Mri_2017.pdf

Magnetic Resonance Imaging 44 (2017) 32–37

Contents lists available at ScienceDirect

Magnetic Resonance Imaging journal homepage: www.mrijournal.com

Original contribution

Diagnostic accuracy of automatic normalization of CBV in glioma grading using T1- weighted DCE-MRI Prativa Sahoo a, Rakesh K. Gupta b,⁎, Pradeep K. Gupta b, Ashish Awasthi c, Chandra M. Pandey c, Mudit Gupta b, Rana Patir d, Sandeep Vaishya d, Sunita Ahlawat e, Indrajit Saha f a

Division of Mathematical oncology, City of Hope National Medical Center, CA, USA Department of Radiology and Imaging, Fortis Memorial Research Institute, Gurgaon, India Department of Biostatistics and Health Informatics, SGPGIMS, Lucknow, India d Department of Neurosurgery, Fortis Memorial Research Institute, Gurgaon, India e SRL Diagnostics, Fortis Memorial Research Institute, Gurgaon, India f Philips Health System, Philips India Limited, Gurgaon, India b c

a r t i c l e

i n f o

Article history: Received 31 March 2017 Accepted 2 August 2017 Available online xxxx Keywords: DCE-MRI Glioma grading CBV normalization

a b s t r a c t Purpose: Aim of this retrospective study was to compare diagnostic accuracy of proposed automatic normalization method to quantify the relative cerebral blood volume (rCBV) with existing contra-lateral region of interest (ROI) based CBV normalization method for glioma grading using T1-weighted dynamic contrast enhanced MRI (DCE-MRI). Material and methods: Sixty patients with histologically confirmed gliomas were included in this study retrospectively. CBV maps were generated using T1-weighted DCE-MRI and are normalized by contralateral ROI based method (rCBV_contra), unaffected white matter (rCBV_WM) and unaffected gray matter (rCBV_GM), the latter two of these were generated automatically. An expert radiologist with N10 years of experience in DCE-MRI and a non-expert with one year experience were used independently to measure rCBVs. Cutoff values for glioma grading were decided from ROC analysis. Agreement of histology with rCBV_WM, rCBV_GM and rCBV_contra respectively was studied using Kappa statistics and intra-class correlation coefficient (ICC). Result: The diagnostic accuracy of glioma grading using the measured rCBV_contra by expert radiologist was found to be high (sensitivity = 1.00, specificity = 0.96, p b 0.001) compared to the non-expert user (sensitivity = 0.65, specificity = 0.78, p b 0.001). On the other hand, both the expert and non-expert user showed similar diagnostic accuracy for automatic rCBV_WM (sensitivity = 0.89, specificity = 0.87, p = 0.001) and rCBV_GM (sensitivity = 0.81, specificity = 0.78, p = 0.001) measures. Further, it was also observed that, contralateral based method by expert user showed highest agreement with histological grading of tumor (kappa = 0.96, agreement 98.33%, p b 0.001), however; automatic normalization method showed same percentage of agreement for both expert and non-expert user. rCBV_WM showed an agreement of 88.33% (kappa = 0.76,p b 0.001) with histopathological grading. Conclusion: It was inferred from this study that, in the absence of expert user, automated normalization of CBV using the proposed method could provide better diagnostic accuracy compared to the manual contralateral based approach. © 2017 Elsevier Inc. All rights reserved.

1. Introduction Gliomas are generally classified into four grades by World Health Organization (WHO) on the basis of histopathologic characteristics. Grades I and II are colloquially referred to as low grade gliomas (LGG) and grades III and IV as high grade gliomas (HGG). Pre operative grading is helpful in prognosticating and selecting the sample for the highest ⁎ Corresponding author at: Department of Radiology and Imaging, Fortis Memorial Research Institute, Gurgaon, Haryana 122002, India. E-mail address: [email protected] (R.K. Gupta).

http://dx.doi.org/10.1016/j.mri.2017.08.003 0730-725X/© 2017 Elsevier Inc. All rights reserved.

grade of the disease during surgery. MR derived phenotyping is performed by T1 weighted imaging (T1w), T2 weighted imaging (T2w), fluid attenuated inversion recovery (FLAIR), diffusion weighted imaging (DWI), and post-contrast T1w imaging. However, the imaging features for such protocols are often non-specific and provide limited understanding of the tumor biology [1,2]. To improve the accuracy of MR-based glioma grading, advanced imaging techniques such as dynamic contrast enhanced (DCE) and dynamic susceptibility contrast (DSC) are being explored to study the tissue perfusion properties in routine clinical practices [3–6]. Several studies have shown that the maximal relative cerebral blood volume

