Intravoxel incoherent motion MR imaging in nasopharyngeal carcinoma: comparison and correlation with dynamic contrast enhanced MR imaging

Objectives To compare accuracy and assess agreement between intravoxel incoherent motion (IVIM) magnetic resonance (MR) perfusion-related parameters and quantitative dynamic contrast-enhanced (DCE) MR parameters in nasopharyngeal carcinoma (NPC). Results D, f, D*, Ktrans, Kep and Vp were significantly lower in the high stage group while Ve was significantly higher in the high stage group. Optimal cut-off values were: D=0.749 × 10−3 mm2/s; f=0.145; D*=100.401 × 10−3 mm2/s; Ktrans=0.571/min; Kep=0.8196/min; Ve=0.6556 %; Vp=0.0757 %. D* (p=0.001), Ktrans (p<0.001), Ve (p=0.014) were all reliable independent predictors for AJCC staging. IVIM-MR perfusion-related (f, D*) and DCE-MR (Ktrans, Kep, Ve, Vp) parameters were significantly correlated (p<0.001). Materials and Methods 75 patients with newly diagnosed NPC were prospectively recruited. Diffusion-weighted MR and DCE-MR imaging were performed with respective IVIM (D, f, D*) and DCE (Ktrans, Kep, Ve, Vp) MR parameters calculated. Patients were stratified into low and high tumor stage groups according to American Joint Committee on Cancer (AJCC) staging for determination of the predictive powers of IVIM-MR and DCE-MR parameters using t–test, ROC curve analyses and multiple logistic regression analysis. Correlation between IVIM-MR perfusion-related and DCE-MR parameters was assessed using Spearman's rank correlation. Conclusion IVIM-MR perfusion-related and quantitative DCE-MR parameters were significantly correlated in the assessment of NPC and were both reliable independent predictors in the prediction of AJCC staging. IVIM-MR perfusion imaging can be a potential useful non-invasive perfusion imaging tool for clinical use in the assessment of NPC.


INTRODUCTION
Noninvasive diagnostic imaging and staging in nasopharyngeal carcinoma (NPC) is crucial in the early diagnosis and treatment planning, especially for patients with high stage disease requiring adjuvant therapy. Magnetic resonance (MR) imaging has been the gold standard and imaging of choice in the detection and delineation of local tumor extent [1][2][3][4]. Diffusion-weighted (DW) MR imaging allows additional information on the functional assessment and characterization of the tumor, showing promising results with successful differentiation between different tumor grading, rendering it an imaging biomarker.
Intravoxel incoherent motion (IVIM) MR imaging is one type of DW MR imaging technique receiving much attraction in that it provides with us both the diffusion and perfusion related information without the use of exogenous contrast. We have previously demonstrated that it is feasible in the differentiation between NPC and post-chemoradiation fibrosis [5], as well as in the pre-treatment staging prediction [6], showing high diagnostic accuracy. While the precision of the derived pure diffusion parameter, D (reflecting tissue cellularity) has been proven to be consistent with high diagnostic confidence, concern has been raised regarding the accuracy of the derived perfusion related parameters (perfusion fraction, f reflecting normal angiogenesis with intact vascular permeability; and pseudodiffusion coefficient, D* reflecting tumoral vascularity) given the limited signalto-noise ratio (SNR). The clinical utility of dynamic contrast enhanced (DCE) MR perfusion had been established in various tumors [7][8][9][10] as well as in NPC [11][12][13], with its quantitative analysis being found useful in the assessment of tumoral vascularity, microcirculation property and hypoxia status since the contrast kinetics is associated with angiogenesis within tumors. In particular, K trans (diffusive transport of Gd-DOTA across capillary endothelium/ volume transfer) and V e (extracellular extravascular volume fraction) are frequently used to assess tumor outcomes as well as patient's prognosis [14]. The relationship between IVIM-MR perfusion-related parameters and semi-quantitative DCE-MR parameters had initially been tested in NPC, showing promising result [15]. However, the relationship between IVIM-MR perfusion-related parameters and quantitative DCE-MR parameters remains unexplored.
The aim of this study was to assess the correlation and agreement of the IVIM-MR perfusion-related parameters with the quantitative DCE-MR parameters, and to compare their diagnostic performance.

