Diagnostic Values of DCE-MRI and DSC-MRI for Differentiation Between High-grade and Low-grade Gliomas

ARTICLE IN PRESS

Original Investigation

Diagnostic Values of DCE-MRI and DSC-MRI for Differentiation Between High-grade and Low-grade Gliomas: A Comprehensive Meta-analysis Jianye Liang, MD, Dexiang Liu, MD, Peng Gao, MD, Dong Zhang, MD, Hanwei Chen, MD, Changzheng Shi, MD, Liangping Luo, MD Rationale and Objectives: This study aimed to collect the studies on the role of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and dynamic susceptibility contrast MRI (DSC-MRI) in differentiating the grades of gliomas, and evaluate the diagnostic performances of relevant quantitative parameters in glioma grading. Materials and Methods: We systematically searched studies on the diagnosis of gliomas with DCE-MRI or DSC-MRI in Medline, PubMed, China National Knowledge Infrastructure database, Cochrane Library, and Embase published between January 2005 and December 2016. Standardized mean differences and 95% confidence intervals were calculated for volume transfer coefficient (Ktrans), volume fraction of extravascular extracellular space (Ve), rate constant of backflux (Kep), relative cerebral blood volume (rCBV), and relative cerebral blood flow (rCBF) using Review Manager 5.2 software. Sensitivity, specificity, area under the curve (AUC), and Begg test were calculated by Stata 12.0. Results: Twenty-two studies with available outcome data were included in the analysis. The standardized mean difference of Ktrans values between high-grade glioma and low-grade glioma were 1.18 (0.91, 1.45); Ve values were 1.43 (1.06, 1.80); Kep values were 0.65 (−0.05, 1.36); rCBV values were 1.44 (1.08, 1.81); and rCBF values were 1.17 (0.68, 1.67), respectively. The results were all significant statistically (P < .05) except Kep values (P = .07), and high-grade glioma had higher Ktrans, Ve, rCBV, and rCBF values than low-grade glioma. AUC values of Ktrans, Ve, rCBV, and rCBF were 0.90, 0.88, 0.93, and 0.73, respectively; rCBV had the largest AUC among the four parameters (P < .05). Conclusion: Both DCE-MRI and DSC-MRI are reliable techniques in differentiating the grades of gliomas, and rCBV was found to be the most sensitive one. Key Words: Gliomas; grading; DCE-MRI; DSC-MRI; meta-analysis. © 2017 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.

INTRODUCTION

G

liomas are the most common primary malignant tumors of the central nervous system. According to the 2016 World Health Organization (WHO) classification of tumors of the central nervous system (1), gliomas are divided into four grades based on their histology and molecular features. Accurate grading of gliomas is critical to the determination of surgery scheme, treatment response, and Acad Radiol 2017; ■:■■–■■ From the Medical Imaging Center, The First Affiliated Hospital of Jinan University, No.613, Huangpu Road West Tianhe District, Guangzhou, 510630 (J.L., P.G., D.Z., C.S., L.L.); Department of Radiology, Guangzhou Panyu Central Hospital, Guangzhou, Guangdong, China (D.L., H.C.). Received August 17, 2017; revised October 15, 2017; accepted October 16, 2017. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. The first two authors contributed equally to this work. Address correspondence to: L.L. e-mail: [email protected]; C.S. e-mail: [email protected] © 2017 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.acra.2017.10.001

prognostic evaluation. On pathology, low-grade gliomas (LGGs) are slowly proliferating tumors that display cytological atypia but no signs of anaplasia, endothelial cell proliferation, or brisk mitotic activity (2). However, in high-grade gliomas (HGGs), substantial hyperplasia of anomalous cells can be observed, resulting in neovascularization and incomplete basement membrane of tumor neovasculature, which in turns leads to augmentation of microvascular permeability, a histologic marker of HGG (3). Furthermore, the abnormal vessels of tumors are usually tortuous and disorganized. The resultant disordered cerebral hemodynamics alter blood volume and blood flow directly. Conventional morphologic magnetic resonance imaging (MRI) can estimate benign and malignant tumors based grossly on the range of cytotoxic edema, hemorrhage, necrosis, signal intensity heterogeneity, and degree of enhancement. However, it has been reported that 9.5% HGG showed no enhancement, whereas 22.72% of LGG enhanced after contrast administration (4). Therefore, quantitative and reliable imaging 1

LIANG ET AL

methods are needed. Dynamic contrast-enhanced MRI (DCEMRI) is a noninvasive technology that provides information about the microcirculation of tumors. It assesses several valuable parameters including volume transfer coefficient (Ktrans), volume fraction of extravascular extracellular space (Ve), and rate constant of backflux (Kep), all of which can reflect the permeability of new vessels and are indicative of malignant grade of tumors (3). Dynamic susceptibility contrast MRI (DSC-MRI) is another advanced technique that provides perfusion information with parameters such as cerebral blood volume (CBV) and cerebral blood flow (CBF). Increased tumor vascularity and tumor grade correlate credibly with relative CBV (rCBV) and relative CBF (rCBF) (5). With the advent of the high-field MR scanner and the development of advanced imaging technologies, increasingly more studies have concentrated on grading of gliomas with DCEMRI and DSC-MRI in recent years. Law et al. (6) found rCBV was the best parameter in discriminating glioma grade, followed by CBF, CBV, and Ktrans. However, Patankar et al. (7) reported that Ktrans had a higher area under the curve (AUC) value than CBV for glioma discrimination (0.979 and 0.966, respectively). Furthermore, Zhang et al. (3) and Sun et al. (8) found that Kep had no significant difference in glioma grading (P > .05), whereas Wang et al. (9) reported that HGG had a lower Kep than LGG in pediatric gliomas (P < .01), which contradicted with the results of Awasthi et al. (10) and Roy et al. (11) The large variations in different studies may be because of various types of scanners, field strength, contrast agents, imaging protocols, parameters, and post-processing methods, etc. In some instances, because of small sample sizes and incomplete parameters of individual studies, the reliability and reproducibility of these two technologies remains unclear. Therefore, we propose a comprehensive meta-analysis with a large sample size to address contradictory findings from different studies and to evaluate the diagnostic performance of relevant parameters in the grading of gliomas, the results of which would provide more reliable information to clinicians.

