Clinical Radiology 2018.pdf

Clinical Radiology xxx (2018) 1e9

Contents lists available at ScienceDirect

Clinical Radiology journal homepage: www.clinicalradiologyonline.net

Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1-perfusion, diffusion, and susceptibility-weighted MRI J. Saini a, y, P. Kumar Gupta b, y, A. Awasthi c, C.M. Pandey c, A. Singh d, R. Patir e, S. Ahlawat f, N. Sadashiva g, A. Mahadevan h, R. Kumar Gupta b, *, y a

Neuroimaging & Interventional Radiology, National Institute of Mental Health and Neurosciences, Bangalore, India Departments of Radiology and Imaging, Fortis Memorial Research Institute, Gurugram, India c Biostatistics, Sanjay Gandhi Post Graduate Institute of Medical Sciences, Lucknow, India d Center for Biomedical Engineering, Indian Institute of Technology Delhi, Delhi, India e Department of Neurosurgery, Fortis Memorial Research Institute, Gurugram, India f SRL Diagnostics, Fortis Memorial Research Institute, Gurugram, India g Department of Neurosurgery, National Institute of Mental Health and Neurosciences, Bangalore, India h Department of Neuropathology, National Institute of Mental Health and Neurosciences, Bangalore, India b

art icl e i nformat ion Article history: Received 5 June 2018 Accepted 31 July 2018

AIM: To compare the diagnostic performance of T1 perfusion magnetic resonance imaging (MRI), diffusion-weighted imaging (DWI), and susceptibility-weighted imaging (SWI) for differentiating primary central nervous system lymphoma (PCNSL) and glioblastoma (GBM). MATERIALS AND METHODS: This retrospective study comprised a cohort of 70 patients with glioblastoma and 30 patients with PCNSL. T1 perfusion MRI-derived rCBV_corr (leakage corrected relative cerebral blood volume), apparent diffusion coefficient (ADC) derived from DWI, and intratumoural susceptibility signals intensity (ITSS) measured on SWI were evaluated in these 100 patients. The ManneWhitney U-test was used for pairwise comparison between groups. The diagnostic performance for differentiating PCNSL from glioblastoma was evaluated by using univariate and multivariable logistic regression analyses and receiver operating characteristic (ROC) analysis. RESULTS: Minimum ADC, maximum rCBVs_corr, kep (back flux exchange rate), and ITSS scores were significantly lower in patients with PCNSL than in those with glioblastoma (p<0.05). On ROC analysis, ITSS showed the best discrimination ability for differentiation of GBM and PCNSL with an area under the ROC curve (AUC) of 0.80. rCBV_corr and ADC showed AUCs of 0.68 and 0.63, respectively. Multiparametric assessment using ADC, rCBV_corr, kep, and ITSS scores significantly increased the diagnostic ability for differentiating PCNSL from GBM as compared to mean ADC, mean rCBV_corr, and ITSS alone or a combination of these parameters. The multiparametric model could correctly discriminate 84% of tumours with a sensitivity and specificity of 90% and 70% with an AUC of 0.92.

* Guarantor and correspondent: R. Kumar Gupta, Department of Radiology and Imaging Fortis Memorial Research Institute, Gurgaon, Haryana, 122002, India. Tel.: þ91 9717988859; fax: þ91 124 496 2222. E-mail address: [email protected] (R. Kumar Gupta). y These authors contributed equally to this work. https://doi.org/10.1016/j.crad.2018.07.107 0009-9260/Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

2

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

CONCLUSION: Multiparametric MRI evaluation using DWI, T1 perfusion MRI, and SWI enabled reliable differentiation of PCNSL and GBM in the majority patients, and these results support an integration of advanced MRI techniques for the diagnostic work-up of patients with these tumours. Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

Introduction Glioblastoma (GBM) and primary central nervous system lymphoma (PCNL) are two important neoplastic entities presenting as focal mass lesions of the brain. Glioblastoma treatment involves surgical resection along with radiation and chemotherapy whereas PCNSL requires biopsy for histological confirmation followed by high-dose chemotherapy.1 The mainstay of glioblastoma treatment is maximal possible safe surgical resection whereas there is no clear role of surgical resection in patients with lymphoma, although a retrospective analysis has shown some survival benefit in patients undergoing surgical debulking of the tumour.1,2 Considering the different treatment strategies for these two tumours, their preoperative differentiation is essential. Conventional MRI features are useful for discrimination of GBM from PCNSL. A hypercellular periventricular lesion with disrupted bloodebrain barrier (BBB) are some of the characteristic features of CNS lymphoma.3 Whereas a solitary, infiltrative supratentorial lesion with non-homogeneous enhancement and necrosis are the typical imaging features of GBM.4 The frequent presence of atypical imaging features may make differentiation of PCNSL from GBM challenging and advanced magnetic resonance imaging (MRI) techniques such as perfusion MRI, diffusion-weighted imaging (DWI) and susceptibilityweighted imaging (SWI) may prove useful in such cases.5 Multiple studies have compared advanced MRI methods, such as perfusion, DWI, and SWI, for discrimination of PCNSL from GBM. Most of these have used the dynamic susceptibility contrast (DSC) perfusion method to calculate the relative cerebral blood volume (rCBV) within the tumour and have found lower rCBV values in cases of lymphoma as compared to those of GBM3,6e11; however, PCNSL are known to show significant leakage and BBB breakdown,12 which leads to signal intensity increase due to T1 effect leading to the masking of the T2*-weighted signal intensity loss. This may lead to underestimation of CBV within leaky tumours.13,14 The problem of leakage has been addressed in DSC perfusion by using leakage-corrected models or use of a preloading technique; however, problems pertaining to susceptibility persist. These problems may be overcome by the use of the T1-perfusion technique, which is relatively resistant to susceptibility artefacts and provides additional information about tumour leakiness. Dynamic contrast-enhanced (DCE) perfusion or T1 perfusion has been used in a few recent studies evaluating PCNSL patients.15,16 These studies have noted increased values of BBB permeability measurements and lower values of plasma volume (vp) in patients with PCNSL as compared to

