Dynamic Contrast-enhanced MRI in Renal Tumors: Common Subtype Differentiation using Pharmacokinetics

Preoperative renal tumor subtype differentiation is important for radiology and urology in clinical practice. Pharmacokinetic data (K trans & V e, etc.) derived from dynamic contrast-enhanced MRI (DCE-MRI) have been used to investigate tumor vessel permeability. In this prospective study on DCE-MRI pharmacokinetic studies, we enrolled patients with five common renal tumor subtypes: clear cell renal cell carcinoma (ccRCC; n = 65), papillary renal cell carcinoma (pRCC; n = 12), chromophobic renal cell carcinoma (cRCC; n = 9), uroepithelial carcinoma (UEC; n = 14), and fat-poor angiomyolipoma (fpAML; n = 10). The results show that K trans of ccRCC, pRCC, cRCC, UEC and fpAML (0.459 ± 0.190 min−1, 0.206 ± 0.127 min−1, 0.311 ± 0.111 min−1, 0.235 ± 0.116 min−1, 0.511 ± 0.159 min−1, respectively) were different, but V e was not. K trans could distinguish ccRCC from non-ccRCC (pRCC & cRCC) with a sensitivity of 76.9% and a specificity of 71.4%, respectively, as well as to differentiate fpAML from non-ccRCC with a sensitivity of 100% and a specificity of 76.2%, respectively. Our findings suggest that DCE-MRI pharmacokinetics are promising for differential diagnosis of renal tumors, especially for RCC subtype characterization and differentiation between fpAML and non-ccRCC, which may facilitate the treatment of renal tumors.

plasma to the EES (K trans ), the efflux rate constant from EES back to plasma (K ep ), the ratio of the EES volume to tissue volume (V e ), and the ratio of blood plasma volume to tissue volume (V p ).
In previous studies, DCE-MRI pharmacokinetics were chiefly used for the central nervous system and fixed organs. Specifically, K trans was used to evaluate histologic grades of intracranial gliomas 19 , and time-signal intensity curves of breast tumors combined with K trans were used to improve the diagnostic accuracy of breast carcinoma 20 . For renal tumors, DCE-MRI pharmacokinetics have been focused on the qualitative diagnosis and evaluation of targeted molecular therapy of metastatic or advanced RCC. However, no comprehensive quantitative analysis has been published. In a previous study, we confirmed that DCE-MRI pharmacokinetic data (K trans & V e ) were reproducible in RCC 21 . Thus, in this study, we used DCE-MRI to perform pharmacokinetic assessments of common renal tumors, and investigated the value of using these pharmacokinetic data for the differentiation of renal tumor subtypes.