P. Sahoo et al. / Magnetic Resonance Imaging 44 (2017) 32–37

33

Table 1 Summary of MR imaging parameters. Sequence

TR (ms)

TE (ms)

TI (ms)

Flip angle

Thickness (mm)

Acquisition matrix

FOV (mm2)

Conventional protocol T2-w TSE T1-w TSE 3D-FLAIR DWI (b = 1000) SWI

3500 2077 4700 4156 31

110 20 288 114 7.2

– 800 1650 – –

90° 90° 90° 90° 17

4 5 0.56 5 2

384 × 384 400 × 342 204 × 204 152 × 121 288 × 288

230 × 230 230 × 184 230 × 184 230 × 184 230 × 230

DCE protocol Dual PD T2-w TSE (pre contrast) T1-w TSE (pre contrast) 3D-T1 FFE Post contrast 3D-T1-w TFE

3500 360 5.0 7.74

23.2, 90 10 1.4 3.54

– – –

90° 90° 10° 8°

6 6 6 1

128 × 128 128 × 128 128 × 128 240 × 222

240 × 240 240 × 240 240 × 240 288 × 288

w-weighted, TSE-turbo spin echo, FLAIR-fluid-attenuated inversion-recovery, DWI-diffusion weighted imaging, SWI-susceptibility weighted imaging, DCE-dynamic contrast enhanced, 3D-FFE-3 dimensional Fast field echo, TFE-turbo field echo, TR-repetition time, TE-echo time, TI-inversion time, FOV-field of view.

(rCBV) correlates with tumor neoangiogenesis and the grade of glioma [4–7]. On the other hand, literature also shows a wide range of inter research group variability in cutoff values of rCBV to discriminate HGG from LGG. One study has reported that DCE derived rCBV cutoff value of 2.5 can classify 90% glioma correctly [8] while other study reported optimal cutoff value for discriminating HGG from LGG may range from 3.75 to 5.58 [9]. Similar variations are also reported for DSC derived rCBV map. Median value of DSC derived rCBV for high grade may vary from 6.2 [6] to 4.37 ± 4.04 [10]. The use of different mathematical models and choice of technique of normalization might account for the above observations [9,11]. Conventionally, the normalization of MR-derived CBV map is based on the measurement of ratio between the maximum CBV area within

the glioma and the CBV of corresponding contralateral unaffected tissue. This value is often referred to as rCBV, and is reported to be significantly high for HGG than LGG [8]. It is quite evident that such a method is very much user dependent and therefore, more prone to user bias. Firstly, the voxel with maximum CBV might be non-representative of the sample because of image noise or computational error. Studies have shown that this error can be eliminated by histogram based techniques [6,9]. Secondly, inclusion of vasculature in the contralateral region of interest (ROI) could influence the results to the extent of changing the grade of glioma. Thirdly, contralateral ROI cannot be defined for midline tumors. Various studies have proposed normalization by a reference region like normal appearing white matter or cerebellum [12]; however, in case of white matter normalization what should be the correct reference location is still speculative since literature shows regional variation of rCBV value in the white matter regions [13]. Cerebellum may not be included in the imaging slab if the tumor is in frontal lobe due to technical limitations of maximal coverage especially in DCE perfusion study [8]. Incorrect selection of reference ROIs for relative CBVs might result in either under or overestimation of rCBV. To date histological analysis of tumor tissues still considered as the gold standard for glioma grading. However, tumor heterogeneity often results in sampling error and inter pathologist variability may also influence erroneous grading [14, 15]. Aim of this study was to develop an automated method for CBV normalization for T1-weighted DCE-MRI and compare its diagnostic accuracy with contralateral ROI based normalization in discriminating HGG from LGG.

2. Material and methods

Fig. 1. Flow chart shows the steps of proposed CBV normalization method.