those with GBM. DWI has also been found to be useful for differentiating PCNSL from other neoplastic lesions of the brain. Lymphoma is known to show low mean diffusivity (MD) within the solid part of the tumour3,15,17,18; however, this finding may also be seen in other cellular brain lesions, and prior administration of steroids can confound imaging findings. SWI shows higher susceptibility signals in GBM as compared to PCNSL.19 Important limitations of the previous imaging studies have been the inclusion of a very small number of PCNSL subjects. In addition, none of these studies have investigated the DCE-derived haemodynamic perfusion parameters for the study of lymphomas, which overcomes the potential limitations of DSC perfusion MRI and additionally gives information about BBB permeability. The aim of the present study was to evaluate the performance of various indices derived from advanced MRI methods. To the authors’ knowledge, no previous study has evaluated multiparametric MRI using DCE-derived haemodynamic indices for differentiating GBM from PCNSL.

Materials and methods A total of 100 (GBM¼70 and PCNSL¼30) adult immunecompetent treatment-naive patients (66 men, 34 women; age range 18e82 years; mean age¼ 51.8115.86 years) were included retrospectively in this study. As a part of the institution informed consent policy, informed consent was obtained before MRI. The ethics committee approved the retrospective analysis and waived the need for informed consent regarding this study from both participating institutions. MRI was performed using a 3 T MRI system (Philips HealthTech, The Netherlands) using 15/32-channel head-coils. Conventional imaging included T2-weighted (TR (repetition time)/TE (echo time)¼3000 ms/110 ms, number of excitations (NEX)¼1, section thickness¼3 mm, matrix¼328258). T1-weighted inversion recovery prepared turbo field echo (TFE) images (TR/TE¼7.8ms/3.6ms, NEX¼1, section thickness¼1mm, flip angle¼80 , acquisition matrix¼240222, field of view (FOV)¼ 240240 mm2, reconstructed matrix¼288288), fluid-attenuated inversion recovery (FLAIR) weighted [TR/TE/TI (Inversion time)]¼4500 ms/279 ms/1600 ms, NEX¼1, section thickness¼1 mm, matrix¼224224). Multi-parametric MRI including DWI, SWI, and contrast-enhanced T1-perfusion was performed on these patients. Following acquisition parameters were used for advanced imaging sequences: pre-contrast two-dimensional (2D) T1-weighted turbo spin echo (TSE; TR/TE¼360/10 ms), flip angle¼90o, acquisition matrix size 244244, section thickness¼6 mm; FOV

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

240240 mm2; sections¼12 and recon matrix size 256256; fast dual spin echo proton-density (PD)weighted and T2-weighted (TR/TE1/TE2¼3500ms/23.2ms/ 90ms), flip angle¼90o and acquisition matrix size 256256, section thickness¼ 6 mm; FOV 240240 mm2; sections¼12 and recon matrix size¼256256; and T1-perfusion (3D TFE): TR/TE¼4.4 ms/2.1 ms, flip angle¼10 , section thickness¼6mm, acquisition matrix size¼128128, temporalresolution¼3.9s, dynamics¼32, section thickness¼ 6 mm; FOV ¼240240 mm2; sections¼12 and recon matrix size¼256256. SWI (TR/TE¼ 31ms/5.6,11.8,18 and 24.2ms, section thickness¼1 mm over contiguous sections, flip angle¼17, matrix 384384, FOV¼240240 mm2), and DWI (TR/TE¼4,156ms/114ms, number of signal averages¼4, flip angle¼90 , section thickness¼5mm, acquisition matrix¼152121, and FOV¼230184 mm2, b-values¼ 0 and 1,000) were acquired for all the included subjects. At the fourth time point of the T1-perfusion data acquisition, 0.1 mmol/kg body weight of gadobenate dimeglumine (Multihance, Bracco, Italy) was administered intravenously at a rate of 3 ml/s, followed by a 30-ml saline flush.

Image processing and data analysis All image analysis was performed independently by two experienced neuroradiologists with more than 15 and 30 years of experience. The apparent diffusion coefficient (ADC) was quantified using a vendor-supplied tool (Intellispace Portal, Philips Medical systems, ver 9.0). T1-perfusion data were analysed using an in-house developed software tool.20,21 The pre-contrast T1 map was computed using a previously described method based on T1W, T2W, and PDW TSE images.20 Three independent TSE images (S1, S2, S3) result in a system of three nonlinear equations in T1, T2, and the product K.r (r is proton density, and K is a constant factor): Sj ¼ K:r:exp



TEj T2

    TRj  1  exp T1

[1]

where j¼ 1, 2, 3. Elimination of K. r and T2 from the above system of equations reduces it to a single nonlinear equation for T1, the solution of which results in T1 estimation. In this study S1, S2 and S3 are dual PD-T2W and T1W images respectively. After pre-contrast T1 estimation, voxel-wise signal intensityetime curves obtained from T1-perfusion data were converted to concentrationetime curves. These concentration curves were fitted to leakage tracer kinetic model (LTKM) model for the estimation of various parameters: volume transfer rate (Ktrans), fractional extracellular extravascular volume (ve), fractional plasma volume (vp), leakage rate (ltr). Parameter kep¼ Ktrans/ve was also computed.21 Arterial input function (AIF) required for the LTKM model was estimated for each patient automatically using the previously described methodology.22 Haemodynamic parameters (rCBV, cerebral blood flow [CBF]) were also estimated by the first-pass analysis of concentrationetime curves.20 The CBV map was also corrected for leakage (rCBV_corr) of contrast medium using a previously reported study.20 the LTKM model was used,