Methods
Patients. The Institutional Review Board of Chinese PLA General Hospital (Beijing, China) approved this study (#S2012-049-01). Study methods were performed in accordance with the Declaration of Helsinki, and written informed consent was obtained from each subject prior to study initiation. Patients with a renal tumor diagnosis were consecutively enrolled from September 2012 to December 2013 and underwent DCE-MRI scans using a 3.0 Tesla MR system (GE Discovery MR 750, GE Healthcare, Milwaukee, WI, USA). Inclusion criteria were as follows: >18 years of age; glomerular filtration rate (GFR) ≥60 mL/min, maximal renal tumor diameter ≥1 cm; pathologic tumor types including clear cell RCC (ccRCC), papillary RCC (pRCC), chromophobic RCC (cRCC), renal pelvic carcinoma (RPC), oncocytoma, and fat-poor renal angiomyolipoma (fpAML) (no fat attenuation on CT or fat signal intensity on MRI). Exclusion criteria included lesions with complete cystic degeneration or necrosis, poor imaging quality (cannot meet imaging analysis requirements); common contraindications for enhanced MRI such as allergy to gadolinium-related contrast agent, metal implants, or claustrophobia.
MRI acquisition. All of the patients underwent an MRI scan within 48 h of the initial diagnosis. MRI examinations were performed on a 3.0 T scanner with a maximum gradient strength of 50 mT and maximum slew rate of 200 mT/s, using an 8-channel surface phased-array coil. Patients practiced breathing techniques before each scan, which included breathing quickly during a non-scanning break and then breath-holding in the same position for as long as possible. Routine clinical axial and coronal T2-weighted imaging was performed for all patients prior to dynamic studies to localize and delineate tumors. The imaging protocol for DCE-MRI consisted of a pre-contrast T1 mapping sequence and a DCE sequence. The former included five consecutive axial 3D spoiled-gradient recalled-echo sequences for liver acquisition with volume acceleration (LAVA) with an array of flip angles (3°, 6°, 9°, 12°, and 15°) in breath-hold mode. It also included an axial DCE sequence (flip angle, 12°): scanning during 12 s of breath-holding for two phases and a subsequent 6 s of breathing was performed repeatedly for up to 4.4 min to monitor contrast passage. Scanning parameters were as follows: repetition time (TR), 2.8 ms; echo time (TE), 1.3 ms; matrix, 288 × 180; field of view (FOV), 38 × 38 cm; slice thickness, 6 mm; number of excitations (NEX), 1; bandwidth, 125 kHz; and parallel imaging acceleration factor, 3. The contrast agent, gadodiamide (0.1 mmol/kg, Omniscan, GE Healthcare) was given intravenously when the second scan was started at a rate of 2 mL/s using a power injector (Spectris; MedRad, Warrendale, PA, USA). The contrast bolus was flushed with 20 mL normal saline, administered at the same rate, to improve bolus coherence.
Image post-processing and analysis. All of the DCE-MRI analyses were conducted using open-source software packages, including the R package (http://dcemri.sourceforge.net/) and a medical image non-rigid registration package (http://cmictig.cs.ucl.ac.uk/wiki/index.php/NiftyReg).
All images were transferred to an Omni-Kinetics workstation (GE Healthcare, LifeScience, China) for analysis. The breath-hold position for each patient differed and the shape of the kidney non-rigidly varied between individuals. It has been shown that image registration methods 22 can be used to handle body motion within the time domain 23,24 . Here, the workstation provided an automatic nonlinear registration framework 25 to remove errors of misalignment between consecutive MRI scans, thereby increasing accuracy. The registration framework used a free-form deformation algorithm 26 as the main registration engine and mutual information as the correspondence metric 27 .

Data Collection. Calculation of pharmacokinetic parameters.
A multiple flip angle method 17, 28 was used to perform T1 mapping to obtain the T1 value of the tissue before and after contrast agent injection. Then the contrast agent concentration in the tissue was computed using tissue signal intensity. A two-compartment extended-Tofts model was used 29 (Eq. 1) with a population-based arterial input function (AIF) 17, 28 (Eq. 2) to calculate parameters. In Equation 1, K trans is the transfer constant from plasma to the EES, V e is the ratio of EES volume to tissue volume, V p is the ratio of blood plasma volume to tissue volume, K ep = K trans /V e is the efflux rate constant from EES to plasma, and C t (t) and Cp(t) represents the contrast agent concentrations in the tissue and plasma, respectively. In Equation 2, D = 1.0 mmol/kg, a 1 = 2.4 kg/l, a 2 = 0.62 kg/l, m 1 = 3.0, and m 2 = 0.016. Region of interest selection. All of the images were transferred to a Sun workstation (Sparc 10, Sun Microsystems, Mountain View, CA, USA), at which pharmacokinetics were measured using ImageJ software (National Institutes of Health, Bethesda, MD, USA). Using reference information from anatomic axial and coronal T2-weighted images and post-contrast T1 images, one radiologist blinded to the pathologic results was instructed to place region of interests (ROIs) on the slice with the largest diameter of tumors according to dynamic images of DCE-MRI, covering the whole tumor where possible but excluding pulsatile artifacts from blood vessels and susceptibility artifacts from adjacent bowels. Then, the same ROI was copied to parametric maps (K trans , V e ).