MATERIALS AND METHODS Data Sources

Two reviewers searched for any literature concerned with grading gliomas with DCE-MRI or DSC-MRI in Medline, PubMed, China National Knowledge Infrastructure database, Cochrane Library, and Embase published between January 2005 and December 2016. Medical subject headings or search keywords were combined into a formula of (astrocytoma or glioblastoma or glial tumor or astrocytic tumor or glioma or oligodendroglioma or oligodendroglial tumor) and (DCEMRI or DSC-MRI or Kep or Ktrans or Ve or rCBV or rCBF), with the searching limitations in the title or abstract of the article. Only studies written in English or Chinese were reviewed. We also scrutinized references in the included studies and searched for newly published studies every 2 months. Manual retrieval was performed if necessary. 2

Academic Radiology, Vol ■, No ■■, ■■ 2017

Studies Selection

The following inclusion criteria were established: (1) DCEMRI or DSC-MRI was applied in differentiating different grades of gliomas; (2) at least one of the quantitative indices for Ktrans, Ve, Kep, rCBV, and rCBF could be extracted or calculated from the study; (3) all the cases had been diagnosed pathologically; (4) neither surgery nor chemotherapy was conducted before magnetic resonance examination; (5) the scores of quality assessment of included studies were at least 9 because the high quality of included studies is the foundation of a credible meta-analysis; the standards for evaluation were stated in the quality assessment section (12); (6) any histologic subtypes of gliomas were included; and (7) the following exclusion criteria were established: (1) animal experiments, such as those using rats; (2) any graduation thesis, meeting records, reviews, duplications, or studies that have not been published; (3) similar studies that were written by the same first authors. Those performing the analysis were blinded to the institution. (4) Lack of key data (eg, standard deviation); and (5) other imaging modalities (eg, computed tomography, positron emission tomography) were used.

Data Abstraction and Quality Assessment

In accordance with 2007 WHO criteria, gliomas of grades I and II were classified into LGG, and grades III and IV into HGG (13), taking the average of grades I and II, and grades III and IV as the mean value if the data had not been merged. Two reviewers extracted the data independently from each study including the author, year of publication, type of MR machine, country, age of patients, types of gliomas, publication journal, DCE and DSC sequences, kinetic models, leakage correction, contrast type, flow rate, dose, numbers of oligoastrocytomas and oligodendrogliomas, post-processing software, mean value, and standard deviation of the related parameters according to the inclusion and exclusion criteria. True positive, false positive, false negative, and true negative data were also necessary to calculate diagnosis values. The revised Quality Assessment of Diagnostic Accuracy Studies checklist was used to assess the quality of each study with 14 criteria in terms of the risk of bias (14). Each criterion was judged as “Yes (low risk of bias),” “No (high risk of bias),” or “Unclear.” When a criterion was judged as Yes, the score increased by one. If the results contradicted each other, especially in terms of quality assessment, another senior clinician or statistician was invited to discuss the results to achieve a consensus.

Data Synthesis

Review Manager software version 5.2 (Cochrane Collaboration, Oxford, UK) was applied to calculate the effect size and the 95% confidence interval (CI). Stata version 12.0

Academic Radiology, Vol ■, No ■■, ■■ 2017

(StataCorp LP, College Station, TX) was used to assess publication bias by calculating the P value. Heterogeneity, which derives from types of study design, age and gender of patients, pathology subtypes, and other variables, is a potential and key factor that can affect the accuracy of the results. We estimated heterogeneity by calculating the inconsistency index (I2) as well as chi-square value; I2 > 50% or P < .05 was recognized as potential heterogeneity. If heterogeneity existed, a random effects model was used to calculate the pooling effect and 95% CI. Otherwise, we used a fixed effects model. As the parameters of each of the studies varied to some extent, we adopted standardized mean differences (SMD) as the pooling effect, which suggested less heterogeneity compared to weighted mean difference. Pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio, and AUC with their 95% CIs were calculated by Stata 12.0. Pooled indices were performed with a bivariate mixed effects binary regression model.

GRADING GLIOMAS WITH DCE-MRI AND DSC-MRI

Figure 1. Flowchart of the selection and identification process based on inclusion and exclusion criteria. After a careful screening, 22 studies that met the inclusion criteria are admitted.