The study is a single-center retrospective cross-sectional study. The data included in the analysis was obtained over 18 months (May-2014 to November-2015). It included a total of 60 consecutive treatment naïve patients with histologically confirmed glioma with gross total resection including 37 high grade glioma (29 glioblastoma multiforme, 8 anaplastic astrocytoma) and 23 low grade glioma (6 oligoastrocytoma, 5 oligodendroglioma, 12 astrocytoma). One patient was excluded from the study as the tumor was located in the thalamus and only stereotactic biopsy was done in that patient. In remaining 59 patients, gross total resection was done. All patients underwent conventional MRI (T1, T2, FLAIR, DWI and SWI) as well as DCE-MRI. Written informed consent from each patient was obtained before MRI study. This study was approved by the institutional ethics committee. All data were classified into three groups depending on the location of the tumor in the gray (gray matter (GM) ≥ 80%) or white matter (white matter (WM) ≥ 80%) or mixed when it was involving both gray and white matter. This localization was done by an experienced neuro-radiologist.

34

P. Sahoo et al. / Magnetic Resonance Imaging 44 (2017) 32–37

Table 2 Optimal relative CBV cutoff values between high grade glioma (HGG) and low grade glioma (LGG), sensitivity and specificity measured using Receiver Operating Characteristic (ROC) analysis.

rCBV_contra (non-expert) rCBV_contra (expert) rCBV_GMa rCBV_WMa

Cutoff

Sensitivity

Specificity

AUC

PPV

NPV

95% CI

p value

2.60 2.62 2.01 2.52

0.65 1.00 0.81 0.89

0.78 0.96 0.78 0.87

0.83 0.99 0.86 0.87

0.83 0.97 0.85 0.91

0.58 1.00 0.69 0.80

0.71–0.92 0.98–1.00 0.75–0.94 0.75–0.96

0.001 0.001 0.001 0.001

CBV-cerebral blood volume, rCBV_contra-CBV normalized with contralateral region, rCBV_GM- automatic CBV normalized with normal appearing gray matter, rCBV_WM- automatic CBV normalized with normal appearing white matter, AUC-area under curve, PPV-positive predicative value, NPV-negative predictive value, CI- confidence level. a Represents the statistical parameter values are same for both expert and non-expert user.

2.1. Imaging protocol All the imaging was performed on a single 3.0 T MRI scanner (Ingenia, Philips Health Systems, The Netherlands) with a 15 channel head coil. Conventional imaging of the brain included T2 weighted, T1 weighted inversion recovery prepared turbo spin echo (TSE) images, 3D fluid-attenuated inversion recovery (FLAIR), diffusion weighted imaging (DWI) and susceptibility weighted (SWI) imaging. For DCE-MRI, pre-contrast 2D T1-weighted TSE, fast dual spin echo proton density (PD)-weighted and T2-weighted fat suppressed images were acquired to quantify voxel-wise pre contrast tissue longitudinal relaxation time T10. Then, dynamic images were acquired using a T1 fast field echo (T1-FFE) with half scan option enabled with factor Y = 0.625. At the fourth time point of the dynamic data acquisition, 0.1 mmol/kg body weight of Gd-BOPTA (Multihance, Bracco, Italy) was administered intravenously with the help of a power injector at a rate of 3.0 ml/s, followed by a bolus injection of a 30-ml saline flush. A series of 384 images at 32 dynamics for 12 slices were acquired with a temporal resolution of 3.9 s. Total DCE-MRI acquisition time was 2 min, 06 s. Table 1 summarizes the imaging parameters protocol.

curve from signal intensity-time curve using in house developed Java based software. An automatic AIF was extracted and corrected for partial volume effect using the algorithm described by Singh et al. [17] and CBV were estimated using first-pass analysis [18]. CBV ¼

Absolute T10 was quantified using the TSE images (T1-weighted, T2weighted and PD-weighted) using the method described by Singh et al. [16], and then the T10 map was used to generate concentration time

ð1Þ

where C(t) is the concentration time curve, Cp(t) is the measured arterial input function, ρ (1.04 g/ml) is the density of brain tissue, H = (1 − Hlv) / (1 − Hcap) is a proportionality constant accounts for the hematocrits level difference between capillaries (Hcap) and large vessels (Hlv). Kinetic parameters (Ktrans, Kep, Ve, Vp and λtr) were estimated using leaky tracer kinetic model (LTKM) [19]: C ðt Þ ¼ V p C P ðt Þ þ V e C e ðt Þ þ C l ðt Þ

ð2Þ

where Vp is the fractional plasma volume, Ve is the fractional EES volume. Combining (Eq. (1)) and (Eq. (2)) CBV can be written as: CBV ¼