Pathologic analysis.
All of the specimens after partial nephrectomy or radical nephrectomy were examined by two uropathologists blinded to MRI findings, and the consensus was used for final decisions. All of the lesions were pathologically characterized according to World Health Organization tumor classification of the kidney 30 .
Statistical Analyses. Test of Normality. Normality of all data was analyzed using Shapiro-Wilk method.
Normality was confirmed when p > 0.05.
Differences in the pharmacokinetics of renal tumor subtypes. All of the pharmacokinetics are expressed as mean ± standard deviation (SD) or median with ranges. Differences in the pharmacokinetics of different renal tumor subtypes were evaluated using an independent samples Kruskal-Wallis test or a one-way analysis of variance (ANOVA).
Difference in pharmacokinetics between benign and malignant tumors. CcRCC, pRCC, cRCC, and UEC were classified as malignant tumors and fpAML and oncoctyoma were defined as benign. Pharmacokinetic differences between benign and malignant tumors were evaluated using an independent samples t-test or a Mann-Whitney U test.
Difference in pharmacokinetics of RCC subtypes. Pharmacokinetic differences among RCC subtypes were evaluated using an independent samples Kruskal-Wallis test or one-way ANOVA and differences between ccRCC and non-ccRCC (pRCC and cRCC) were analyzed using an independent samples t-test or Mann-Whitney U test. A receiver operating characteristic (ROC) curve was used to analyze the diagnostic sensitivity and specificity, and to calculate Youden's index.

Differences in pharmacokinetics between fpAML and non-ccRCCs and between RCCs and UECs.
Pharmacokinetic differences between fpAML and non-ccRCCs and between RCCs and UECs were evaluated using an independent samples t-test or Mann-Whitney U test, and an ROC was used to analyze the diagnostic sensitivity and specificity, and to calculate Youden's index. All of the statistical analyses were performed with SPSS software (IBM SPSS Statistics for Macintosh, Version 22.0. IBM Corp., Armonk, NY, USA) and p values less than 0.05 were considered statistically significant. However, for multiple samples compared with ANOVA or the Kruskal-Wallis test, p values less than 0.01 were considered statistically significant.

Patient information and lesion characterization. Patient information, surgical and pathologic data
were collected by a senior attending radiologist, and data for the subjects appear in Table 1. Of the enrolled subjects, 82 patients underwent partial nephrectomy and 37 underwent radical nephrectomy. The interval between DCE-MRI scanning and surgery was 7.2 ± 3.8 days. After excluding renal adenoma (n = 2), renal metastasis (n = 1), solitary fibroma (n = 1), juxtaglomerular cell tumor (n = 1), and cases with poor imaging quality (n = 4), a total of 110 patients underwent DCE-MRI pharmacokinetic analysis (Fig. 1). We did not enroll patients with oncocytomas. Comparison of DCE-MRI pharmacokinetics among renal tumor subtypes. The K trans and V e parametric maps of five renal tumor subtypes are shown in Fig. 2. Differences in K trans among five renal tumors were statistically significantly different (p < 0.001) and pairwise comparisons appear in Table 2 and Fig. 3. Differences in V e among the five renal tumors were not statistically significantly different (p = 0.044; Fig. 4).

Comparison of DCE-MRI pharmacokinetics between fpAML and non-clear cell RCCs. K trans val-
ues for fpAML and non-ccRCCs were statistically significantly different (p < 0.001). Threshold K trans values to distinguish fpAML from non-ccRCCs as well as sensitivity and specificity (Youden's index 0.762) and AUC data appear in Fig. 6. V e values for fpAML and non-ccRCCs were not statistically significantly different (p = 0.069).

Comparisons of DCE-MRI pharmacokinetics between RCC and UEC. K trans of RCCs and UECs were
statistically significantly different (p = 0.015). Threshold K trans values to distinguish RCC from UEC appear in Fig. 7 along with sensitivity and specificity data (Youden's index 0.762). AUC data appear in Fig. 7 as well. V e for RCCs and UECs were not statistically significantly different (p = 0.396).

