3). Similarly, histogram distribution parameter skewness was significantly lower in ccrcc compared with prcc ( vs ; P,.0001) (Table 2 and Fig 3). With a skewness cutoff of 1.3 or less, ccrcc could be differentiated from prcc with a sensitivity and specificity of 89.1% and 84.2%, respectively (AUC of 0.938). There were no significant differences among the histogram distribution and other histogram enhancement parameters in the ability to discriminate ccrcc from prcc (P..09). Combination of histogram enhancement and distribution parameters did not improve the ability to use each measure to independently predict whether a patient had papillary or clear cell subtype of RCC. There was near perfect interreader agreement in discriminating ccrcc from prcc with k coefficients ranging from 0.91 to 1.0 for the WL enhancement parameters. Kurtosis had the highest interreader agreement, with a k coefficient of 1.0. Comparison of Three Methods The WL parameter of third quartile enhancement had significantly higher sensitivity and accuracy for diagnosing ccrcc when compared with qualitative assessment (P,.05) (Figs 4, 5). There was no difference in specificity between the two methods (P =.62). Qualitative assessment resulted in discordant diagnosis in 14.9% of cases when compared with third quartile enhancement. ROI enhancement had significantly higher sensitivity (P =.01) for diagnosis of ccrcc when compared with qualitative assessment, but there was no significant difference in specificity or accuracy (P..05). Although WL enhancement had higher sensitivity, specificity, and accuracy for diagnosis of ccrcc compared with ROI enhancement, this was not statistically significant (P..05). Discussion ccrccs enhance avidly and heterogeneously when compared with papillary subtypes, which demonstrate low-level homogeneous contrast enhancement Radiology: Volume 265: Number 3 December 2012 n radiology.rsna.org 795

7 Table 3 Receiver Operating Characteristic Analysis of WL Enhancement Histogram Parameters in Discriminating ccrcc from prcc WL Enhancement Measure Cutoff Value Sensitivity (%) Specificity (%) Accuracy (%) AUC Mean (49/55) [73.5, 92.6] 100 (19/19) [87.3, 100] Median (46/55) [71.4, 91.3] 100 (19/19) [87.3, 100] Third quartile (53/55) [86.6, 99.1] 89.5 (17/19) [89.5, 89.5] Kurtosis (50/55) [80.0, 96.2] 89.5 (17/19) [43.8, 98.9] Skewness (49/55) [77.8, 95.0] 84.2 (16/19) [37.4, 97.9] Note. Numbers in parentheses are raw data used to calculate percentages, and numbers in brackets are 95% confidence intervals. Figure 3 Figure 3: Boxplots of (a) kurtosis and (b) skewness in ccrcc and prcc. Top and bottom of boxes = 25th and 75th percentiles, line in box = median. (17,19,21,28). Although qualitative evaluation of contrast enhancement can help discriminate ccrcc from prcc, this assessment is subjective with considerable interreader variability in sensitivity and specificity as demonstrated in our study, in which there were discordant readings in greater than 10% of cases. Quantitative measurement of enhancement may improve overall accuracy as well as overcome subjective interreader variability in making a diagnosis of ccrcc. The WL enhancement parameters of mean, median, and third quartile enhancement were all significantly higher in ccrcc compared with prcc (P,.001), and the histogram parameter third quartile enhancement had the highest accuracy in discriminating ccrcc from prcc, which was higher than qualitative assessment by the radiologists. Interestingly, histogram distribution parameters kurtosis and skewness were significantly lower in ccrcc. Although such histogram distribution parameters have been shown to be useful in evaluation of white matter disease in brain and in assessment of treatment response in breast cancer (29,30), to our knowledge such histogram distribution parameters have not been evaluated in discriminating ccrcc from prcc. Histogram distribution parameters are independent of the absolute measure of signal intensity and thus help overcome the important limitation of using signal intensity cutoff values in making a diagnosis. Signal intensity cutoff values are difficult to apply in clinical practice, as there is variability within and between MR systems in signal intensity measures on T1-weighted images. Furthermore, change in signal intensity after contrast agent injection depends on the type, dose, and rate of gadolinium-based contrast agent injected. Thus, use of histogram distribution parameters may be a more appropriate comparator