RESULTS Literature Search and Selection of Studies

Fifty-seven studies were obtained after a primary search based on study titles and abstracts. Twenty-one studies were eliminated because they were duplications, animal experiments, or had not been confirmed by pathology. Downloading and reading full texts of the remaining 36 studies led to the exclusion of an additional four studies based on the lack of key data. A quality assessment score lower than nine excluded five studies, and four studies were excluded because treatment such as chemotherapy had been executed before examination. One study was excluded for the examination was performed on computed tomography. Eventually, a total of 22 studies (19 studies in English and three studies in Chinese) with 1123 patients who fulfilled all of the inclusion criteria were accepted for the analysis. The studies had been performed in the following regions: China (n = 6), USA (n = 3), India (n = 3), UK (n = 1), Korea (n = 2), Norway (n = 1), Thailand (n = 1), Italy (n = 1), Sweden (n = 1), Canada (n = 1), Brazil (n = 1), and Denmark (n = 1). The patients included both adults and children, and the ages of onset ranged from 4 to 85 years old. The quantitative results were denoted by effect sizes and 95% CIs in parentheses. A flowchart portraying the selection and identification process based on inclusion and exclusion criteria is shown in Figure 1. The main characteristics of the included studies are presented in Table 1 and Table 2. Quantitative Analysis

Ktrans Seventeen of the included studies that assessed Ktrans between HGG and LGG were analyzed. Heterogeneity tests showed χ2 = 37.60, I2 = 60%, P = .001, which indicated existence of

moderate heterogeneity. Therefore, Ktrans estimates were pooled using a random effects model, and the pooled SMD of Ktrans was 1.18 (0.91,1.45), P < .001 (Fig 2). No obvious publication bias effect was found using Begg test (P = .303). Ve Ten of the included studies that assessed Ve between HGG and LGG were analyzed. Heterogeneity tests showed χ2 = 27.10, I2 = 67%, P = .001, which indicated existence of mild heterogeneity. The Ve estimates were pooled using a random effects model, and the pooled SMD of Ve was 1.43 (1.06,1.80), P < .001 (Fig 3). No obvious publication bias was found using Begg test (P = .210). Kep Six of the included studies that assessed Kep between HGG and LGG were analyzed. Heterogeneity tests showed χ2 = 38.29, I2 = 87%, P < .001, which indicated existence of obvious heterogeneity. The Kep estimates were pooled using a random effects model, and the pooled SMD of Kep was 0.65 (−0.05,1.36), but the result was not statistically significant, P = .07 (Fig 4). No obvious publication bias was found using Begg test (P = .195). rCBV Twelve of included studies that assessed rCBV between HGG and LGG were analyzed. Heterogeneity tests showed χ2 = 34.92, I2 = 68%, P < .001, which indicated existence of mild heterogeneity. The rCBV estimates were pooled using a random effects model, and the pooled SMD of rCBV was 1.44 (1.08,1.81), P < .001 (Fig 5). No obvious publication bias was found using Begg test (P = .631). rCBF Seven of included studies that assessed rCBF between HGG and LGG were analyzed. Heterogeneity tests showed 3

LIANG ET AL

4 TABLE 1. Characteristics of Studies Included in the Meta-analysis

Year

Machine Type

Country

Age (y)

Type (Count)

Journal

Quality Assessment

Awasthi et al. (10) Falk et al. (2) Tietze et al. (15) Choi et al. (16) Roy et al. (11) Sahoo et al. (17) Jia et al. (18) Nguyen et al. (19) Arevalo-Perez et al. (20) Li et al. (21) Zhang et al. (3) Sun et al. (8) Wang et al. (9) Huang et al. (22) Server et al. (23) de Fatima Vasco Aragao et al. (24) Direksunthorn et al. (25) Law et al. (6) Santarosa et al. (26) Kim et al. (27) Patankar et al. (7) Boxerman et al. (28)

2012 2014 2015 2013 2013 2013 2015 2012 2015 2015 2012 2015 2015 2015 2011 2014 2013 2006 2016 2013 2005 2006

1.5T GE 3T Philips 3T Philips 3T Philips 3T GE 1.5T GE 3T Siemens 1.5T Siemens 1.5T GE 3T Siemens 1.5T Siemens 3T GE 3T Siemens 3T Siemens 3T GE 1.5T GE 3T Philips 1.5T Siemens 3T Philips 3T Siemens 1.5T Philips 1.5T GE

India Sweden Denmark Korea India India China Canada USA China China China China China Norway Brazil Thailand USA Italy Korea UK USA

16–65 22–79 NA 51.79 ± 18.34 43 21–63 46 ± 12 NA 54.3 42.6 ± 14.3 47.11 ± 14.18 45.7(14–73) 12.7 ± 4.6 45(17–72) 57.73 ± 12.95 36.23 ± 16.95 45.9(12–74) 42(4–85) 55.4(22–79) NA 52.9(31–77) 52(19–80)

I + II (21), III +IV (55) II(18), III(7) II(10), IV(23) I (1), II(9), III(8), IV(15) I(3), II(23),III (9), IV(29) I + II (45), III + IV (102) II (24), III (7), IV (26) II (8), III (4), IV (19) II (20), III (10), IV (33) I + II (15), III (8), IV (9) I (8), II (6), III (6), IV (8) I (2), II (10), III (7), IV (9) I + II (11), III + IV (5) I (2), II (34), III (21), IV (28) II (18), III (14), IV (47) I + II (9), III + IV (20) I + II (18), III + IV (26) II (31), III (16), IV (26) II (9), III (4), IV (13) II (9), III (16), IV (38) II (10), III (6), IV (23) II (11), III (9), IV (23)

Neuroradiology Neuroradiology Neuroradiology Korean J Radiol J Comput Assist Tomogr J Magn Reson Imaging Eur J Radiol Am J Neuroradiol J Neuroimaging Cancer Imaging J Magn Reson Imaging Acta acad Med sin Chin J Med Imaging Technol J Third Mil Med Univ Neuroradiology Am J Neuroradiol J Med Assoc Thai Am J Neuroradiol Eur J Radiol PLOS ONE Am J Neuroradiol Am J Neuroradiol

12 10 13 10 13 13 14 12 11 10 13 12 9 9 14 13 9 11 13 9 10 12

NA, not available.

