Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Apparent diffusion coefficient for molecular subtyping of non-gadolinium-enhancing WHO grade II/III glioma: volumetric segmentation versus two-dimensional region of interest analysis

Apparent diffusion coefficient for molecular subtyping of non-gadolinium-enhancing WHO grade... Objectives To investigate if quantitative apparent diffusion coefficient (ADC) measurements can predict genetic subtypes of non- gadolinium-enhancing gliomas, comparing whole tumour against single slice analysis. Methods Volumetric T2-derived masks of 44 gliomas were co-registered to ADC maps with ADC mean (ADC )calculated. mean For the slice analysis, two observers placed regions of interest in the largest tumour cross-section. The ratio (ADC )between ratio ADC in the tumour and normal appearing white matter was calculated for both methods. mean Results Isocitrate dehydrogenase (IDH) wild-type gliomas showed the lowest ADC values throughout (p < 0.001). ADC in mean the IDH-mutant 1p19q intact group was significantly higher than in the IDH-mutant 1p19q co-deleted group (p <0.01). A −6 2 volumetric ADC threshold of 1201 × 10 mm /s identified IDH wild-type with a sensitivity of 83% and a specificity of 86%; mean a volumetric ADC cut-off value of 1.65 provided a sensitivity of 80% and a specificity of 92% (area under the curve (AUC) ratio 0.9–0.94). A slice ADC threshold for observer 1 (observer 2) of 1.76 (1.83) provided a sensitivity of 80% (86%), specificity of ratio 91% (100%) and AUC of 0.95 (0.96). The intraclass correlation coefficient was excellent (0.98). Conclusions ADC measurements can support the distinction of glioma subtypes. Volumetric and two-dimensional measurements yielded similar results in this study. Key Points � Diffusion-weighted MRI aids the identification of non-gadolinium-enhancing malignant gliomas � ADC measurements may permit non-gadolinium-enhancing glioma molecular subtyping � IDH wild-type gliomas have lower ADC values than IDH-mutant tumours � Single cross-section and volumetric ADC measurements yielded comparable results in this study . . . . Keywords Brain Diffusion magnetic resonance imaging Isocitrate dehydrogenase Glioma Neuroimaging Abbreviations DKI Diffusion kurtosis imaging ADC Apparent diffusion coefficient DTI Diffusion tensor imaging AUC Area under the curve DWI Diffusion-weighted imaging CS Centrum semiovale IDH Isocitrate dehydrogenase * S. C. Thust Lysholm Department of Neuroradiology, National Hospital for [email protected] Neurology and Neurosurgery, Queen Square, London WC1N 3BG, UK 1 Department of Clinical Neurology, National Hospital for Neurology Neuroradiological Academic Unit, Department of Brain Repair and and Neurosurgery, London, UK Rehabilitation, UCL Institute of Neurology, London, UK Department of Neurodegenerative Disease, UCL Institute of 2 Neurology and Division of Neuropathology, London, UK Imaging Department, University College London Foundation Hospital, London, UK Queen Square MS Centre. Department of Neuroinflammation, UCL Institute of Neurology, University College London, London, UK 3780 Eur Radiol (2018) 28:3779–3788 wt mut del IDH Isocitrate dehydrogenase wild-type IDH 1p19q belong to the oligodendroglioma group mut int wt IDH 1p19 Isocitrate dehydrogenase-mutant 1p19q [13]. IDH gliomas probably constitute a genetically hetero- intact geneous category of lesions, but often exhibit aggressive be- mut del IDH 1p19 Isocitrate dehydrogenase-mutant 1p19q co- haviour and have been suspected to represent early glioblas- deleted toma [14–17]. In the emerging literature on MR imaging fea- wt ICC Intraclass correlation coefficient tures of IDH glioma, initial lack of enhancement has been LGG Low grade glioma reported in some of these tumours [6, 18, 19]. NAWM Normal