3

which is an extension of the generalised tracer kinetic model (GTKM) by incorporating an additional tissue uptake leakage compartment in the extracellular extravascular space, as it is more suited to short-duration DCE-MRI data.21

ITSS scoring on SWI ITSS refers to low signal intensity linear or dot-like foci of susceptibility seen within the tumour on SWI. The presence of ITSS was assessed in each tumour and the semiquantitative analysis method as proposed by Park et al. was used.23 The degree of ITSS has been divided into various grades, as described below: grade 0¼no ITSS; grade I¼1e5 dot like or fine linear ITSSs; grade 2¼ 6e10 dot like or fine linear ITSSs; and grade 3¼11 or more dot-like or fine linear ITSSs within the tumour. All the images/maps were registered before analysis. The CBV value was normalised by the mean of normal white matter CBV value to generate the normalised CBV_corr map.24 Multiple regions of interest (ROIs) of 10e30 mm2 were drawn on the normalised CBV_corr map on all sections displaying the maximal values of CBV to extract the highest CBV_corr values from the tumour, and the mean of all sections in each tumour was considered for analysis. Care was taken while placing the ROI to avoid placement over large vessels and veins seen on SWI and T2-weighted images. Similarly, values for other kinetic parameters were extracted from the co-registered maps and ROIs were placed in the regions showing the highest values and mean values form all sections were recorded accordingly. DWIderived ADC maps showing the regions with lowest values were used by placing the ROI and recording the values using similar methodology avoiding necrotic regions. Two experienced radiologists blinded to the final histopathology performed the ROI analysis and recorded values. In cases where there was discordance, the final value was recorded in consensus.

Statistical analysis Considering histopathology findings as the reference standard, to differentiate GBM and lymphoma in brain tumours based on perfusion parameters, the minimum sample size was calculated for ROC analysis under the following assumptions: the area under the curve (AUC) was 0.5 for null and 0.7 for alternate hypothesis at a 5% level of significance, 80% power for a two-tail test with a design effect of 1.25. Under these assumptions the minimum sample size required for lymphoma was 30 and for GBM it was 70. Normality of data was evaluated using the ShapiroeWilk test. Most of perfusion parameters being non-normally distributed, non-parametric tests were used for analysis. The ManneWhitney U-test was used for pairwise comparison between groups. Univariate logistic regression analysis was used to assess the diagnostic value of each imaging parameter. For each parameter, the ability of the logistic regression model to discriminate between GBM and lymphoma was evaluated by the AUC of the ROC analysis.

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

4

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

Multivariable logistic regression models including these imaging parameters were also used to determine the performance of ITSS. Final multivariable model using rCBV_ corr, ADC, ITSS, and kep were analysed to determine its ability to discriminate GBM from lymphoma. All the analyses were performed using Stata 12.0 software (Strata Corp, Lakeway Drive College Station, Texas, USA). A p-value of <0.05 indicated a statistically significant difference.

Results Out of a total 100 excised or biopsied tumour tissues, 70 were confirmed to be GBM at histopathology while the remaining 30 were found to be B-cell PCNSL. Six of the GBM cases were found to have the isocitrate dehydrogenase (IDH) mutant and the remaining cases were IDH wild-type. All patients found to have PCNSL were immunocompetent and did not have clinical or imaging evidence of malignancy elsewhere. All the tumours included in the analysis were in the supratentorial region, with GBM most commonly involving the frontal lobe (30) followed by temporal lobe (24). Lymphoma was most commonly seen in the basal ganglia and thalamic regions (9) followed by lobar location. A contrast-enhanced study showed heterogeneous enhancement in the majority of gliomas while 15 patients with PCNSL showed heterogeneous enhancement. Necrosis was visible in 61/70 patients with GBM while it was noted in 9/30 patients with PCNSL. In two patients, GBM was nonenhancing whereas all cases of PCNSL showed enhancement. The median and interquartile range of the rCBV values, permeability metrics, ADC, and ITSS grading of all tumours are summarised in Table 1. Significant differences in the rCBV_corr, ADC, ITSS, and kep values were seen in comparing the GBM with PCNSL (Fig 1). On comparing rCBV_corr, PCNSL was found to have significantly lower values (3.89) as compared to GBM (5.67; p¼0.029); however, considerable overlap in the perfusion values of both tumour types was noted. Haemodynamic/kinetic parameters derived from T1-perfusion MRI, ADC values derived Table 1 Descriptive statistics of multiparametric imaging parameters for glioblastoma and lymphoma. Variables

rCBV_corr ADC* Ktrans kep ve vp

ltr

ITSS

Glioblastoma

Lymphoma

Median IQR (Q1,Q3)

Median IQR (Q1,Q3)

5.67 729.09 0.22 4.61 0.06 0.01 0.02 3.00

3.89 600.81 0.22 3.08 0.09 0.02 0.04 2.00

4.19e6.75 536.74e857.46 0.11e0.44 3.19e6.36 0.03e0.12 0.00e0.02 0.02e0.04 2.00e3.00

2.50e5.54 519.83e712.96 0.10e0.36 1.56e3.71 0.03e0.14 0.00e0.04 0.00e0.08 1.00e2.25

p-Value

0.029 0.004 0.626 <0.001 0.053 0.278 0.068 <0.001

Units of Ktrans, kep, ltr are per minute, whereas ve and vp are unit-less. IQR, inter quartile range, rCBV_corr, relative cerebral blood volume with leakage correction, ADC, apparent diffusion coefficient (105 mm2/s), Ktrans, transfer coefficient, kep, back flux exchange rate ve, fractional extracellular extravascular volume, vp, fractional plasma volume and the ltr, leakage rate, ITSS, intratumoural susceptibility signal intensity.