DCE-MR parameters
K trans , K ep and V p were all statistically significantly lower in the high stage group as compared with low stage group in AJCC staging. V e on the contrary, was found to be positively correlated, showing statistically significantly higher value in the high stage group as compared with low stage group. Their respective mean values +/− SD were summarized in Table 1. Respective optimal cut-off values upon ROC curve analyses (with respective sensitivity, specificity, positive likelihood ratio and negative likelihood ratio) were summarized in Table 2 Figure 2. K trans was the most powerful parameter based on AUC (0.996).

Comparison between IVIM-MR and DCE-MR parameters
Stepwise regression analysis revealed D (p<0.001), f (p=0.018), D* (p=0.001), K trans (p<0.001) and V e (p=0.014) were all reliable independent predictors. K trans was statistically significantly more powerful than f and D* (p=0.001) on direct comparison. No statistical significance could be reached when V e was compared against D, f and D*.

Correlation between IVIM and DCE-MR parameters
There was significant correlation between IVIM-MR perfusion-related parameters, f and D* with all DCE-MR parameters, K trans , K ep , V e and V p (Spearman's rank correlation coefficients: -0.628 to 0.533, p <0.001 (Bonferroni corrected p value = 0.002)). Their results were summarized and represented by scatter pots as shown in Figure 3.

DISCUSSION
Accurate staging in NPC is imperative for treatment planning as the differences in treatment paradigms and prognosis are based on tumor aggressiveness. This emphasizes the need for a noninvasive imaging tool that can help in the prediction of staging. IVIM MR imaging has been an attractive and promising alternative imaging tool that can simultaneously assess tissue vascularity and cellularity without the need of exogenous contrast. Despite the initial promising result of correlation between IVIM-MR perfusion-related and semi-quantitative DCE-MR parameters, the relationship between IVIM-MR perfusionrelated and quantitative DCE-MR parameters remains unclear.
Our results are in concordance with other published data on NPC [11][12][13][14][15], showing similar range, trend and distribution of the parameters across different tumor stages. The improvement in results as compared with our earlier DCE-MR study on NPC [12] is likely attributed to the larger cohort of patients and     and DCE-MR parameters (K trans and V e ) were found to be reliable independent predictors in the prediction of AJCC staging. On direct comparison of their diagnostic accuracy, K trans was more robust as compared with f and D* (p=0.001). The robustness of quantitative DCE-MR imaging (in particular for K trans ) is likely attributed to its direct reflection of the tissue physiology based on the concentration of contrast and are more closely linked to perfusion and permeability, hence related to tumor angiogenesis [14]. The exact relationship between IVIM-MR perfusion-related parameters and DCE-MR parameters remains unclear, but is likely to demonstrate close correlation as Le Bihan et al. [16] reasoned that tracer delivery to tissue is dependent on intravascular flow. Our results confirmed that IVIM-MR perfusion-related parameters (f and D*) were significantly correlated with the quantitative DCE-MR parameters (K trans , K ep , V e and V p ) (Spearman rank correlation p<0.001). This helped to suggest the usefulness of the IVIM-MR perfusion-related parameters in that they are potentially reflecting the microvessel density or contrast kinetics as observed with the quantitative DCE-MR parameters, rendering it a potential useful non-invasive perfusion imaging tool in the assessment of angiogenesis or neovasculature in NPC. High stage NPC exhibited significantly lower perfusion parameters (f, D*, K trans , K ep and V p ) as compared with low stage