appearing white matter Diffusion-weighted imaging (DWI) is a technique of great PACS Picture archiving and communications interest in cancer, because water diffusivity is impaired in system highly cellular tissues, which reflects tumour proliferative rate ROC Receiver operating characteristic and aggressiveness [20]. The phenomenon of reduced diffu- ROI Centrum semiovale region of interest sion preceding fulminant radiological progression of pre- CS ROI Tumour region of interest sumed LGG has been observed prior to molecular typing tum wt TE Echo time [7], evoking later descriptions of IDH glioma serial imaging TR Repetition time findings [4]. Quantitative apparent diffusion coefficient VOI Centrum semiovale volume of interest (ADC) values have demonstrated high accuracy for glioma CS VOI Tumour volume of interest grading through meta-analysis [21]. For the non-invasive tum wt WHO World Health Organization identification of low to intermediate IDH glioma, diffusion 2HG D2-hydroxyglutarate tensor imaging (DTI) and diffusion kurtosis imaging (DKI) have shown potential, suggesting that reduced and heteroge- wt nous diffusivity are IDH features [22–24]. However, ad- Introduction vanced diffusion techniques are not universally available out- side academic hospital institutions, may require longer scan Gadolinium contrast uptake was previously considered the times and dedicated post-processing. best MR imaging predictor of glioma histological grade and Mean ADC measurement could be a rapid and practicable malignancy[1–3]. On the basis of this, it has been common approach to assess glioma diffusivity, being computationally practice to interpret non-enhancing intrinsic tumours as prob- non-demanding compared to histograms or texture analysis. able low grade gliomas (LGG) [4]. But conventional MRI has Although theoretically superior, there is no conclusive evi- proven to be unreliable in predicting subsequent tumour be- dence that whole lesion analysis outperforms region-of- haviour, whereby a proportion of presumed LGG may rapidly interest placement for the identification of malignant gliomas progress with development of malignant features such as en- [25]. hancement and necrosis [4–8]. The study presented sought to (i) investigate whether ADC The discovery of several key genetic alterations as princi- measurements from routine clinical DWI were associated with pal determinants of glioma prognosis has challenged the ref- glioma molecular subtype and (ii) to compare the performance erence standard of glioma grouping by histology [9]. of volumetric whole tumour ADC with single slice ADC Mutations in isocitrate dehydrogenase (IDH) represent a com- measurements. mon (> 70%) defining event in the development of LGG, conversely more than 90% of glioblastomas belong to the IDH wild-type group [10, 11]. Despite its oncogenic effect Materials and methods through production of a toxic metabolite D2- hydroxyglutarate (2HG), the presence of an IDH mutation is Patients associated with a favourable prognosis. The revised 2016 World Health Organization (WHO) clas- Following institutional board approval for a retrospective sification of brain tumours for the first time incorporates mo- study, we searched the neuropathology records revealing 37 wt lecular data to augment the diagnosis [12]. For WHO grade II/ patients with WHO grade II/III IDH glioma between 2009 III gliomas, three molecular subgroups have been defined: and 2016. For comparison of the molecular groups, control wt IDH wild-type glioma (IDH ) with survival similar to that samples of IDH (IDH1-R132H) mutant gliomas (34 mut int mut del of glioblastoma, IDH-mutant glioma with intact 1p19q IDH 1p19q and 32 IDH 1p19q )wererandomlyse- mut int (IDH 1p19 ) and an intermediate prognosis, and IDH- lected. We sought to evaluate