Academic Radiology, Vol ■, No ■■, ■■ 2017

Author

Study

DCE Sequence

Model

Leakage Correction of rCBV

DSC Sequence

Contrast Type

Flow rate

Corrected NA Corrected

NA T2*W SS-GE-EPI NA

Omniscan Gadovist Gadovist

5 mL/s 3 mL/s 2.5 mL/s

NA

NA

Gadovist

Corrected NA

NA NA

NA

Post-Processing Software

0.2 mmol/kg 0.05 mmol/kg 0.05 mmol/kg

NA 7/8 0/6

NA MATLAB SPM8 and MATLAB

2 mL/s

0.1 mmol/kg

3/2

Omniscan Gd-DTPA

5 mL/s 5 mL/s

NA 0.2 mmol/kg

NA NA

NA

Omniscan

4 mL/s

0.1 mmol/kg

5/8

NA

NA

Magnevist

4 mL/s

0.1 mmol/kg

3/1

PRIDE tools(Philips Medical Systems) NA in-house-developed software Tissue-4D software (Siemens Syngo) NordicICE software

NA

NA

Magnevist

2–3 mL/s

0.2 mmol/kg

4/6

NordicICE software

NA

NA

Omniscan

4 mL/s

0.1 mmol/kg

2/6

NA

NA

Gd-DTPA

4 mL/s

0.1 mmol/kg

NA

Tissue-4D software (Siemens Syngo) MATLAB

NA NA

NA NA

Omniscan Gd-DTPA

2 mL/s 2–3 mL/s

0.1 mmol/kg 0.1 mmol/kg

6/1 0/0

NA

T2W SE-EPI

Gd-DTPA

4 mL/s

0.1 mmol/kg

NA

NA Corrected

T2*W SS-GE-EPI NA

Magnevist Magnevist

5 mL/s 4 mL/s

18 ml 0.1 mmol/kg

3/2 0/0 NA

Awasthi et al. (10) Falk et al. (2) Tietze et al. (15)

3D-SPGR T1W SPGR Turbo FLASH

Choi et al. (16)

NA

Roy et al. (11) Sahoo et al. (17)

3D-SPGR 3D-FSPGR

Jia et al. (18)

NA

Nguyen et al. (19)

2D FLASH

Arevalo-Perez et al. (20) Li et al. (21)

3D-SPGR

Zhang et al. (3) Sun et al. (8) Wang et al. (9)

3D Turbo FLASH NA T1-twist

Huang et al. (22)

T1-twist

Server et al. (23) de Fatima Vasco Aragao et al. (24) Direksunthorn et al. (25) Law et al. (6)

NA SE

Pharmacokinetic model Tikhonov method extended two compartment exchange model classic Tofs-Kermode model Pharmacokinetic model leaky tracer kinetic model classic Tofs-Kermode model Phase-Derived VIF with Bookend T1 Correction classic Tofs-Kermode model classic Tofs-Kermode model modified Tofts' two compartment model extended Tofts Linear model classic Tofs-Kermode model classic Tofs-Kermode model NA NA

NA

NA

NA

3D-PRESTO

Gadovist

4–5 mL/s

NA

NA

Corrected

GE-EPI

Magnevist

5 mL/s

0.1 mmol/kg

0/10

Santarosa et al. (26) Kim et al. (27) Patankar et al. (7)

GE

Corrected

T2*W GE-EPI

Gadovist

2–5 mL/s

10 ml

4/3

Corrected Corrected

SS-GE-EPI NA

Gadovist Omniscan

4 mL/s NA

0.1 mmol/kg 0.1 mmol/kg

0/0 NA

Boxerman et al. (28)

NA

classic Tofs-Kermode model classic Tofs-Kermode model NA biexponential model of PCCF NA

Corrected

SS-GE-EPI

Omniscan

3–5 mL/s

0.15–0.25 mmol/kg

2/5

T1-twist

NA 3D T1 FFE

Omni Kinetics(GE) Tissue-4D software (Siemens Syngo) Tissue-4D software (Siemens Syngo) NordicICE software FuncTool (GE AW4.2 workstation) proprietary analytic software((Philip) in-house-developed software NordicICE software NordicICE software in-house-developed software NA

5

DCE, dynamic contrast-enhanced; DSC, dynamic susceptibility contrast; EPI, echo planar imaging; FFE, fast-field echo sequence; FLASH, fast low angle shot; FSPGR, fast spoiled gradient echo images; GE, gradient echo; NA, Not available; OA, oligoastrocytomas; OG, oligodendrogliomas; SS, single shot.

GRADING GLIOMAS WITH DCE-MRI AND DSC-MRI

OA/OG

Dose

Academic Radiology, Vol ■, No ■■, ■■ 2017

TABLE 2. The Specific Parameters of DCE- and DSC-MRI of the Included Studies

LIANG ET AL

Academic Radiology, Vol ■, No ■■, ■■ 2017

Figure 2. Forest plot of the mean value of volume transfer coefficient (Ktrans) between high-grade glioma (HGG) and low-grade glioma (LGG) (left chart listed by mean Ktrans ± SD min−1). The confidence intervals (CIs) of most studies are on the right side of the central axis (x = 0), which indicates a significant difference between HGG and LGG.

Figure 3. Forest plot of the mean value of volume fraction of extravascular extracellular space (Ve) between high-grade glioma (HGG) and low-grade glioma (LGG) (left chart listed by mean Ve ± SD). A statistical difference is observed in Ve.