2.2. Quantification of perfusion parameters

R H C ðt Þ R dt ρ C p ðt Þ

H ρ

 R V p C P ðt Þ þ V e C e ðt Þ þ C l ðt Þ dt R C P ðt Þdt

ð3Þ

CBV can be corrected for disruption of BBB by considering only the plasma volume in (Eq. (3)) Corrected CBV = Hρ V p . 2.3. Automatic normalization T1 weighted images were segmented into gray matter, white matter and CSF using 5-class-K-mean clustering from Image J [20]. The slices which appeared to be normal in FLAIR, T2w and post-contrast T1 images were selected for segmentation according to the user's input. The slices selected for white matter/gray matter segmentation of the contra-lateral region where there is no abnormality on contra-lateral side or a slice above or below the abnormal region if the contra-lateral region is also involved. Histogram analysis of CBV was obtained from segmenting gray matter and white matter. Mean CBV value was extracted from the histogram by using Gaussian fit. The whole CBV stack was divided by the mean of normal white matter CBV value and mean normal gray matter CBV to generate the rCBV_WM and rCBV_GM map respectively. Absolute color scale was used for visualization of rCBV map. Details of the method are summarized in the flow chart (Fig. 1). 2.4. ROI placement

Fig. 2. Receiver Operating Characteristic curves of rCBV_contra measured by expert user and non-expert user, rCBV_WM and rCBV_GM.

One expert and one non-expert radiologist independently placed all ROI's. Expert user has 10 years' experience in DCE perfusion imaging and 35 years' experience in Radiology while non-expert user has 1 years' experience in DCE perfusion imaging and 5 years' experience in Radiology. Both the observers were blinded to each other and histopathological diagnosis of the patients. Tumor boundary was defined on T2/FLAIR images as hyperintensity with or without enhancement on post contrast study. Perfusion maps were overlaid on these

P. Sahoo et al. / Magnetic Resonance Imaging 44 (2017) 32–37

35

Fig. 3. 45-year-old male patient with glioblastoma multiforme shows T1-weighted (a), FLAIR (b), T2 weighted (c) MR images, automatic segmentation of normal white matter (d), gray matter (e) and perfusion images: CBV (f), CBV automatically normalized with segmented white matter (g), CBV automatically normalized with segmented gray matter (h). ROIs on CBV map (f) shows the ROI placement in tumor region (ROI 1) and contralateral region by non-expert user (ROI 2) and expert user (ROI 3). Contralateral ROI by non-expert user involves major vessel resulting miss classification of this tumor as low grade.

conventional images. Three type of rCBV were estimated by both the observers. rCBV_contra: ratio between maximum appearing CBV value in tumor region and the corresponding contralateral region CBV value. rCBV_GM: maximum appearing CBV automatic normalized with gray matter in tumor region. rCBV_WM: maximum appearing CBV automatic normalized with white matter in tumor region. The maximum CBV in the tumor region were identified in each slice, elliptical ROIs of size 15-30 mm2 were placed and mean CBV of the ROI was estimated. From this, the slice which provided maximum mean CBV value was considered for the statistical analysis. This method was followed for all patients.

2.5. Sample size estimation To test the inter method reliability between two tests at 5% level of significance and 95% power with assumption at null hypothesis H0(Intra-class correlation coefficient = 0) and alternative hypothesis H1(intra-class correlation coefficient = 0.50), total 38 subjects were needed. To test the agreement between both the methods using kappa statistics at 5% level of significance and 95% power on the basis of assumption of H0: Kappa = 0.50, a total of 57 subject were needed with two observations per subject. Thus minimum sample size of the study was 57 which were rounded off to 60.

Fig. 4. Bland-Altman plot for rCBV_GM (a), rCBV_WM (b) by expert verses non-expert user shows repeatability with the mean difference falling within the repeatability coefficient (RC) limits for the majority of subjects. X-axis represents the averaged rCBV value of expert and non-expert user and Y-axis represents the difference in rCBV value measured by Expert and nonexpert users.

36

P. Sahoo et al. / Magnetic Resonance Imaging 44 (2017) 32–37

Table 3 Diagnostic accuracy of contra-lateral based relative cerebral blood volume (rCBV) and normal white matter and gray matter based normalization method using histo-pathology results as reference. N = 60

Correctly classified

KAPPA

95% CI⁎

Agreement

rCBV_contra Expert Non-expert

59 42

0.96 0.40

0.81–0.99 0.14–0.60

98.33% 70.00%

rCBV_GM Expert Non-expert

48 48

0.58 0.58

0.34–0.75 0.34–0.75

80.00% 80.00%

rCBV_WM Expert Non-expert

53 53

0.76 0.76

0.54–0.88 0.54–0.88

88.33% 88.33%

ICC-intra-class correlation coefficient, CI-confidence interval. ⁎ Represents p b 0.001 for the confidence intervals.