NPC, consistent with earlier IVIM-MR findings [5][6]. This can be attributed to the increasing tumoral heterogeneity due to the presence of microscopic necrosis or hypoxia in association with high stage NPC, as supported by prior studies showing reduced K trans with increasing hypoxia [17]. Sumi and Nakamura [18] also suggested that different stages of tumors would have distinctive stromal components with varying degrees of vascularization, and hence perfusion level, reflecting the relative area of stromal tissues with varying levels of vessel attenuation. Low stage NPC is less aggressive and has tightly packed stromal tissue due to the absence of necrosis, therefore increased vascularity overall. Our finding of higher V e in high stage NPC suggests tumor Although V e reflects the extracellular extravascular space but not the vascular compartment, it had been found to correlate with tumor aggressiveness and played a role in stimulating angiogenesis as well as strengthening tumor cells migration into peripheral tissue [11,14,17]. Chin et al. [14] also hypothesized that tumor microenvironment with a large V e would promote metastatic dissemination as it reflected the presence of abundant vascularized extracellular space.
The major limitation of our study is the lack of histopathological correlation such that the physiological basis of the perfusion status between different stages of NPC cannot be established. In particular, K trans does not necessarily linearly correlated to perfusion as it is also determined by the permeability of the blood vessels. Measurement reliability of both the IVIM-MR and DCE-MR perfusion parameters is another major challenge, given the intrinsic wide variability and SNR variation as reflected by the large SD and heterogeneous parametric maps accompanying f, D*, V e and V p [19,20]. The extended Tofts model used in the analysis of DCE-MR perfusion parameters is technically challenging, which can be confounded by many factors in both the acquisition and post-processing stages. In particular, the extremely low V p value raises potential concern of overfitting and doubtfulness of its clinical use. In addition, it may not be able to sufficiently integrate the intricate microenvironment of biological tissues [13]. Estimations of IVIM-MR parameters are known to depend upon bvalue selection as different b values can result in higher or lower parameter estimations [21,22]. The 13 b-values utilized in our study were based on our initial experience in our earlier studies, achieving reasonable results. We believe the use of optimal b-values distribution and signal averaging using ROI techniques have helped in achieving satisfactory results. The major drawback was the long duration of scanning time (12 minutes) of the DW sequence rendering it undesirable for clinical use. Future study can be directed on different combinations of b-values as to search for an optimal number or range for clinical utilization within a reasonable scanning time. Although we were able to demonstrate the feasibility and correlation of IVIM-MR perfusion related parameters, larger sample study is necessary to assess and confirm the diagnostic performance. Larger cohort from different tumor stages of NPC wound also be helpful to look for its discriminating power in individual tumor stages, in particular within same T stage.
In conclusion, we have demonstrated significant correlation between IVIM-MR perfusion-related parameters (f and D*) and quantitative DCE-MR parameters (K trans , K ep , V e and V p ) in the assessment of NPC. D, f, D*, K trans and V e were all found to be reliable independent predictors in the prediction of AJCC staging though K trans was more robust as compared with f and D* (p=0.001). IVIM-MR perfusion imaging can be a potential useful non-invasive perfusion imaging tool for clinical use in the assessment of NPC.