χ2 = 22.79, I2 = 74%, P < .001, which indicated existence of obvious heterogeneity. The rCBF estimates were pooled using a random effects model, and the pooled SMD of rCBF was 1.42 (0.65,2.19), P < .001 (Fig 6). Publication bias may exist statistically using Begg test (P = .049).

characteristic curve of each index was symmetric. The sensitivity of Ktrans, rCBV, and rCBF showed mild heterogeneities (I2 = 63.80%, 61.22%, and 57.33% respectively), and specificity of four indices had no obvious heterogeneity. The results suggested that rCBV had the highest diagnostic odds ratio 46 (23,94) and AUC values 0.93 (0.90,0.95) (Fig 7).

Diagnosis Values

Corresponding pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio, and AUC are shown in Table 3. There were no obvious threshold effects in the accuracy estimation among the four indices (P > .05). The summary receiver operating 6

Sensitivity Analysis

Sensitivity analysis is an important method of dealing with heterogeneity and publication bias. We eliminated an individual study and calculated the pooled effect of the rest of

Academic Radiology, Vol ■, No ■■, ■■ 2017

GRADING GLIOMAS WITH DCE-MRI AND DSC-MRI

Figure 4. Forest plot of the mean value of rate constant of backflux (Kep) between high-grade glioma (HGG) and low-grade glioma (LGG) (left chart listed by mean Kep ± SD min-1). Studies are distributed on both sides of the central axis, indicating no obvious difference between HGG and LGG.

Figure 5. Forest plot of the mean value of relative cerebral blood volume (rCBV) between high-grade glioma (HGG) and low-grade glioma (LGG) (left chart listed by mean rCBV ± SD). A positive result is observed between HGG and LGG.

Figure 6. Forest plot of the mean value of relative cerebral blood flow (rCBF) between high-grade glioma (HGG) and low-grade glioma (LGG) (left chart listed by mean rCBF ± SD). A positive difference exists in different grades of gliomas.

studies. Comparing the results to the pooled effect of all the included studies, we would determine the influence of individual study on the overall effect. Results of this meta-analysis revealed that the included studies had no significant change on the pooled value of SMD. However,

the Ktrans of Zhang et al. (3) showed significant influence on heterogeneity and publication bias before it was eliminated (I2 = 71% to I2 = 64% calculated by Revman5.2, P value of Egger test: P = .038 to P = .132 calculated by Stata 12.0). 7

Academic Radiology, Vol ■, No ■■, ■■ 2017

LIANG ET AL

TABLE 3. Diagnosis Values of Ktrans, Ve, rCBV, and rCBF Combined Index Studies trans

K Ve rCBV rCBF

13 6 8 5

Sensitivity (95% CI)

Specificity (95% CI)

PLR (95% CI)

NLR (95% CI)

DOR (95% CI)

AUC (95% CI)

0.88(0.81,0.93) 0.85(0.73,0.92) 0.91(0.83,0.95) 0.88(0.77,0.94)

0.80(0.72,0.86) 0.84(0.75,0.91) 0.82(0.71,0.90) 0.68(0.56,0.77)

4.3(3.0,6.2) 5.5(3.3,9.2) 5.1(3.1,8.6) 2.7(1.9,3.9)

0.15(0.09,0.25) 0.18(0.09,0.34) 0.11(0.06,0.20) 0.18(0.09,0.36)

28(14,58) 31(12,78) 46(23,94) 15(6,40)

0.90(0.87,0.92) 0.88(0.85,0.91) 0.93(0.90,0.95) 0.73(0.69,0.77)

I2 Sensitivity Specificity 63.80% 44.87% 61.22% 57.33%

39.88% 0 42.05% 36.16%

AUC, area under the curve; CI, confidence interval; DOR, diagnostic odds ratio; Ktrans, volume transfer coefficient; NLR, negative likelihood ratio; PLR, positive likelihood ratio; rCBF, relative cerebral blood flow; rCBV, relative cerebral blood volume; Ve, volume fraction of extravascular extracellular space. Ktrans (2,6–9,15,16,19–22,26,29); Ve (3,8,9,16,21,22); rCBV (6,15,22–27); rCBF (2,15,22,23,25). Kep was not listed because no statistic difference was seen in SMD.

DISCUSSION Currently, increasingly more histologic changes and molecular features are investigated to grade gliomas. Providing a precise preoperative determination of tumor grade is helpful for clinicians choosing surgical and subsequent treatment protocols (30). Glioma is a kind of malignant tumor that is not only characterized by a variable population of atypical cells and abnormal mitoses, but also by obvious hyperplasia of new vessels, disruption of the blood brain barrier (BBB), and simultaneous changes in local hemodynamics (31). The malignant degrees of brain gliomas and biological characteristics have been found to have obvious positive correlation with the degree of neovascularization (32). Compared to microvessels of normal brain, the tumors tend to be immature, manifested by incomplete basement membranes and higher permeability, resulting in leakage of blood components to the extravascular extracellular space (22). Microvessels of HGG might be inadequate to meet the requirements for the further growth of glioma, resulting in formation of new immature microvessels to acquire additional nutrition, whereas fewer new microvessels are required in LGG. In addition, the mature microvessels of LGG usually have lower permeability, with architecture and density similar to those of normal brain tissue (18). Furthermore, because of the existence of the intact BBB, no contrast medium could leak from intravascular to extravascular space, and Ktrans and Ve tend to be low in LGG. Therefore, microvascular permeability can reflect the new microvessel proliferation of the tumor indirectly, which can be used to evaluate the malignant degree of glioma. In Zhang et al.’s study (33), heparin-binding cytokine pleiotrophin was identified as a stimulus of vascular proliferation in glioma. Pleiotrophin was found much richer in high-grade astrocytomas and closely related with poor survival. Neovascular endothelial cells of tumor had been shown to overexpress the CD105, suggesting that CD105 is closely related to the growth of tumor. Furthermore, CD105positive microvessel density, on behalf of the immature microvessels, moderately correlated with Ktrans and Ve, which indicates Ktrans and Ve could be used to assess immature 8