from DWI, and ITSS scores measured on SWI images are shown as box plots (Fig 2). The AUC for rCBV_corr was found to be 0.68 (95% CI: 0.55e0.80), which correctly classified 69 lesions with sensitivity and specificity of 75.7 and 53.3, respectively (Table 2, Fig 3). Regarding ADC, lower values of ADC were noted in cases of PCNSL (600.81105 versus 729.09105 mm2/s; p¼0.004), but again substantial overlap in the values was seen. The AUC for ADC was found to be 0.63 (95% CI: 0.52e0.74), which correctly classified 60 tumours with sensitivity and specificity of 70% and 36.7%. kep values were found to be significantly higher in GBM patients (4.61) as compared to PCNSL (3.08; p<0.001). The AUC for kep was found to be 0.74 (95% CI: 0.63e0.85), which correctly predicted histology in 70 patients. ITSS values found to be higher in cases of GBM (3.00) with an AUC for ITSS of 0.80 (95% CI: 0.70e0.89), which correctly predicted histology in 75 patients with sensitivity and specificity of 74.3 and 76.7, respectively (Table 2). Multivariable logistic regression analysis combined various imaging parameters to predict the correct histology. rCBV_corr, ADC, kep, and ITSS scores showed the highest AUC and were best discriminators of GBM and Lymphoma. Hence, combinations of ITSS with diffusion and perfusion parameters were assessed to predict the histological nature of the tumours. In addition, the discriminative ability of all the four parameters combined was also assessed. Combination ITSS with r-CBV_corr, kep, and ADC significantly increased the differentiation of PCNLS and GBM in comparison with a univariate approach (p<0.01). Combining all four parameters resulted in further improvement with an AUC of 0.92 (95% CI: 0.87e0.97) and correct prediction of 84 cases with sensitivity and specificity of 90 and 70 respectively (Table 2).

Discussion The present study investigated the possible imaging measurements that may discriminate between GBM and PCNSL by pooling various indices derived from T1-perfusion MRI, DWI, and SWI. ITSS derived from SWI was found to have the best classification accuracy followed by DCEderived kep, rCBV_corr, and ADC for discriminating GBM from PCNSL. Combining the parameters derived from T1perfusion MRI and DWI with SWI improved the classification accuracy as well as showed higher AUC for differentiating the two tumours. The present results show that combining multiparametric MRI is useful for discrimination of GBM and PCNSL. Conventional imaging allows differentiation between these two diseases as GBM usually appears as a ringshaped contrast-enhanced mass lesion with central necrosis on contrast-enhanced T1W MRI, whereas PCNSL typically presents as a solid mass lesion with homogeneous contrast enhancement in immunocompetent patients4,5; however, this pattern is not reliable in some clinical scenarios because atypical, solid, enhancing GBM without visible necrosis may mimic PCNSL, and atypical PCL with evident necrosis may mimic GBM25 In these

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

5

Figure 1 (Iaei) Left frontal lobe glioblastoma in a 77-year-old woman showing hyperintense mass on T2W/FLAIR images with perilesional oedema (Iaeb), the lesion shows no ITSS on SWI images (Ic), and on the ADC map (Id). The lesion shows minimum ADC (550105 mm2/s) value. Post-contrast T1W image shows heterogeneous enhancement of the mass lesion (Ie), T1 perfusion MRI shows a low rCBV_corr value within the tumour except for small foci of increased perfusion (3.2, If). Kinetic parametric maps are showing increased permeability within the solid part of the tumour: Ktrans (Ig), kep (Ih), and ve (Ii). (IIaei) A 67-year-old man with right frontal lobe GBM appearing as a well-demarcated, heterogeneous, lesion with a hyperintense signal on T2W/FLAIR images with mild perilesional oedema (IIaeb). SWI shows few foci of susceptibility within the tumour (ITSS 3, IIc). The lesion shows low ADC (541105mm2/s) within the solid part of the tumour (IId) and a large necrotic core with heterogeneous enhancement (IIe). The rCBV_corr map shows increased perfusion (5.9) in the solid peripheral part of the tumour (IIf). Permeability maps show areas of BBB disruption within the tumour: Ktrans (IIg), kep (IIh), and ve (IIi). (IIIaei) A large, solid, T2W/FLAIR (IIIaeb) mildly hyperintense mass lesion in a 70-year-old male patient with lymphoma seen growing along the ventricular margin with involvement of the corpus callosum, showing no foci of susceptibility within the tumour on SWI (IIIc) and appearing hypointense on ADC map (529105 mm2/ s, IIId). Profuse perilesional oedema is noted around the tumour mass. The contrast-enhanced T1W image shows homogeneous enhancement of the mass lesion (IIIe). Areas of increased rCBV_corr value (3.3) are seen on the CBV map. Kinetic maps show areas of elevated permeability indices within the tumour: Ktrans (IIIg), kep (IIIh), and ve (IIIi). (IVaei) A 37-year-old male patient with lymphoma shows a T2/FLAIR (IVa/IVb) hyperintense, large, periventricular mass lesion that enhances on the post-contrast study. Significant perilesional oedema and few scattered foci of susceptibility are seen within the lesion on SWI (ITSS 3, IVc). On the ADC map, the solid enhancing (IIe) part of the tumour shows low ADC (573105 mm2/s, IVd). Marked elevation of rCBV_corr ((IVf) values is seen within the tumour (5.8). Kinetic parametric maps, Ktrans (IVg), kep (IVh), and ve (IVi), show elevated values within the tumour.