2.6. Statistical analysis To decide the cutoff value of rCBV for glioma grading, the Receiver Operating Characteristic (ROC) curve analysis was used. Histopathological analysis data were considered as the standard to derive the cutoff value for each variable in ROC analysis. For cross validation, 1000 simulations of sample size 60 were generated. The 95% Confidence Interval for area under curve (AUC) for each variable was calculated using bias corrected bootstrap method [21]. Intra method reliability between both observers was tested using intra-class correlation coefficient (ICC). Kappa statistics was used to evaluate the level of agreement between histopathological grading and perfusion parameter based grading. With the cutoff value from the ROC analysis, numbers of misclassification from each group (WM tumor, GM tumor and mix tumor) were estimated. To check the reproducibility of the automatic normalization method, Bland-Altman (BA) plots were prepared between the two observer expert and non-expert estimated rCBV_GM and rCBV_WM. All statistical analysis was done using SPSS 22.0 for Windows (SPSS Inc., IBM, Chicago, IL) and R version 3.3.2 (R Core Team (2016)).

3. Result Optimal rCBV cutoff value, sensitivity and specificity of the contralateral based method and automatic normalization method for differentiating LGG from HGG is given in Table 2. ROC curves are shown in Fig. 2. In the contralateral based method, result of expert user shows higher sensitivity and specificity than the non-expert user; however, automatic normalization based techniques produces the same result for both the expert and non-expert user. Fig. 3 shows the difference in ROI placement by expert and non-expert user. For the automatic normalization method, the diagnostic accuracy was observer independent. There was no statistical difference (p = 0.65) in sensitivity and specificity for rCBV_WM (89% and 87%) and rCBV_GM (81% and 78%). However, sensitivity and specificity for rCBV_contra by non-expert user (65% and 78%) differ significantly with rCBV_GM (p b 0.01) and rCBV_WM (p b 0.01). Bland Altman plot showed that rCBV_GM and rCBV_WM derived by automated normalization method were reproducible with the mean Table 4 Showing the intra class correlation between rCBV values derived using various methods by expert and non-expert user.

rCBV_contra (expert v/s non-expert) rCBV_GM (expert v/s non-expert) rCBV_WM (expert v/s non-expert) rCBV_GM v/s rCBV_WM

ICC

95% CI

p value

0.23 0.997 0.997 0.72

0.02–0.46 0.994–0.998 0.995–9.998 0.57–0.82

0.030 b0.001 b0.001 b0.001

ICC-intra-class correlation coefficient, CI-confidence interval.