ADC for suspected LGG prior mut del mutant 1p19q co-deleted glioma (IDH 1p19q )with the to tissue diagnosis. To replicate the clinical situation, only best prognosis and greatest chemosensitivity [11]. There is gliomas without gadolinium enhancement were included (2 partial overlap with histomorphology, whereby many non-enhancing gliomas were excluded because of missing mut int IDH 1p19 are astrocytic and the majority of images and degraded DWI, respectively). The study sample Eur Radiol (2018) 28:3779–3788 3781 wt consisted of 14 IDH (7 WHO II and 7 WHO III), 16 90.5 [69.5, 137] ms; TR = 4000 [2837, 10,000] ms, in-plane mut int 2 IDH 1p19q (8 WHO II and 8 WHO III) and 14 resolution = 1.25 × 1.25 [0.5 × 0.5, 2.5 × 2.5] mm ; slice mut del IDH 1p19q (11 WHO II and 3 WHO III), amounting to thickness = 5 [4, 6] mm; gap between slices = 1.5 [0, 2] 44 non-enhancing gliomas for the three molecular groups mm. For each patient, the imaging study was performed on (patient selection diagram shown in Fig. 1). No haemorrhagic average (standard deviation, sd) 2.3 (2.8) months prior to the or necrotic gliomas were featured in the study. tissue diagnosis. Image examples for the glioma molecular subgroups are shown in Fig. 2. MRI acquisition Post-processing and ADC analysis Ours is a quaternary neurosurgical centre; therefore the stan- dard (structural and DWI) MRI sequences in this study orig- ADC map calculation inated from 10 different referring institutions (institution 1 to institution 10): 29 from our own institutions, 4 from institution In a spin echo diffusion-weighted sequence, the signal S [S = b b 2, 3 from institution 3, 2 from institution 4, and one each from (−b ADC) S e ] from each pixel in an image is formed of a first the remaining six institutions. The studies were acquired on 18 component (S ) dependent on tissue properties (i.e. ‘spin den- different scanners (31 at 1.5 Tesla, and 13 at 3 Tesla) from all sity’,T and T relaxation times) and sequence properties (e.g. 1 2 major vendors: four General Electric scanners [Discovery −b ADC repetition time, TR); and a second component (e )de- MR450 (number of patients n = 5), 2× Signa Excite (n =1 pendent on the diffusion gradients (b, in units of s/mm )and each), Genesis Signa (n = 2)], seven Siemens scanners [3× the apparent diffusion coefficient (ADC, in units of mm /s). Avanto (n =7, n =2, n =1), aTrio (n = 9), Symphony (n = The ADC is obtained by dividing the image acquired with- 4), Skyra (n =3), Espree(n = 1)], six Philips scanners [Ingenia out diffusion gradients (S =S ) by the image acquired with b =0 0 (n =2), 5× Achieva (n = 1 each)] and one Toshiba scanner (n = diffusion gradients (S ): 1). All acquisitions included axial T2-weighted images, and axial standard 3-directional whole brain DWI. The median ADC ¼ðÞ 1=b lnðÞ S =S ð1Þ 0 b [min, max] values of the parameters of the T2-weighted im- In this division, the dependence of ADC from S (and ages were echo time (TE) = 99.5 [80, 141] ms; repetition time 0 therefore from T ,T and TR) is eliminated [26]. The ADC (TR) = 4610 [2500, 7480] ms, in-plane resolution = 0.5 × 0.5 1 2 maps were calculated using Eq. 1 and the utility fslmaths from [0.3 × 0.3, 0.9 × 0.9] mm ; slice thickness = 5 [1, 6] mm; gap the software library fsl (version 5.0) [27]. Offline whole tu- between slices = 1.5 [0, 2] mm. All DWI acquisitions included mour analysis and single slice analysis were subsequently diffusion gradient weighting values b = 0 s/mm and b =1000 performed. s/mm ; the median [min, max] of other parameters were TE = Fig. 1 Flow diagram of patients included and excluded from the analyses 3782 Eur Radiol (2018) 28:3779–3788 Fig. 2 WHO II/III molecular subgroup examples showing T2- weighted images, b1000, ADC maps and T1-weighted post gadolinium images of non- wt enhancing