for across different MR systems as long as imaging is performed in the corticomedullary phase of enhancement. The accuracy of WL enhancement parameters was not significantly different from that of the ROI-based method in the ability to discriminate ccrcc from prcc. However, there was higher interobserver agreement for WL enhancement parameters compared with ROI-based analysis in discriminating ccrcc from prcc. One of the limitations of such ROI-based methods as previously mentioned is inter- and intraobserver variability in ROI placement. This method requires one to qualitatively assess the lesion and then find a small region that demonstrates most avid enhancement for ROI placement. Other technical problems such as misregistration between pre- and postcontrast acquisitions and difficulty with appropriate ROI placement on a precontrast image for T1-isointense tumors can also lower interobserver agreement. However, the advantage of the ROI-based method is that it is easy to perform and requires no special postprocessing or workstation. Historically, performing WL evaluation, although more informative, has been technically challenging and hence not been adopted in clinical practice. The tool such as the one we tested in this study can make the process of performing WL evaluation simpler by performing automated registration and segmentation to generate histogram of WL enhancement on a per-voxel basis. The retrospective nature of this study was a limitation. Furthermore, we evaluated the two most common subtypes of RCC and did not evaluate other subtypes, which have lower incidence 796 radiology.rsna.org n Radiology: Volume 265: Number 3 December 2012

8 Figure 4 Figure 4: Findings in a 58-year-old woman with an enhancing lesion in the lower pole of the right kidney (arrow), which on review of (a) precontrast and (b) postcontrast T1-weighted fat-saturated gradient-echo VIBE MR images was diagnosed as prcc by all three radiologists, owing to low-level homogeneous contrast enhancement at corticomedullary phase. (c) WL enhancement histogram analysis demonstrated third quartile enhancement of 226 (.100), kurtosis of 1.01 ( 2.016), and skewness of ( 1.296), which are all consistent with a diagnosis of ccrcc. At pathologic evaluation, the lesion was proved to be ccrcc. of occurrence. Histogram enhancement and distribution parameters in these other less-common subtypes remain to be evaluated. Accuracy of enhancement parameters in discriminating ccrcc may be overestimated in our study since cutoff values were generated and evaluated on the same data set. In conclusion, our study has shown that histogram analysis of WL enhancement as well as histogram distribution parameters kurtosis and skewness can be used to discriminate ccrcc from prcc with high accuracy and high interreader agreement. Disclosures of Conflicts of Interest: H.C. No relevant conflicts of interest to disclose. A.B.R. Financial activities related to the present article: Figure 5 Figure 5: Flow diagram shows the patient population, index test results (WL third quartile enhancement), and reference standard (histopathologic evaluation at surgery). = one patient had one ccrcc and one prcc lesion, for a total of 74 lesions in 73 patients. Radiology: Volume 265: Number 3 December 2012 n radiology.rsna.org 797

9 none to disclose. Financial activities not related to the present article: grants/grant pending to institution from Department of Defense. Other relationships: none to disclose. T.C.M. No relevant conflicts of interest to disclose. S.K. No relevant conflicts of interest to disclose. A.A.A. No relevant conflicts of interest to disclose. S.D.R. No relevant conflicts of interest to disclose. J. McMenamy No relevant conflicts of interest to disclose. J. Melamed. No relevant conflicts of interest to disclose. J.S.B. No relevant conflicts of interest to disclose. B.K. Financial activities related to the present article: employee of Siemens. Financial activities not related to the present article: none to disclose. Other relationships: none to disclose. A.B.R. Financial activities related to the present article: employee of Siemens. Financial activities not related to the present article: none to disclose. Other relationships: none to disclose. References 1. Sun M, Thuret R, Abdollah F, et al. Ageadjusted incidence, mortality, and survival rates of stage-specific renal cell carcinoma in North America: a trend analysis. Eur Urol 2011;59(1): Hock LM, Lynch J, Balaji KC. Increasing