microvessel density in glioma (18). Using CD34 staining, Nguyen et al. (34) also found a strong correlation between microvessel area and Vp and blood volume, and a moderate correlation between microvessel area and Ktrans. Zhang et al. (3) found that between high- (III and IV) and low- (I and II) grade gliomas, the discriminative values of sensitivity and specificity were 92% and 85% by Ktrans, and 90% and 78% by Ve, respectively. The AUCs were 0.964 for Ktrans and 0.929 for Ve, which were in strong agreement with the findings in this meta-analysis. Because HGG are more aggressive and have luxuriant blood supply compared to LGG, hemodynamic perfusion indices (rCBV and rCBF) would manifest notable increases. Awasthi et al. (10) observed a significant positive correlation between immunohistochemical expression of vascular endothelial growth factor (VEGF) with rCBV and rCBF. Haris et al. (4) found the corrected rCBV correlated better with MVD and VEGF. The analysis of Xyda et al. (29) revealed that the discriminating values of CBV and CBF have comparable diagnostic accuracy with a sensitivity of 93% and 90%, respectively, and a specificity of 94% for both parameters. Awasthi et al. (10) also found that the degree of BBB disruption was associated with the aggressiveness and progression of the tumor. Quantification of rCBV suffered from leakage effect because of disrupted BBB, which may result in underestimating rCBV of glioma (35). This leads to the combination of permeability parameters along with rCBV and rCBF in discriminant analysis and may improve the accuracy of gliomas grading. In this meta-analysis, the pooled SMD of Ktrans, Ve, rCBF, rCBV of HGG were higher than LGG with the results significant statistically, which were generally consistent with the published studies. The pooled SMD of Kep of HGG was also slightly higher than LGG; however, the result was not statistically significant. In Awasthi et al.’s study (36), Kep, Ktrans, and Ve were found significantly correlated with matrix metalloproteinase-9 expression, and matrix metalloproteinase-9 was best estimated by Kep using a quadratic model. Higher matrix metalloproteinase-9 expression was associated with higher Kep, and lower 1-year survival. Although the result seemed unsatisfactory in this meta-analysis, Kep still played an important role in evaluating progression and prognostication of

Academic Radiology, Vol ■, No ■■, ■■ 2017

GRADING GLIOMAS WITH DCE-MRI AND DSC-MRI

Figure 7. Summary receiver operating characteristic curve of volume transfer coefficient (Ktrans) (a), volume fraction of extravascular extracellular space (Ve) (b), relative cerebral blood volume (rCBV) (c), and relative cerebral blood flow (rCBF) (d) in the glioma grading. The largest area under the curve (AUC) among the four parameters was rCBV, followed by Ktrans, Ve, and rCBF. SROC, summary receiver operating characteristic.

glioblastoma multiforme. More studies evaluating Kep should be included to verify the result in the future. The major augmentation of the aforementioned parameters in HGG in comparison to LGG supports the notion that hemodynamic and permeable parameters are valuable diagnostic markers for the classification of gliomas before histopathologic diagnosis. In addition, rCBV had the highest differential value with the largest AUC among the four

parameters (AUC = 0.93), with results similar to Nguyen et al.’s study (37) in the preoperative grading of gliomas. Heterogeneity is common in meta-analysis. An exploration of the contributions for heterogeneity is an important goal rather than the calculation of a single summary measure. First, statistical analysis with Review Manager 5.2 indicated mild heterogeneity of Ktrans (I2 = 60%), whereas Stata 12.0 indicated no heterogeneity of Ktrans (I2 = 49.6%), but indicated 9

LIANG ET AL

heterogeneity of Ve (I2 = 67% versus I2 = 35.2%). The heterogeneity of rCBF decreased from I2 = 91% to 74% after excluding the study of Sahoo et al. (17). Second, the heterogeneity of Kep and rCBF had already surpassed 70%,which indicated obvious heterogeneity; caution should be taken when making a clinical decision based on these results. Except for the limited number of studies available, a possible explanation was imbalance of the composition of HGG and LGG. For example, there were 11 studies lacking grade I gliomas, one study lacking grade III, and one study lacking grade IV. Third, different types of pharmacokinetic models used in DCEMRI likely influenced the accuracy of the results. Most studies were based on the Tofts and Kermode model; however, Sahoo et al. (17) did some modifications by introducing an additional tissue uptake leakage compartment in extracellular extravascular space, whereas Harrer et al. (38) applied an individually measured equation from an automatically calculated arterial input function. Fourth, the accuracy of rCBV measurements may be affected by non-correction of contrast leakage with disruption of BBB. In this meta-analysis study, nine studies performed corrections with a preloaded doses of contrast agent (24,28), or by using some mathematical algorithms (6,26,27). Although such corrections had not been found in other studies, which may result in underestimation of the rCBV (26). For DCE permeability studies, different post-processing software may affect the accuracy of Ktrans and Ve measurements as well. Lastly, oligodendrogliomas included in most studies may have abnormally elevated rCBV even at lower grades (39).