a IDH , b mut int IDH 1p19q and c mut del IDH 1p19q glioma Whole tumour (volumetric) ADC analysis Single slice ADC analysis Tumour volumes of interest (VOI ) were outlined by a Standard picture archiving and communication systems tum neuroradiology resident (S.H.) using ITK snap Toolbox (PACS) software (IMPAX 6.5.1.1008, Agfa-Gevaert, Mortsel, version 3.6 (www.itksnap.org [28]), covering the entire Belgium) was used to exploit tools routinely available for T2 signal abnormality with each segmentation optimised reporting of MR images. Two observers blinded to by a board-certified neuroradiologist specialised in brain histomolecular results (J.A.M. general radiology trainee = ob- tumour imaging (S.C.T.). For multicentric gliomas, the server 1 and S.C.T. = observer 2) located the tumour on the T2- total volume of signal abnormality was treated as one weighted sequence, selecting two round regions of interest on lesion. ADC maps were co-registered to T2 imaging using the ADC map viewed side-by-side: The first region of interest the FLIRT toolbox [29, 30]performingarigidbodytrans- (ROI ) was drawn in the largest lesion cross-section sparing tum formation with a six-parameter model and ‘Normalised the tumour margin to avoid partial volume effects. The second Mutual Information’ as cost function. Subsequently, round ROI aiming for a similar size to ROI was placed in CS tum ADC measurements were obtained for each tumour, contralateral centrum semiovale NAWM, taking care to exclude mean using the fslstats utility from fsl [25–27]. images with visible ventricular surfaces, cortex and/or sulcal To consider possible interindividual variations in brain spaces at measurement level. Three patients were excluded diffusivity, we assessed the ADC in normal appearing from the single slice analysis because of non-availability of an mean white matter (NAWM). For each patient, a standardised ADC map on PACS. The ratio between the ADC in the mean second volume of interest (VOI ) was drawn in the con- tumour and CS was calculated [PACS_ADC = CS ratio tralateral centrum semiovale (CS). This VOI was used ADC (ROI )/ADC (ROI )]. No absolute ADC CS mean tum mean CS to calculate the ADC =ADC (VOI )/ values were measured by the single slice method, as their work- ratio mean tum wt ADC (VOI )(Fig. 2). For two IDH tumours, the station display can vary depending on the referring institution. mean CS NAWM analysis was omitted because of bilateral tumour An example of the volumetric segmentation and single slice infiltration. ADC measurement is demonstrated in Fig. 3. Eur Radiol (2018) 28:3779–3788 3783 Fig. 3 Image examples demonstrating the whole lesion volumetric segmentation (mask overlaid on right frontal mut int IDH 1p19q glioma), single slice largest tumour cross-section ROI and comparative tum contralateral NAWM ROI CS placements Histopathology and molecular analysis Results wt Paraffin blocks containing tissue were analysed at our institu- The mean age was greater in the IDH group than in the mut mut int tion’s neuropathology department according to WHO 2016 IDH groups (p = 0.0001 for IDH 1p19q , p = 0.005 mut del guidance and previously published data [16]. IDH R132H for IDH 1p19q ). The larger proportion of WHO II glio- mut del immuno-negative tumours underwent multiple gene Sanger mas in the IDH 1p19q was not statistically significant wt sequencing. A quantitative polymerase chain reaction-based (Pearson chi-square test p = 0.115 for IDH and p =0.105 mut int copy number assay was used to determine 1p/19q status. for IDH 1p19q ). The patient demographic data and tu- mour volumes are reported in Table 1. Statistical analysis Association between molecular subtype and ADC For the volumetric and single slice data, the statistical analysis values consisted of two steps each: (i) linear regression to assess the wt mut int wt association between the tumour type (IDH ,IDH 1p19q , In the volumetric analysis, IDH tumours showed significantly mut del IDH 1p19q ) and ADC values, followed by (ii) logistic lower whole tumour volume ADC (VOI )than mean tum mut int mut del regression to determine if ADC values can differentiate IDH 1p19q (p < 0.0005) and IDH 1p19q (p =0.001). wt mut mut int IDH from IDH gliomas. A receiver operating character- The ADC (VOI )intheIDH 1p19q group was signif- mean tum mut del istic (ROC) analysis was used to quantify the performance of icantlyhigherthanintheIDH 1p19q group (p = 0.0047). wt the logistic regression. For the identification of a cut-off point IDH gliomas had a significantly lower whole tumour mut int for the logistic regression the ‘nearest to (0,1)’ method was ADC than IDH 1p19q (p < 0.0005) and ratio mut del performed. Statistical significance was set at 5%. The inter- IDH 1p19q (p =0.019). TheADC in the ratio mut int rater agreement was expressed as an intraclass correlation co- IDH 1p19q group was significantly higher than in the mut del efficient (ICC) using a two-way random effects model. All IDH 1p19q group (p =0.0054). statistical analyses were performed using Stata version 14 On single slice assessment, a significantly lower mean wt (College Station, TX: StataCorp LP). PACS_ADC was observed for IDH than for ratio Table 1 Patient demographic data and tumour volumes Whole tumour ADC (VOI ) mean tum Patient group Nr of patients Age in years Tumour volume CS NAWM volume Tumour volume for patients 3 3 total (male) (mean ± sd) (years) (mean ± sd) (cm ) (mean ± sd) (cm ) with bilateral infiltration (mean ± sd) (cm ) wt IDH 14 (9) 53 (± 14) 64 (± 68) (n = 12) 11.6 (± 2.5) (n = 12) 366 (± 46) (n =2) mut int IDH 1p19 16 (6) 33.9 (± 8.6) 60 (± 44) 10.9 (± 2.3) N/A mut del IDH 1p19 14 (7) 38.9 (± 8.3) 48 (± 50) 10.8 (± 2.5) N/A 3784 Eur Radiol (2018) 28:3779–3788 mut int mut int Table 3 F test for the difference between IDH 1p19 and IDH 1p19q (p <0.0005 observer 1; p <0.0005 observer mut del IDH 1p19 mut del 2) and for IDH 1p19q (p = 0.001 observer 1; p =0.001 mut int observer 2). The PACS_ADC in the IDH 1p19q group ratio Analysis type p mut del was higher than in the IDH 1p19q group (p =0.0008 for ADC (VOI)0.0047 mean tum observer 1 and p = 0.0025 for observer 2). No statistical asso- Whole tumour ADC 0.0054 ratio ciations were demonstrated between the NAWM ADC mean PACS_ADC 1st observer 0.0008 ratio values and molecular subtype. PACS_ADC 2nd observer 0.0025 ratio The intra-rater agreement for the PACS_ADC mea- ratio surements was very high: the correlation of measurements made on the same individual was 0.96, while the correla- Diagnostic performance of ADC values tion between mean observer ratings was 0.98. The correla- tion of measurements equaled the consistency agreement, For ADC (VOI ), a ROC analysis quantified the accuracy mean tum indicating no systematic difference between the two ob- of correctly classifying tumour type to an area under the curve servers. The single slice ADC values were slightly but ratio (AUC) of 0.94. The cut-off point for the ADC (VOI )was mean tum systematically higher than the volumetric ADC .The −6 2 ratio 1201 × 10 mm /s, with a sensitivity of 0.83 and a specificity numerical results of the association between tumour type of 0.86. For a decrease in the ADC (VOI ) value by 1.0 × mean tum and ADC values for the volumetric and single slice analy- −5 2 wt 10 mm /s, the odds of IDH increasedby78% (p = 0.003). ses are reported in Table 2.InTable 3, the difference be- For the volumetric ADC , the ROC analysis yielded an ratio mut int tween the ADC values in IDH 1p19q and in AUC of 0.90 with a sensitivity of 0.80 and a specificity of mut del IDH 1p19q is shown. In Table 4 