incidence of all stages of kidney cancer in the last 2 decades in the United States: an analysis of surveillance, epidemiology and end results program data. J Urol 2002;167(1): Chow WH, Devesa SS, Warren JL, Fraumeni JF Jr. Rising incidence of renal cell cancer in the United States. JAMA 1999;281(17): Bosniak MA. The small (less than or equal to 3.0 cm) renal parenchymal tumor: detection, diagnosis, and controversies. Radiology 1991;179(2): Lam JS, Shvarts O, Leppert JT, Figlin RA, Belldegrun AS. Renal cell carcinoma 2005: new frontiers in staging, prognostication and targeted molecular therapy. J Urol 2005;173(6): Reuter VE. The pathology of renal epithelial neoplasms. Semin Oncol 2006;33(5): Young AN, Dale J, Yin-Goen Q, et al. Current trends in molecular classification of adult renal tumors. Urology 2006;67(5): Cheville JC, Lohse CM, Zincke H, Weaver AL, Blute ML. Comparisons of outcome and prognostic features among histologic subtypes of renal cell carcinoma. Am J Surg Pathol 2003;27(5): Leibovich BC, Lohse CM, Crispen PL, et al. Histological subtype is an independent predictor of outcome for patients with renal cell carcinoma. J Urol 2010;183(4): Teloken PE, Thompson RH, Tickoo SK, et al. Prognostic impact of histological subtype on surgically treated localized renal cell carcinoma. J Urol 2009;182(5): Escudier B, Eisen T, Stadler WM, et al. Sorafenib in advanced clear-cell renal-cell carcinoma. N Engl J Med 2007;356(2): Powles T, Chowdhury S, Jones R, et al. Sunitinib and other targeted therapies for renal cell carcinoma. Br J Cancer 2011;104(5): Chowdhury S, Choueiri TK. Recent advances in the systemic treatment of metastatic papillary renal cancer. Expert Rev Anticancer Ther 2009;9(3): Volpe A, Kachura JR, Geddie WR, et al. Techniques, safety and accuracy of sampling of renal tumors by fine needle aspiration and core biopsy. J Urol 2007;178(2): Maturen KE, Nghiem HV, Caoili EM, Higgins EG, Wolf JS Jr, Wood DP Jr. Renal mass core biopsy: accuracy and impact on clinical management. AJR Am J Roentgenol 2007;188(2): González-Michaca L, Chew-Wong A, Soltero L, Gamba G, Correa-Rotter R. Percutaneous kidney biopsy, analysis of 26 years: complication rate and risk factors; comment [in Spanish]. Rev Invest Clin 2000;52(2): Pedrosa I, Chou MT, Ngo L, et al. MR classification of renal masses with pathologic correlation. Eur Radiol 2008;18(2): Pedrosa I, Sun MR, Spencer M, et al. MR imaging of renal masses: correlation with findings at surgery and pathologic analysis. RadioGraphics 2008;28(4): 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(3): Herts BR, Coll DM, Novick AC, et al. Enhancement characteristics of papillary renal neoplasms revealed on triphasic helical CT of the kidneys. AJR Am J Roentgenol 2002;178(2): Zhang J, Lefkowitz RA, Ishill NM, et al. Solid renal cortical tumors: differentiation with CT. Radiology 2007;244(2): Kim JK, Kim TK, Ahn HJ, Kim CS, Kim KR, Cho KS. Differentiation of subtypes of renal cell carcinoma on helical CT scans. AJR Am J Roentgenol 2002;178(6): Hecht EM, Israel GM, Krinsky GA, et al. Renal masses: quantitative analysis of enhancement with signal intensity measurements versus qualitative analysis of enhancement with image subtraction for diagnosing malignancy at MR imaging. Radiology 2004;232(2): Wernecke J. The inventor mentor: Programming object-oriented 3D graphics with open inventor. Boston, Mass?: Addison- Wesley, Chefd hotel C, Hermosillo G, Faugeras O. Flows of diffeomorphisms for multimodal image registration. Proceedings of IEEE International Symposium on Biomedical Imaging, 2002; Grady L. Random walks for image segmentation. IEEE Trans Pattern Anal Mach Intell 2006;28(11): Joanes DN, Gill CA. Comparing measures of sample skewness and kurtosis. J R Stat Soc Ser D (Statistician) 1998;47(1): Sheir KZ, El-Azab M, Mosbah A, El-Baz M, Shaaban AA. Differentiation of renal cell carcinoma subtypes by multislice computerized tomography. J Urol 2005;174(2): ; discussion Della Nave R, Foresti S, Pratesi A, et al. Whole-brain histogram and voxel-based analyses of diffusion tensor imaging in patients with leukoaraiosis: correlation with motor and cognitive impairment. AJNR Am J Neuroradiol 2007;28(7): Chang YC, Huang CS, Liu YJ, Chen JH, Lu YS, Tseng WY. Angiogenic response of locally advanced breast cancer to neoadjuvant chemotherapy evaluated with parametric histogram from dynamic contrast-enhanced MRI. Phys Med Biol 2004;49(16): radiology.rsna.org n Radiology: Volume 265: Number 3 December 2012