LIMITATIONS Some limitations of this meta-analysis should be taken into account. First, the applicable studies of Kep and rCBF were too few to be analyzed. The next step is to collect more relevant studies to expand the sample sizes. Second, the classification of gliomas in most studies were based on the histopathology according to 2007 WHO criteria. However, certain astrocytomas diagnosed as grade III with specific genetic expression do not show significant increase of rCBV because of the lack of microvascular proliferation, which could potentially affect their inclusion in the same perfusion group with the glioblastomas. The updated 2016 WHO classification criteria emphasize tumor genomics given their better outcome prognostication compared to tumor histopathology. Metaanalysis of imaging studies with glioma grading based on 2016 WHO criteria is warranted in the future. The selection of region of interest varied from study to study; the parameters may be affected by the selection of the region of interest, especially with respect to necrotic portions of the tumor, which is easily seen in HGG.

CONCLUSIONS Both DCE-MRI and DSC-MRI are useful noninvasive imaging techniques for differentiating HGGs from LGGs, and 10

Academic Radiology, Vol ■, No ■■, ■■ 2017

this analysis suggests that rCBV may be regarded as the best surrogate for the characterization of gliomas preoperatively. REFERENCES 1. Louis DN, Perry A, Reifenberger G, et al. The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol 2016; 131:803–820. https://doi.org/10.1007/ s00401-016-1545-1. 2. Falk A, Fahlstrom M, Rostrup E, et al. Discrimination between glioma grades II and III in suspected low-grade gliomas using dynamic contrastenhanced and dynamic susceptibility contrast perfusion MR imaging: a histogram analysis approach. Neuroradiology 2014; 56:1031–1038. https://doi.org/10.1007/s00234-014-1426-z. 3. Zhang N, Zhang L, Qiu B, et al. Correlation of volume transfer coefficient Ktrans with histopathologic grades of gliomas. J Magn Reson Imaging 2012; 36:355–363. https://doi.org/10.1002/jmri.23675. 4. 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; 32:955– 965. https://doi.org/10.1097/RCT.0b013e31816200d1. 5. Hussain NS, Moisi MD, Keogh B, et al. Dynamic susceptibility contrast and dynamic contrast-enhanced MRI characteristics to distinguish microcystic meningiomas from traditional Grade I meningiomas and highgrade gliomas. J Neurosurg 2016; 1–7. https://doi.org/10.3171/ 2016.3.JNS14243. 6. Law M, Young R, Babb J, et al. Comparing perfusion metrics obtained from a single compartment versus pharmacokinetic modeling methods using dynamic susceptibility contrast-enhanced perfusion MR imaging with glioma grade. AJNR Am J Neuroradiol 2006; 27:1975–1982. 7. Patankar TF, Haroon HA, Mills SJ, et al. Is volume transfer coefficient (K(trans)) related to histologic grade in human gliomas? AJNR Am J Neuroradiol 2005; 26:2455–2465. 8. Sun SJ, Qian HF, Li FQ, et al. Diagnostic value of combining permeability with T1 perfusion parameters in quantitative dynamic contrastenhanced magnetic resonance imaging for glioma grading. Acta Academiae Medicinae Sinicae 2015; 37:674–680. https://doi.org/10 .3881/j.issn.1000-503X.2015.06.007. 9. Wang JL, Fan M, Chu JP, et al. Application of parameters of dynamic contrast-enhanced MRI and ADC in grading pediatric gliomas. Chin J Med Imaging Technol 2015; 31:678–682. 10. Awasthi R, Rathore RK, Soni P, et al. Discriminant analysis to classify glioma grading using dynamic contrast-enhanced MRI and immunohistochemical markers. Neuroradiology 2012; 54:205–213. https://doi.org/10 .1007/s00234-011-0874-y. 11. 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; 37:321–326. https://doi.org/10.1097/ RCT.0b013e318282d7e2. 12. Hu Q, Wang XY, Zhu SY, et al. Meta-analysis of contrast-enhanced ultrasound for the differentiation of benign and malignant breast lesions. Acta Radiol 2015; 56:25–33. https://doi.org/10.1177/0284185113517115. 13. Louis DN, Ohgaki H, Wiestler OD, et al. The 2007 WHO classification of tumours of the central nervous system. Acta Neuropathol 2007; 114:97– 109. https://doi.org/10.1007/s00401-007-0243-4. 14. Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med 2011; 155:529–536. https://doi.org/10.7326/0003-4819-155-8 -201110180-00009. 15. Tietze A, Mouridsen K, Mikkelsen IK. The impact of reliable prebolus T 1 measurements or a fixed T 1 value in the assessment of glioma patients with dynamic contrast enhancing MRI. Neuroradiology 2015; 57:561– 572. https://doi.org/10.1007/s00234-015-1502-z. 16. Choi HS, Kim AH, Ahn SS, et al. Glioma grading capability: comparisons among parameters from dynamic contrast-enhanced MRI and ADC value on DWI. Korean J Radiol 2013; 14:487–492. https://doi.org/ 10.3348/kjr.2013.14.3.487. 17. Sahoo P, Rathore RK, 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