Discussion
The accurate diagnosis of renal masses can be accomplished by analyzing the imaging features of renal masses. Although diagnostic imaging is often used to diagnose renal masses, it comes with a number of challenges. Thus, in this study, we used DCE-MRI pharmacokinetics to characterize renal masses among five renal tumor subtypes to determine if kinetic measurements could be used as an alternative diagnostic tool for the differential diagnosis of renal tumors. Of these, fpAML had the greatest K trans followed by ccRCC, cRCC, UEC, and pRCC. The fpAML and ccRCC values were not different statistically but the K trans of ccRCC was greater than that of pRCC, a finding that is in accordance with the literature 31 . fpAML had the greatest K trans , likely due to its thick-walled blood vessels that lack arterial elasticity 30,32 . ccRCC tumors have a rich and regular network of small thin-walled blood vessels, which may create high K trans . pRCC tumors have few blood vessels, which may contribute to the low K trans .
Using K trans and V e to distinguish between renal benign and malignant tumors produced no statistically significant differences, which may be explained by the fact that ccRCCs accounted for most of the malignant tumors and their pharmacokinetics were similar to those of fpAMLs. For ccRCC and non-ccRCC, K trans was statistically significantly different and K trans had a large area under the ROC curve for diagnosing ccRCC (0.819); however, the V e values were not significantly different.
Differentiating fpAML from non-ccRCC is of interest, but previous studies have shown that CT is of little value in this regard 33 and that fpAML and pRCC often overlap in images 34,35 ; specifically, both renal masses can appear hypointense on T2-weighted images. Although MRI has been used to analyze imaging differences between fpAML and RCCs 10,36,37 , the analyses were done by grouping ccRCC and non-ccRCC together instead of analyzing them separately, the latter of which is the ideal way to analyze these two different types of tumors 10,38 . ccRCCs have many distinguishing features compared to fpAML, so positive results would be expected. Here, we focused on the differentiation between fpAML and non-ccRCCs, and noted that K trans was statistically significantly different between these tumor subtypes, with an area under the ROC curve of 0.924. When the threshold for K trans of 0.365 min −1 was selected, the sensitivity and specificity of fpAML were high. Increasing the threshold K trans value to 0.427 min −1 , improved specificity and worsened sensitivity, which may allow preoperative distinctions between fpAML and non-ccRCC. Uroepithelial carcinoma of the renal pelvis or renal pelvic carcinoma that invades the renal parenchyma may mimic RCC in the center of the kidney. Wehrli's group 11 pointed out that T2 weighted image signal intensity and uncorrected apparent diffusion coefficient values were not different between RPC and RCC. We observed that   RCCs had larger K trans than RPCs, likely because RCCs have a higher microvascular density than RPCs. With a threshold of 0.228 min −1 , K trans can distinguish RCCs from RPC (AUC 0.766; sensitivity 86%; specificity 71.4%), which may be useful for distinguishing between these tumors. However, V e was not different between RCC and RPC, so this value cannot be used as a distinguishing index. For the DCE-MRI technique, we chose a population averaged arterial input function (AIF) instead of a personal AIF to perform pharmacokinetic calculations. Personal or individual AIFs, if calculated accurately, can improve pharmacokinetic studies, but personal AIFs require a high temporal resolution and may be influenced by physiology, ROI placement, partial volume effects, and inflow effects. Due to the non-continuous scanning mode of the DCE-MRI (See "MRI technique" in Methods) for balancing clinical practice and scientific research needs, the temporal resolution of DCE-MRI was limited. Thus, we used a population-based AIF method, which addressed temporal resolution difficulties and reduced AIF ROI location and sizing errors as previously reported 39 . In addition, population-based AIF works as well as individual AIF for estimating pharmacokinetics, as confirmed by several investigators [40][41][42] .
The limitations of this study include the necessity of image registration and establishment of kinetic parametric maps, which was time-consuming and is not ideal for clinical practice. Thus, more user-friendly software or an accelerating method should be investigated. Second, ROIs covering the entire tumor on the slice with its maximum diameter was the most reproducible method for drawing ROI in DCE-MRI analysis, but this method ignores necrosis, cystic changes, and hemorrhages, which may induce errors in analysis. In the future, histogram  . ROC curve (blue line) comparison of K trans in fpAML and non-ccRCCs. The AUC was 0.924. When the threshold K trans value was 0.365 min −1 , the sensitivity and specificity were 100% and 76.2%, respectively. When the K trans value was greater than 0.427 min −1 , the sensitivity and specificity were 70.0% and 95.2%, respectively.
analysis of pharmacokinetics should be attempted. Third, we were unable to enroll patients with renal oncocytoma, as this is a relatively rare disease, making it difficult to obtain an appreciable sample size. Previous work indicates that oncocytoma has a similar K trans and V e as ccRCC 31 , but our sample size was small (n = 3). Thus, additional research is required to validate our findings. Finally, a few patients in our center chose CT instead of MRI for imaging, and of those who agreed to undergo MRI examination, many could not endure the lengthy DCE-MRI process.
In conclusion, DCE-MRI kinetic measurements are promising for the differential diagnosis of renal tumors, especially for RCC subtype characterization, and for distinguishing between fpAML and non-ccRCC.