situations, advanced MRI techniques may be useful for discriminating the two conditions. DWI has been extensively studied for differentiating GBM from PCNSL. In multiple studies, it has been found that PCNSL has lower ADC values when compared to the GBM.18,25e27 In few other studies, significant overlap in ADC values of GBM and PCNSL were reported.28 Hence, many studies have tried using multiparametric MRI or combining MRI features with other techniques to accurately classify these tumours.15,25,29 In the present study, mean values of ADC reported for lymphoma and GBM agree with previous studies, and PCNSL lesions had lower ADC values as compared to the GBM; however, significant overlap in the ADC values resulted in lower classification accuracy. Few previous studies have also reported less accuracy of diffusion for discriminating PCNSL and GBM. In the study of Suh et al., atypical PCNSL lesions were studied, and the authors reported an AUC of 70% with a sensitivity of 68% and

specificity of 82%.30 Makino et al. reported no significant differences in the ADC values of PCNSL and GBM.31 All the patients included in the current study were immunocompetent, contrary to some of the previous studies.32 Use of the histogram method might overcome this limitation, and may help in the better prediction of the tumour type as shown in the recent study of Choi et al.33 Various studies have shown the utility of DSC perfusion MRI for discrimination of GBM and PCNSL. Previous studies have consistently shown that GBM has higher rCBV values as compared to those of PCNSL.6e9,15,26,29,32,34 The haemodynamic parameters, CBV and CBF, have been shown to show significant correlation with tumour,neoangiogenesis which is an important biological process often seen in the malignant neoplasms of the brain. It has been shown to be mediated by vascular endothelial growth factor (VEGF) in gliomas, and its expression has been correlated with the perfusion measurements.35,36 however, similar studies have

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

6

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

Figure 2 Box plots summarising the values of various haemodynamic/kinetic parameters derived from T1-perfusion MRI, ADC values derived from DWI, and ITSS scores measured on SWI images.

not been performed for PCNSL. Perfusion studies using DSC perfusion have shown low rCBV in these tumours. Previous DCE studies have focused on the evaluation of the permeability measurements.15,16,33,34,37 In the current study, statistically significant differences in the rCBV_corr measured using T1-perfusion MRI were found; GBM showing higher CBV as compared to the lymphomas and these findings are consistent with the previously published literature. Perfusion measurements of PCNSL are lower as compared to

gliomas because of lower vascularity in PCNSL lesions as compared to GBM; however, it was noted that PCNSL rCBV_corr values showed overlap with GBM values and closely mimicked high-grade gliomas on perfusion MRI. Immunohistochemical studies in lymphomas have shown that VEGF expression can be significantly high in a subset of lymphoma patients and it has been associated with poor prognosis.12 In one study, VEGF expression and BBB permeability were investigated in patients with PCNSL.

Table 2 Classification of glioblastoma and lymphoma using ROC and logistic regression models with double and multiple imaging parameters. Models with Cut-off Correctly Sensitivity Specificity AUC 95% CI of one parameter classified (%) (%) AUC (%)

Models with 2 or more parameters

Correctly Sensitivity Specificity AUC 95% CI of classified (%) (%) AUC (%)

rCBV_corr ADC Ktrans kep ve vp

rCBV_corr and ITSS ADC and ITSS Ktrans and ITSS kep and ITSS ve and ITSS vp and ITSS ltr and ITSS Model with multi parameters: rCBV_corrr, ADC, kep, ITSS#

82 77 81 80 80 81 80 84

ltr

ITSS

4.245 566.12 0.123 3.195 0.034 0.007 0.018 3.0

69 60 58 70 57 60 61 75

75.7 70.0 70.0 75.7 70.0 71.4 74.3 74.3

53.3 36.7 30.0 56.7 26.7 33.3 30.0 76.7

0.68 0.63 0.55 0.75 0.42 0.44 0.45 0.80

0.55e0.80 0.52e0.74 0.43e0.67 0.64e0.85 0.30e0.55 0.32e0.57 0.30e0.59 0.70e0.89

91.4 88.6 97.1 91.4 95.7 97.1 91.4 90.0

60.0 50.0 43.3 53.3 43.3 43.3 53.3 70.0

0.83 0.86 0.81 0.86 0.84 0.78 0.82 0.92

0.73e0.93 0.79e0.94 0.71e0.91 0.77e0.94 0.76e0.92 0.67e0.90 0.71e0.92 0.87e0.98

A cut-off for classification probability was taken as 0.50. kep, Ktrans, ve and leakage adjusted. Numbers in parenthesis are percentages. GBM correctly classified percent shows sensitivity, and lymphoma correctly classified percent shows sensitivity of parameter/model. Units of Ktrans, kep, ltr are per minute, whereas ve and vp are unit-less. rCBV_corr, relative cerebral blood volume with leakage correction, ADC, apparent diffusion coefficient, Ktrans, transfer coefficient, kep, back flux exchange rate ve, fractional extracellular extra vascular volume, vp, fractional plasma volume and the ltr, leakage rate, ITSS, Intratumoural susceptibility signal intensity.

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

7

Figure 3 ROC curves demonstrate the discriminating performance of T1-derived haemodynamic and kinetic parameters, ADC, and ITSS for differentiating PCNSL and GBM.

Authors reported high VEGF expression in at least 50% of the patients and these patients also showed markedly higher mean vessel density. They also noted BBB disruption in those with high VEGF expression while it was relatively preserved in those with lower VEGF.12 These results indicate that some of the patients with PCNSL have significant angiogenesis and these subsets of patients with higher angiogenesis are expected to have higher perfusion

measurements. These histological studies support our observation that lymphoma patients may have high rCBV. DSC perfusion MRI might be underestimating the CBV values due to significant BBB disruption causing contrast medium leakage. This finding is supported by the higher accuracy of uncorrected DSC-derived CBV then leakagecorrected CBV for discriminating the PCNSL from GBM.26 In future studies, correlation of VEGF expression and