Patient selection
A prospective study was carried out after institutional review board approval with written informed consent was obtained. Consecutive patients with newly diagnosed NPC seen in our institution from January 2012 to March 2014 were recruited. Major inclusion criterion was patients with newly diagnosed biopsied proven NPC. Major exclusion criteria were: a) treatment of any form (surgery, chemoradiation) already given; b) concurrent nasopharyngeal disease or other tumors.
Eighty patients with newly diagnosed biopsied proven undifferentiated NPC were initially recruited. No other histological subtypes of NPC were identified. Five patients were excluded due to small tumor volumes (<30.00 mm 3 ) or significant motion artifacts leading to inaccurate assessment. Therefore, a total of seventy-five patients (51 male, 24 female; mean age 53.0 +/− 13.4 years, range 25-90 years) were included. American Joint Committee on Cancer (AJCC) and TNM stages for all patients were determined by two independent clinical oncologists subspecialized in NPC, based on all available imaging findings and endoscopic findings. Discrepancy if any, was resolved after subsequent consensus. Distribution of all patients' respective AJCC and TNM stages together with their clinical demographic data were listed in Table 3.

Image analysis
Image analysis of the DW MRI data was performed using an in-house software program developed in IDL 6.3 (ITT Visual Information Solutions, Boulder, CO) and fitted on a pixel-by-pixel basis by Levenberg-Marquardt algorithm, as described before [23][24][25][26]. DW images from all 3 directions were automatically averaged and used for analysis. The bi-exponential model from an IVIM sequence was described by Le Bihan [27] as S b /S 0 = (1-f) exp(-bD)+fexp[-b(D*+D)], where S b is the signal intensity in the pixel with diffusion gradient b, S 0 is the signal intensity in the pixel without diffusion gradient, D is the true diffusion as reflected by pure molecular diffusion, f is the fractional perfusion related to microcirculation, and D* is the pseudodiffusion coefficient representing perfusion-related diffusion or incoherent microcirculation. D was obtained by a simplified linear fit equation (S b = S 0 x exp-b.D ) using b values > 200 s/mm 2 ; while f and D* were calculated by a nonlinear regression algorithm using all b values. Using Matlab (The MathWorks, Natick, Massachusetts), only pixels with signals safely above background noises were included for calculation to avoid fitting pixels with low SNR. Therefore, pixels with low SNR were removed from subsequent analysis.
Image analysis of the DCE MRI data was performed by in-house software program (DCE-Tool, Version 5.1, Philips Healthcare, Best, The Netherlands) based on a twocompartment model (Extend Tofts model) [28,29]. Motion correction and co-registration were initially performed and both the precontrast T1 map and dynamic T1 map were calculated as described previously [30,31]. Based on the assumption of a linear relation between Gd-DOTA concentration and 1/T1, the Gd-DOTA concentration map was calculated. Arterial input function (AIF) was selected and obtained from the vertebral artery, with the parametric maps of K trans (diffusive transport of Gd-DOTA across capillary endothelium/ volume transfer), K ep (rate constant of Gd-DOTA/ reflux rate constant), V e (extracellular extravascular volume fraction) and V p (intravascular plasma volume fraction) generated automatically. Again, Using Matlab (The MathWorks, Natick, Massachusetts), only pixels with signals safely above background noises were included for calculation to avoid fitting pixels with low SNR. Any unphysiological data such as zero were excluded. Kinetic pharmacokinetic model fitting curves were used for validation of the data.
Region-of-interest (ROI) was manually drawn by two independent head-and-neck radiologists to contour the border of NPC on each slice of the axial STIR T2weighted images ( Figure 4A), avoiding the inclusion of air and adjacent anatomic structures, such that the total tumor volume was analyzed. Total tumor volume was calculated using segmentation method as described previously [32]. It was then co-registered to the diffusion ( Figure 4B) and perfusion ( Figure 4C-4H) maps for subsequent analysis using Image J (NIH, Bethesda, MD). Each lesion was measured twice in two separate sessions at 2 weeks apart to ensure reproducibility. The obtained IVIM-MR and DCE-MR parameters for each tumor were calculated on a pixel-by-pixel basis and expressed as mean values of all pixels within the volume of interest.

Statistical analysis
Patients were stratified into two groups for the evaluation of AJCC stage by combining AJCC stage I and II (n=29) into low stage group, and AJCC stage III and IV (n=46) into high stage group for statistical analysis.
Normality test was performed using Shapiro-Wilk test, with the DCE and IVIM parameters showing approximately normal distribution (all P>0.05). Univariate analysis using Student's t-test was performed to compare IVIM-MR parameters (D, f, D*) and DCE-MR parameters (K trans , K ep , V e and V p ) between low and high AJCC stages. Multiple logistic regression analysis was performed to gauge their independent predictive values. Receiver operating characteristic (ROC) curves were derived with respective cut-off values determined to accommodate best diagnostic accuracy. Comparison of the accuracy of different ROC curves was carried out by given p values. Correlation between IVIM-MR perfusion-related parameters and DCE-MR parameters were determined by using Spearman's rank correlation. All statistical analyses were performed by SPSS version 21 (SPSS Inc, Chicago, IL, USA). Bonferroni correction was performed to account for the multiple comparisons.

Abbreviations
AIF -arterial input function AJCC -American Joint Committee on Cancer AUC -area under curve D -pure diffusion D* -pseudodiffusion coefficient DCE -dynamic contrast enhanced DW -diffusion weighted f -perfusion fraction FFE -fast field echo FOV -field of view K trans -volume transfer constant of Gd-DOTA K ep -rate constant of Gd-DOTA IVIM -intravoxel incoherent motion MR -magnetic resonance NPC -nasopharyngeal carcinoma ROC -receiver operating characteristic ROI -region of interest SD -standard deviation SNR -signal-to-noise ratio SPIR -spectral presaturation inversion recovery STIR -short T1 inversion recovery TFE -turbo-field-echo TR/TE -repetition time/echo time TSE -turbo-spin-echo V e -extracellular volume fraction V p -blood volume fraction