difference falling within the repeatability coefficient (RC) limits for the majority of subjects (Fig. 4). Result of agreement analysis and Kappa statistic is given in Table 3. As there was no significant difference in area under curve found in ROC analysis between rCBV_GM and rCBV_WM, a value near white matter cutoff was considered for deciding cutoff for both the observers. Kappa statistics showed that expert user results highly correlated with histological grading with kappa value 0.96; however non-expert user's results correlated poorly with histological grade with kappa value 0.40. The grading using normalization technique showed similar result for both expert and non-expert user with kappa value as 0.58 for gray matter normalization and 0.76 for white matter normalization. For rCBV_GM intra class correlation between two independent observers was 0.997 (p b 0.001) and for rCBV_WM intraclass correlation between two independent observers was 0.997 (p b 0.001) Significantly low correlation (ICC = 0.23, p = 0.03) was observed between rCBV_contra values measured by expert user and non-expert user, while significantly good correlation (ICC = 0.72, p = 0.001) was observed between rCBV_GM and rCBV_WM (Table 4). ICC for automated method is significantly different from ICC for rCBV_contra (p b 0.01). Measurement of rCBV_contra differs significantly with the experience of the radiologist (p = 0.038). However, the agreement with histological grading was increased with rCBV_WM (88.33%, kappa: 0.76) as compared with rCBV_GM (80.0%, kappa: 0.58) (Table 3). Out of the 60 tumors, 27 were located exclusively in white matter region, 4 exclusively in gray matter and rest 28 were mixed in location. Tumors located in the gray matter were not considered for further statistical analysis as the number was small. Of the rest of 55 cases, all were correctly characterized by rCBV_contra measured by expert user (Fig. 5). rCBV_WM correctly classified all WM tumor; however, misclassified 6 mixed tumors (Fig. 5b, e). Normalization using normal gray matter underestimates rCBV_GM value for the tumors located in white matter region as a result rCBV_GM misclassified 9 WM tumors (Fig. 5f); however, it misclassified only 6 mix tumors (Fig. 5c). 4. Discussion In contralateral based rCBV quantification, the results of expert and non-expert user differs significantly (p = 0.038); however, in case of normalized CBV, no significant difference was found between expert and non-expert user's results (p = 0.64). CBV normalization by white matter shows higher agreement in the classification of glioma grading as compared to that by gray matter. Although the diagnostic accuracy of the white matter normalization method was slightly lesser than the expert user's contralateral based rCBV (p = 0.25), it was found to be significantly higher than non-expert user's contralateral based rCBV measurements (p = 0.03). In the present study, rCBV_contra measured by expert user showed higher agreement with the histopathological grading of tumors. The expert user's ability to select an appropriate contralateral region (with similar proportion of gray and white matter volume, and avoiding major vessels) was responsible for higher accuracy in glioma grading. On the other hand, normalizing CBV with pure gray matter and pure white matter caused underestimation or overestimation of rCBV. It is quite evident that an expert user can place the ROIs very accurately while analyzing perfusion indices; however, the process is knowledge based and time consuming. In the case of non-expert user, rCBV_WM has showed the highest sensitivity and specificity in glioma grading implying that rCBV_WM may be used as an alternative method in absence of expert user. As the CBV values in GM is always found to be high than that in WM [22], normalizing CBV with normal gray matter underestimates rCBV_GM in WM tumor as gliomas are raised principally from the cerebral white matter [23]. In case of mixed tumors, both gray matter and white matter normalization provides similar results. As normal white matter can vary from region to region and also with age [13], selecting

P. Sahoo et al. / Magnetic Resonance Imaging 44 (2017) 32–37

37

Fig. 5. Scatter plot shows CBV value normalized with contralateral region (rCBV_contra) (a, d), CBV automatically normalized with white matter (rCBV_WM) (b, e) and CBV automatically normalized with gray matter (rCBV_GM) (c, f) of the mixed group tumor and white matter tumor group respectively. Dotted line shows the cutoff value decided from ROC analysis. Note the underestimation of rCBV normalization with gray matter misclassifies the tumors in the white matter region (f).

a reference region of interest in the white matter for normalization of CBV may be less reliable, than the proposed rCBV_WM method. Comparison of rCBV values between pure WM tumor and pure GM tumor could not be performed because of their small sample size. This might be considered as a limitation of this study. The evaluation of this automatic normalization method needs to be carried out in prospective studies for further validation in future. In conclusion we propose an alternative user-independent CBV normalization method for grading gliomas. The normalization of CBV by WM using the proposed method could provide better diagnostic accuracy than the manual contralateral based approach in the absence of expert user. References [1] Verma N, Cowperthwaite MC, Burnett MG, Markey MK. Differentiating tumor recurrence from treatment necrosis: a review of neuro-oncologic imaging strategies. Neuro Oncol 2013;15(5):515–34. [2] Mullins ME, Barest GD, Schaefer PW, Hochberg FH, Gonzalez RG, Lev MH. Radiation necrosis versus glioma recurrence: conventional MR imaging clues to diagnosis. AJNR Am J Neuroradiol 2005;26(8):1967–72. [3] Awasthi R, Rathore RKS, Soni P, et al. Discriminant analysis to classify glioma grading using dynamic contrast-enhanced MRI and immunohistochemical markers. Neuroradiology 2011;54(3):205–13. [4] Roy B, Awasthi R, Bindal A, et al. Comparative evaluation of 3-dimensional pseudocontinuous arterial spin labeling with dynamic contrast-enhanced perfusion magnetic resonance imaging in grading of human glioma. J Comput Assist Tomogr 2013;37(3):321–326.3. [5] Wang X, Zhang H, Tan Y, Qin J, Wu X, Wang L, et al. Combined value of susceptibilityweighted and perfusion-weighted imaging in assessing WHO grade for brain astrocytomas. J Magn Reson Imaging 2013;39(6):1569–74. [6] Santarosa C, Castellano A, Conte GM, et al. Dynamic contrast-enhanced and dynamic susceptibility contrast perfusion MR imaging for glioma grading: preliminary comparison of vessel compartment and permeability parameters using hotspot and histogram analysis. Eur J Radiol 2016;85(6):1147–56. [7] Haris M, Husain N, Singh A, et al. Dynamic contrast-enhanced derived cerebral blood volume correlates better with leak correction than with no correction for vascular endothelial growth factor, microvascular density, and grading of astrocytoma. J Comput Assist Tomogr 2008;32(6):955–65. [8] Jain KK, Sahoo P, Tyagi R, et al. Prospective glioma grading using single-dose dynamic contrast-enhanced perfusion MRI. ClinRadiol 2015;70(10):1128–35. [9] Emblem KE, Nedregaard B, Nome T, et al. Glioma grading by using histogram analysis of blood volume heterogeneity from MR-derived cerebral blood volume maps. Radiology 2008;247(3):808–17.