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

8

J. Saini et al. / Clinical Radiology xxx (2018) 1e9

haemodynamic measurements may be done to look for the relationship between these parameters. Multiple studies have evaluated permeability indices with Kickingereder et al. showing higher Ktrans and kep in PCNSL as compared to GBM15 while Abe et al. showed higher ve in PCNSL and the rest of the parameters were found to have overlapping values,38 whereas, Lin et al. found no significant difference in the Ktrans values, while vp was found to be higher for GBM as compared to the lymphoma.16 In the present study, higher kep values were found in GBM, whereas other parameters were not significantly different between the tumours. kep is the ratio of Ktrans and ve, and higher kep in GBM may be the result of either higher Ktrans, lower ve, or a combination of the two. Lower median ve values were noted in GBM, which were not statistically significant. Overall PCNSL is believed to have higher permeability compared to GBM, and possible reasons are the presence of relatively preserved vasculature in gliomas and perivascular infiltration of tumour cells in PCNSL.15 Tumour permeability, measured as Ktrans, has been shown to be higher in PCNSL in multiple older studies15,34; however; no significant difference in the Ktrans values were observed in the present study. The possible reason for this may be the differences in DCE acquisition and post-processing methods used in the current study. Other reasons may be the ROI-based method used for recording the quantitative values and the use of the histogram-based method might show differences as was noted by Murayama et al.34 in their study. SWI-derived ITSS scoring has been used for grading of gliomas and differentiation of gliomas from PCNSL.25,39e41 In this method, foci of susceptibility are counted and, based on the number of susceptibility foci lesions, are graded from 0 to 3. The ITSS score was able to discriminate 80% of the tumours correctly and ROC analysis showed an AUC of 79%. In the initial studies of Kim et al. and Peters et al., GBM could be discriminated from PCNSL with high sensitivity and specificity.39,41 Radbruch et al. also drew similar conclusions and found SWI useful for discrimination of PCNSL and lymphoma.40 Lee et al. noted haemorrhage in approximately 18% patients of lymphoma, and they correlated this finding along with the presence of necrosis with presence of underlying EpsteineBarr virus (EBV) positivity of PCNSL tumours.6 In the recent study by Kickingerder et al., haemorrhage or susceptibility within the tumour was noted in one-third of patients and all these were negative for EBV.25 Isolated reports of frankly haemorrhagic PCNSL exist in literature and some of this strong immunoreactivity for VEGF was noted.42,43 T-cell PCNSL are rare, and in one of the previous reports, it was noted that they are more often associated with haemorrhage44; however, in the present study B-cell PCNSL was also associated with significant susceptibility and this observation is in line with the observations of Kickingerder et al. The present findings suggest that higher ITSS scores in PCNSL are possible and make their discrimination from GBM difficult. Tests for EBV was not undertaken in the present cohort because previous studies from India have shown no association between EBV and PCNSL in immune-competent patients45,46; however, it may be considered a limitation of the present study.

Combining parameters derived from DCE perfusion, DWI, and SWI improved the diagnostic accuracy. The AUC of discriminant function was better than either of the parameters alone, and it showed best sensitivity and specificity for discriminating PCNSL and GBM. Similar results were reported by Kickingerder et al., who used a multivariable regression model and found that combination of rCBV, ADC, and ITSS could significantly improve the diagnostic accuracy for classification of these lesions.25 In conclusion, T1-perfusion-derived CBV and ADC values show substantial overlap between GBM and PCNSL and appear inferior to the ITSS scores derived from SWI for discriminating between them. Combining ITSS scores with CBV and ADC improves the diagnostic accuracy of MRI for discriminating GBM and PCNSL. Multiparametric analysis improves the discrimination of these two different tumour types more than individual imaging techniques such as perfusion, DWI, and SWI.

References 1. Alifieris C, Trafalis DT. Glioblastoma multiforme: pathogenesis and treatment. Pharmacol Ther 2015;152:63e82. https://doi.org/10.1016/ j.pharmthera.2015.05.005. 2. Hoang-Xuan K, Bessell E, Bromberg J, et al. Diagnosis and treatment of primary CNS lymphoma in immunocompetent patients: guidelines from the European Association for Neuro-Oncology. Lancet Oncol 2015;16(7):e322e32. 3. Haldorsen IS, Espeland A, Larsson E-M. Central nervous system lymphoma: characteristic findings on traditional and advanced imaging. AJNR Am J Neuroradiol 2011;32(6):984e92. https://doi.org/10.3174/ajnr.A2171. 4. Malikova H, Koubska E, Weichet J, et al. Can morphological MRI differentiate between primary central nervous system lymphoma and glioblastoma? Cancer Imaging Off Publ Int Cancer Imaging Soc 2016;16(1):40. https://doi.org/10.1186/s40644-016-0098-9. nez de la Pen ~ a MDM, Vicente LG, Alonso RC, et al. The multiple faces 5. Jime of nervous system lymphoma. atypical magnetic resonance imaging features and contribution of the advanced imaging. Curr Probl Diagn Radiol 2017;46(2):136e45. https://doi.org/10.1067/j.cpradiol.2016. 04.004. 6. Lee IH, Kim ST, Kim H-J, et al. Analysis of perfusion weighted image of CNS lymphoma. Eur J Radiol 2010;76(1):48e51. https://doi.org/10.1016/ j.ejrad.2009.05.013. 7. Nakajima S, Okada T, Yamamoto A, et al. Differentiation between primary central nervous system lymphoma and glioblastoma: a comparative study of parameters derived from dynamic susceptibility contrastenhanced perfusion-weighted MRI. Clin Radiol 2015;70(12):1393e9. https://doi.org/10.1016/j.crad.2015.08.004. 8. Liao W, Liu Y, Wang X, et al. Differentiation of primary central nervous system lymphoma and high-grade glioma with dynamic susceptibility contrast-enhanced perfusion magnetic resonance imaging. 1987 Acta Radiol Stockh Swed 2009;50(2):217e25. https://doi.org/10.1080/ 02841850802616752. 9. Xing Z, You RX, Li J, et al. Differentiation of primary central nervous system lymphomas from high-grade gliomas by rCBV and percentage of signal intensity recovery derived from dynamic susceptibility-weighted contrast-enhanced perfusion MRI. Clin Neuroradiol 2014;24(4):329e36. https://doi.org/10.1007/s00062-013-0255-5. €hr O, et al. MR perfusion in and around the 10. Blasel S, Jurcoane A, Ba contrast-enhancement of primary CNS lymphomas. J Neurooncol 2013;114(1):127e34. https://doi.org/10.1007/s11060-013-1161-z. 11. Nakajima S, Okada T, Yamamoto A, et al. Primary central nervous system lymphoma and glioblastoma: differentiation using dynamic susceptibility-contrast perfusion-weighted imaging, diffusion-weighted imaging, and (18)F-fluorodeoxyglucose positron emission tomography. Clin Imaging 2015;39(3):390e5. https://doi.org/10.1016/j.clinimag.2014. 12.002.