the ICC values are 0.92 for a threshold ADC of 1.65. For a decrease in the ratio detailed. The boxplots of the ADC and ADC values wt mean ratio volumetric ADC value by 0.1, the odds of IDH increased ratio are depicted in Fig. 4. by 46% (p = 0.004). wt Table 2 Results of the linear regression between ADC and tumour type (IDH is the reference group) Whole tumour ADC (VOI ) mean tum Patient group ADC (VOI ) Regression 95% CI of the p mean tum mean (sd) coefficient regr. coeff. −6 2 −6 2 −6 2 (10 mm /s) (10 mm /s) (10 mm /s) wt IDH 1032 (168) 1032 922–1141 0.0005 mut int IDH 1p19 1543 (254) 511 361–661 0.0005 mut del IDH 1p19 1321 (162) 289 134–444 0.001 Whole tumour ADC ratio Patient group ADC Regression 95% CI of the p ratio mean (sd) coefficient regr. coeff. wt IDH 1.49 (0.32) 1.49 1.32–1.66 0.0005 mut int IDH 1p19 2.09 (0.34) 0.59 0.37–0.82 0.0005 mut del IDH 1p19 1.77 (0.20) 0.28 0.05–0.51 0.019 Single slice PACS_ADC first observer ratio Patient group PACS_ADC Regression 95% CI of the p ratio mean (sd) coefficient regr. coeff. wt IDH 1.50 (0.21) 1.50 1.33–1.68 0.0005 mut int IDH 1p19 2.37 (0.35) 0.87 0.63–1.10 0.0005 mut del IDH 1p19 1.96 (0.27) 0.45 0.20–0.70 0.001 Single slice PACS_ADC second observer ratio Patient group PACS_ADC Regression 95% CI of the p ratio mean (sd) coefficient regr. coeff. wt IDH 1.48 (0.19) 1.48 1.28–1.68 0.0005 mut int IDH 1p19 2.37 (0.38) 0.88 0.62–1.14 0.0005 mut del IDH 1p19 1.96 (0.36) 0.47 0.20–0.75 0.001 mut Regression coefficient represents the difference in the dependent variable (ADC) between each of the two IDH groups and the reference group wt (IDH ) Eur Radiol (2018) 28:3779–3788 3785 Table 4 Inter-rater agreement expressed as intraclass correlation Discussion coefficient (ICC) In this analysis, we observed that ADC values obtained from Correlation ICC Consistency ICC (95% CI) (95% CI) standard clinical DWI are a highly significant predictor of non-enhancing glioma IDH status and may permit non- Observer 1 vs observer 2 - PACS_ADC ratio invasive molecular subtyping in accordance with the 2016 Individual ICC 0.96 (0.92–0.98) 0.96 (0.92–0.98) WHO classification. Average ICC 0.98 (0.96–0.99) 0.98 (0.96–0.99) Two recent surveys highlighted clinical practices in caring Observer 1 PACS_ADC vs volumetric ADC ratio ratio for patients with presumed LGG, with approximately 50% of Individual ICC 0.80 (0.35–0.92) 0.87 (0.77–0.93) neurosurgeons adopting a ‘wait and see’ approach balanced Average ICC 0.89 (0.52–0.96) 0.93 (0.87–0.96) against surgical risk [31], and only 21% performing an upfront Observer 2 PACS_ADC vs volumetric ADC wt ratio ratio biopsy [32]. Consequently, innocuous appearing IDH glio- Individual ICC 0.79 (0.43–0.91) 0.85 (0.74–0.92) mas may reveal their aggressive nature through progression Average ICC 0.88 (0.60–0.95) 0.92 (0.85–0.96) and receive treatment with a delay. Low ADC values are associated with increased glioma cel- lularity and worse prognosis, supported by comparisons of dif- fusivity, histological specimens and clinical data in multiple A ROC analysis quantified the accuracy of the studies [5, 33–37]. Low diffusivity predicts poor astrocytoma survival independent from WHO grade [38], although no linear PACS_ADC logistic regression in correctly classifying tu- ratio mour type to an AUC of 0.96 for observer 1 and 0.95 for relation exists between ADC and glioma prognosis [39]. Past studies to distinguish astrocytoma and oligodendroglioma observer 2. The cut-off point for the PACS_ADC for ob- ratio using ADC values yielded variable success [40, 41],andinret- server 1 (observer 2) was 1.83 (1.76) with a sensitivity of 0.80 rospect may have been influenced by the incomplete overlap (0.86) and a specificity of 1.00 (0.91) at the cut-off point. For a between histological and molecular groups. Diagnostic