Academic Radiology, Vol ■, No ■■, ■■ 2017

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

(DCE) MRI. J Magn Reson Imaging 2013; 38:677–688. https://doi.org/ 10.1002/jmri.24021. Jia ZZ, Gu HM, Zhou XJ, et al. The assessment of immature microvascular density in brain gliomas with dynamic contrast-enhanced magnetic resonance imaging. Eur J Radiol 2015; 84:1805–1809. https://doi.org/ 10.1016/j.ejrad.2015.05.035. Nguyen TB, Cron GO, Mercier JF, et al. Diagnostic accuracy of dynamic contrast-enhanced MR imaging using a phase-derived vascular input function in the preoperative grading of gliomas. AJNR Am J Neuroradiol 2012; 33:1539–1545. https://doi.org/10.3174/ajnr.A3012. Arevalo-Perez J, Peck KK, Young RJ, et al. Dynamic contrast-enhanced perfusion MRI and diffusion-weighted imaging in grading of gliomas. J Neuroimaging 2015; 25:792–798. https://doi.org/10.1111/jon.12239. Li X, Zhu Y, Kang H, et al. Glioma grading by microvascular permeability parameters derived from dynamic contrast-enhanced MRI and intratumoral susceptibility signal on susceptibility weighted imaging. Cancer Imaging 2015; 15:4. https://doi.org/10.1186/s40644-015-0039-z. Huang J, Li XG, Kang HY, et al. Dynamic contrast enhanced magnetic resonance imaging and dynamic susceptibility contrast magnetic resonance imaging in grading of glioma. J Third Mil Med Univ 2015; 37:672– 677. Server A, Graff BA, Orheim TE, et al. Measurements of diagnostic examination performance and correlation analysis using microvascular leakage, cerebral blood volume, and blood flow derived from 3T dynamic susceptibility-weighted contrast-enhanced perfusion MR imaging in glial tumor grading. Neuroradiology 2011; 53:435–447. https://doi.org/10.1007/ s00234-010-0770-x. de Fatima Vasco Aragao M, Law M, Batista de Almeida D, et al. Comparison of perfusion, diffusion, and MR spectroscopy between lowgrade enhancing pilocytic astrocytomas and high-grade astrocytomas. AJNR Am J Neuroradiol 2014; 35:1495–1502. https://doi.org/10 .3174/ajnr.A3905. Direksunthorn T, Chawalparit O, Sangruchi T, et al. Diagnostic performance of perfusion MRI in differentiating low-grade and high-grade gliomas: advanced MRI in glioma, A Siriraj project. J Med Assoc Thai 2013; 96:1183–1190. 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; 85:1147–1156. https://doi.org/10.1016/j.ejrad.2016.03.020. Kim H, Choi SH, Kim JH, et al. Gliomas: application of cumulative histogram analysis of normalized cerebral blood volume on 3 T MRI to tumor grading. PLoS ONE 2013; 8:e63462. https://doi.org/10.1371/ journal.pone.0063462. Boxerman JL, Schmainda KM, Weisskoff RM. Relative cerebral blood volume maps corrected for contrast agent extravasation significantly

GRADING GLIOMAS WITH DCE-MRI AND DSC-MRI

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

correlate with glioma tumor grade, whereas uncorrected maps do not. AJNR Am J Neuroradiol 2006; 27:859–867. Xyda A, Haberland U, Klotz E, et al. Brain volume perfusion CT performed with 128-detector row CT system in patients with cerebral gliomas: a feasibility study. Eur Radiol 2011; 21:1811–1819. https://doi.org/10.1007/ s00330-011-2150-2. Grabner G, Kiesel B, Wohrer A, et al. Local image variance of 7 Tesla SWI is a new technique for preoperative characterization of diffusely infiltrating gliomas: correlation with tumour grade and IDH1 mutational status. Eur Radiol 2017; 27:1556–1567. https://doi.org/10.1007/s00330-016 -4451-y. Jimenez-Heffernan JA, Freih Fraih A, Alvarez F, et al. Cytologic features of pleomorphic xanthoastrocytoma, WHO grade II. A comparative study with glioblastoma. Diagn Cytopathol 2017; 45:339–344. https://doi.org/10.1002/dc.23660. Jia ZZ, Geng DY, Liu Y, et al. Microvascular permeability of brain astrocytoma with contrast-enhanced magnetic resonance imaging: correlation analysis with histopathologic grade. Chin Med J 2013; 126:1953–1956. Zhang L, Kundu S, Feenstra T, et al. Pleiotrophin promotes vascular abnormalization in gliomas and correlates with poor survival in patients with astrocytomas. Sci Signal 2015; 8:ra125. https://doi.org/10.1126/ scisignal.aaa1690. Nguyen TB, Cron GO, Bezzina K, et al. Correlation of tumor immunohistochemistry with dynamic contrast-enhanced and DSC-MRI parameters in patients with gliomas. AJNR Am J Neuroradiol 2016; 37:2217–2223. https://doi.org/10.3174/ajnr.A4908. 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; 247:808–817. https://doi.org/10 .1148/radiol.2473070571. Awasthi R, Pandey CM, Sahoo P, et al. Dynamic contrast-enhanced magnetic resonance imaging-derived kep as a potential biomarker of matrix metalloproteinase 9 expression in patients with glioblastoma multiforme: a pilot study. J Comput Assist Tomogr 2012; 36:125–130. https://doi.org/ 10.1097/RCT.0b013e31823f6c59. Nguyen TB, Cron GO, Perdrizet K, et al. Comparison of the diagnostic accuracy of DSC- and dynamic contrast-enhanced MRI in the preoperative grading of astrocytomas. AJNR Am J Neuroradiol 2015; 36:2017– 2022. https://doi.org/10.3174/ajnr.A4398. Harrer JU, Parker GJ, Haroon HA, et al. Comparative study of methods for determining vascular permeability and blood volume in human gliomas. J Magn Reson Imaging 2004; 20:748–757. https://doi.org/10.1002/ jmri.20182. Cha S, Tihan T, Crawford F, et al. Differentiation of low-grade oligodendrogliomas from low-grade astrocytomas by using quantitative blood-volume measurements derived from dynamic susceptibility contrastenhanced MR imaging. AJNR Am J Neuroradiol 2005; 26:266–273.

11