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107

J. Saini et al. / Clinical Radiology xxx (2018) 1e9 12. Takeuchi H, Matsuda K, Kitai R, et al. Angiogenesis in primary central nervous system lymphoma (PCNSL). J Neurooncol 2007;84(2):141e5. https://doi.org/10.1007/s11060-007-9363-x. 13. Boxerman JL, Schmainda KM, Weisskoff RM. Relative cerebral blood volume maps corrected for contrast agent extravasation significantly correlate with glioma tumour grade, whereas uncorrected maps do not. AJNR Am J Neuroradiol 2006;27(4):859e67. 14. Zaharchuk G. Theoretical basis of hemodynamic MRI techniques to measure cerebral blood volume, cerebral blood flow, and permeability. AJNR Am J Neuroradiol 2007;28(10):1850e8. https://doi.org/10.3174/ ajnr.A0831. 15. Kickingereder P, Sahm F, Wiestler B, et al. Evaluation of microvascular permeability with dynamic contrast-enhanced MRI for the differentiation of primary CNS lymphoma and glioblastoma: radiologice pathologic correlation. AJNR Am J Neuroradiol 2014;35(8):1503e8. https://doi.org/10.3174/ajnr.A3915. 16. Lin X, Lee M, Buck O, et al. Diagnostic accuracy of T1-weighted dynamic contrast-enhanced-MRI and DWI-ADC for differentiation of glioblastoma and primary CNS lymphoma. AJNR Am J Neuroradiol 2017;38(3): 485e91. https://doi.org/10.3174/ajnr.A5023. 17. Zacharia TT, Law M, Naidich TP, et al. Central nervous system lymphoma characterisation by diffusion-weighted imaging and MR spectroscopy. J Neuroimaging Off J Am Soc Neuroimaging 2008;18(4):411e7. https:// doi.org/10.1111/j.1552-6569.2007.00231.x. 18. Guo AC, Cummings TJ, Dash RC, et al. Lymphomas and high-grade astrocytomas: comparison of water diffusibility and histologic characteristics. Radiology 2002;224(1):177e83. https://doi.org/10.1148/radiol. 2241010637. € pf V, Preusser M, et al. Non-invasive assessment of 19. Furtner J, Scho intratumoural vascularity using arterial spin labeling: a comparison to susceptibility-weighted imaging for the differentiation of primary cerebral lymphoma and glioblastoma. Eur J Radiol 2014;83(5):806e10. https://doi.org/10.1016/j.ejrad.2014.01.017. 20. Singh A, Haris M, Rathore D, Purwar A, Sarma M, Bayu G, et al. Quantification of physiological and hemodynamic indices using T(1) dynamic contrast-enhanced MRI in intracranial mass lesions. J Magn Reson Imag 2007;26(4):871e80. 21. 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 contrastenhanced (DCE) MRI. J Magn Reson Imaging JMRI 2013;38(3):677e88. https://doi.org/10.1002/jmri.24021. 22. Singh A, Rathore RK, 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 Imag 2009;29(1):166e76. 23. Park MJ, Kim HS, Jahng G-H, et al. Semiquantitative assessment of intratumoural susceptibility signals using non-contrast-enhanced highfield high-resolution susceptibility-weighted imaging in patients with gliomas: comparison with MR perfusion imaging. AJNR Am J Neuroradiol 2009;30(7):1402e8. https://doi.org/10.3174/ajnr.A1593. 24. Sahoo P, Gupta RK, Gupta PK, et al. Diagnostic accuracy of automatic normalization of CBV in glioma grading using T1- weighted DCE-MRI. Magn Reson Imaging 2017;44:32e7. https://doi.org/10.1016/j.mri.2017.08.003. 25. Kickingereder P, Wiestler B, Sahm F, et al. Primary central nervous system lymphoma and atypical glioblastoma: multiparametric differentiation by using diffusion-, perfusion-, and susceptibility-weighted MRI. Radiology 2014;272(3):843e50. https://doi.org/10.1148/radiol.14132740. 26. Toh CH, Wei K-C, Chang C-N, et al. Differentiation of primary central nervous system lymphomas and glioblastomas: comparisons of diagnostic performance of dynamic susceptibility contrast-enhanced perfusion MRI without and with contrast-leakage correction. AJNR Am J Neuroradiol 2013;34(6):1145e9. https://doi.org/10.3174/ajnr.A3383. €gele T, et al. Water diffusivity: comparison of 27. Horger M, Fenchel M, Na primary CNS lymphoma and astrocytic tumour infiltrating the corpus callosum. AJR Am J Roentgenol 2009;193(5):1384e7. https://doi.org/ 10.2214/AJR.09.2486. 28. Server A, Kulle B, Maehlen J, et al. Quantitative apparent diffusion coefficients in the characterization of brain tumours and associated peritumoural oedema. 1987 Acta Radiol Stockh Swed 2009;50(6):682e9. https://doi.org/10.1080/02841850902933123.