[10] Alcaide-Leon P, Pareto D, Martinez-Saez E, Auger C, Bharatha A, Rovira A. Pixel-bypixel comparison of volume transfer constant and estimates of cerebral blood volume from dynamic contrast-enhanced and dynamic susceptibility contrast-enhanced MR imaging in high-grade gliomas. AJNR Am J Neuroradiol 2015;36(5): 871–6. [11] Wetzel SG, Cha S, Johnson G, et al. Relative cerebral blood volume measurements in intracranial mass lesions: interobserver and intraobserver reproducibility study. Radiology 2002;224(3):797–803. [12] Thomsen H, Steffensen E, Larsson E. Perfusion MRI (dynamic susceptibility contrast imaging) with different measurement approaches for the evaluation of blood flow and blood volume in human gliomas. Acta Radiologica 2012;53(1):95–101. [13] Helenius J, Perkiö J, Soinne L, et al. Cerebral hemodynamics in a healthy population measured by dynamic susceptibility contrast MR imaging. ActaRadiol 2003;44(5): 538–46. [14] Revesz T, Scaravilli F, Coutinho L, Cockburn H, Sacares P, Thomas DG. Reliability of histological diagnosis including grading in gliomas biopsied by image-guided stereotactic technique. Brain 1993 Aug;116(4):781–93. [15] Prayson RA, Agamanolis DP, Cohen ML, et al. Interobserver reproducibility among neuropathologists and surgical pathologists in fibrillary astrocytoma grading. J Neurol Sci 2000 Apr 1;175(1):33–9. [16] Singh A, Haris M, Rathore D, et al. Quantification of physiological and hemodynamic indices using T(1) dynamic contrast-enhanced MRI in intracranial mass lesions. J Magn Reson Imaging 2007;26(4):871–80. [17] Singh A, Rathore RKS, Haris M, Verma SK, Husain N, Gupta RK. Improved bolus arrival time and arterial input function estimation for tracer kinetic analysis in DCE-MRI. J Magn Reson Imaging 2009;29(1):166–76. [18] Ostergaard L, Weisskoff RM, Chesler D, et al. High resolution measurement of cerebral blood flow using intravascular tracer bolus passages. Part II: experimental comparison and preliminary results. Magn Reson Med 1996;36(5):726–36. [19] Sahoo P, Rathore RKS, Awasthi R, et al. Subcompartmentalization of extracellular extravascular space (EES) into permeability and leaky space with local arterial input function (AIF) results in improved discrimination between high- and low-grade glioma using dynamic contrast-enhanced (DCE) MRI. J Magn Reson Imaging 2013; 38(3):677–88. [20] Jain Anil K, Dubes Richard C. Algorithms for clustering data. Upper Saddle River, NJ, USA: Prentice-Hall, Inc.; 1988 http://ij-plugins.sourceforge.net/plugins/segmentation/kmeans.html. [21] Carpenter J, Bithell J. Bootstrap confidence intervals: when, which, what? A practical guide for medical statisticians. Stat Med 2000;19(9):1141–64. [22] Lee MC, Cha S, Chang SM, Nelson SJ. Partial-volume model for determining white matter and gray matter cerebral blood volume for analysis of gliomas. J Magn Reson Imaging 2006;23(3):257–66. [23] Filley C. The behavioral neurology of cerebral white matter. Neurology 1998;50(6): 1535–40.