9

29. Yamashita K, Yoshiura T, Hiwatashi A, et al. Differentiating primary CNS lymphoma from glioblastoma multiforme: assessment using arterial spin labeling, diffusion-weighted imaging, and 18F-fluorodeoxyglucose positron emission tomography. Neuroradiology 2013;55(2):135e43. https://doi.org/10.1007/s00234-012-1089-6. 30. Suh CH, Kim HS, Lee SS, et al. Atypical imaging features of primary central nervous system lymphoma that mimics glioblastoma: utility of intravoxel incoherent motion mri. Radiology 2014;272(2):504e13. https://doi.org/10.1148/radiol.14131895. 31. Makino K, Hirai T, Nakamura H, et al. Does adding FDG-PET to MRI improve the differentiation between primary cerebral lymphoma and glioblastoma? Observer performance study. Ann Nucl Med 2011;25(6):432e8. https://doi.org/10.1007/s12149-011-0483-1. 32. Wang S, Kim S, Chawla S, et al. Differentiation between glioblastomas, solitary brain metastases, and primary cerebral lymphomas using diffusion tensor and dynamic susceptibility contrast-enhanced MRI. AJNR Am J Neuroradiol 2011;32(3):507e14. https://doi.org/10.3174/ajnr.A2333. 33. Choi YS, Lee H-J, Ahn SS, et al. Primary central nervous system lymphoma and atypical glioblastoma: differentiation using the initial area under the curve derived from dynamic contrast-enhanced MR and the apparent diffusion coefficient. Eur Radiol 2017;27(4):1344e51. https:// doi.org/10.1007/s00330-016-4484-2. 34. Murayama K, Nishiyama Y, Hirose Y, et al. Differentiating between central nervous system lymphoma and high-grade glioma using dynamic susceptibility contrast and dynamic contrast-enhanced MRI with histogram analysis. Magn Reson Med Sci MRMS Off J Jpn Soc Magn Reson Med 2017, https://doi.org/10.2463/mrms.mp.2016-0113. 35. Barajas RF, Phillips JJ, Vandenberg SR, et al. Pro-angiogenic cellular and genomic expression patterns within glioblastoma influences dynamic susceptibility weighted perfusion MRI. Clin Radiol 2015;70(10): 1087e95. https://doi.org/10.1016/j.crad.2015.03.006. 36. Awasthi R, Rathore RKS, Soni P, et al. Discriminant analysis to classify glioma grading using dynamic contrast-enhanced MRI and immunohistochemical markers. Neuroradiology 2012;54(3):205e13. https:// doi.org/10.1007/s00234-011-0874-y. 37. Lu S, Wang S, Gao Q, et al. Quantitative evaluation of diffusion and dynamic contrast-enhanced magnetic resonance imaging for differentiation between primary central nervous system lymphoma and glioblastoma. J Comput Assist Tomogr 2017;41(6):898e903. https:// doi.org/10.1097/RCT.0000000000000622. 38. Abe T, Mizobuchi Y, Sako W, et al. Clinical significance of discrepancy between arterial spin labeling images and contrast-enhanced images in the diagnosis of brain tumours. Magn Reson Med Sci MRMS Off J Jpn Soc Magn Reson Med 2015;14(4):313e9. https://doi.org/10.2463/mrms.2014-0083. 39. Kim HS, Jahng G-H, Ryu CW, et al. Added value and diagnostic performance of intratumoural susceptibility signals in the differential diagnosis of solitary enhancing brain lesions: preliminary study. AJNR Am J Neuroradiol 2009;30(8):1574e9. https://doi.org/10.3174/ajnr.A1635. 40. Radbruch A, Wiestler B, Kramp L, et al. Differentiation of glioblastoma and primary CNS lymphomas using susceptibility weighted imaging. Eur J Radiol 2013;82(3):552e6. https://doi.org/10.1016/j.ejrad.2012.11.002. € ß N, Wodarg F, et al. Glioblastomas versus lymphomas: 41. Peters S, Kno more diagnostic certainty by using susceptibility-weighted imaging (SWI). ROFO Fortschr Geb Rontgenstr Nuklearmed 2012;184(8):713e8. https://doi.org/10.1055/s-0032-1312862. 42. Kim I-Y, Jung S, Jung T-Y, et al. Primary central nervous system lymphoma presenting as an acute massive intracerebral hemorrhage: case report with immunohistochemical study. Surg Neurol 2008;70(3):308e11. https://doi.org/10.1016/j.surneu.2007.04.009. 43. Rubenstein J, Fischbein N, Aldape K, et al. Hemorrhage and VEGF expression in a case of primary CNS lymphoma. J Neurooncol 2002;58(1):53e6. 44. Kim EY, Kim SS. Magnetic resonance findings of primary central nervous system T-cell lymphoma in immunocompetent patients. 1987 Acta Radiol Stockh Swed 2005;46(2):187e92. 45. Tandon A, Challa S, Shanmugam M, et al. EpsteineBarr virus as a possible etiologic agent in primary central nervous system lymphoma in immunocompetent individuals. Neurol India 2009;57(1):36e40. 46. Sharma MC, Gupta RK, Kaushal S, et al. A clinicopathological study of primary central nervous system lymphomas & their association with EpsteineBarr virus. Indian J Med Res 2016;143(5):605e15. https:// doi.org/10.4103/0971-5916.187109.

Please cite this article in press as: Saini J, et al., Multiparametric imaging-based differentiation of lymphoma and glioblastoma: using T1perfusion, diffusion, and susceptibility-weighted MRI, Clinical Radiology (2018), https://doi.org/10.1016/j.crad.2018.07.107