focus decrease in the single slice ADC value by 0.1, the odds of ratio wt has shifted to genetic typing, yet immunohistochemistry tests IDH increased by 62% (p = 0.005) for observer 1 and 57% are complex and not infallible, requiring interpretation in in the (p = 0.004) for observer 2. The numerical results for glioma subtype prediction are reported in Table 5. The ROC curves context of morphological criteria and test type performed to avoid interpretational errors [42]. are depicted in Fig. 5. Fig. 4 Boxplot of the values of the a whole tumour ADC (VOI ), b whole mean tum tumour ADC , c single slice ratio PACS_ADC first observer and ratio d single slice PACS_ADC ratio first observer 3786 Eur Radiol (2018) 28:3779–3788 Table 5 Cut-off point estimation Method Cut-off point Sensitivity at cut-off point Specificity at cut-off point AUC at cut-off point ADC (VOI)1201 0.83 0.86 0.85 mean tum −6 2 (10 mm /s) Whole tumour ADC 1.65 0.80 0.92 0.86 ratio PACS_ADC 1st observer 1.83 0.86 1.00 0.93 ratio PACS_ADC 2nd observer 1.76 0.86 0.91 0.88 ratio Recently, Leu et al. were able to assign gliomas to the WHO type. The quicker and easier single slice analysis even performed 2016 molecular groups using ADC; however, their method dif- marginally better. This is in line with results of previous imaging fered from ours by including enhancing lesions and ADC median research, which suggested that whole lesion diffusivity measure- values derived from b700–1000 gradients with DTI analysed for ment is not always superior to ROI analysis [25, 44]. some patients [43]. To our best knowledge, this is the first IDH The ability of ADC to predict glioma subtypes and optimum typing study to focus on non-enhancing gliomas, using b1000 thresholds may be subject to ROI placement technique with pre- values derived from 3-directional DWI. This is particularly im- vious research focusing on minimum ADC value analysis: Xing portant, as such tumours are usually assumed to be less aggres- et al. showed a statistical correlation between ADC and IDH sive in common clinical practice. status using a multiple (≥ 5) ROI technique with the mean of We found ADC values to be closely reproducible when the lowest ADC measurement chosen as minimum ADC in con- ratio comparing whole lesion measurements against single slice region sensus [45]. In a similar fashion, a previous DTI study for IDH of interest placements, for which there was near complete inter- typing used multiple ROI placements and a two-reader consen- observer agreement. The similarity of our volumetric and single sus method to obtain minimum ADC values [24]. slice results could be explained by a relative homogeneity of As a reference ROI, we chose the centrum semiovale for its these non-enhancing, non-necrotic gliomas. Both the absolute potentially greater reproducibility compared to a ‘mirror’ ROI ADC values and ADC appear valuable for this lesion [45], because this could be influenced by tumour location. We mean ratio Fig. 5 ROC curves for the a whole tumour ADC (VOI ), b whole tumour ADC , c single slice PACS_ADC observer 1 and d observer 2 mean tum ratio ratio Eur Radiol (2018) 28:3779–3788 3787 Statistics and biometry One of the authors (C.T. MD PhD MSc Medical avoided the internal capsule [24], which is a smaller structure Statistics) has significant statistics expertise. and more difficult to locate by an untrained rater. Lee et al. found ADC mean and ADC histograms useful for Informed consent Written informed consent was waived by the institu- IDH typing of WHO grade III and IV gliomas [46]. However, tional review board. for glioblastoma IDH typing alone, a recent study identified no difference in ADC values [47]. In Tan et al.’s study of grade Ethical approval Institutional review board approval was obtained. II–IV gliomas, the accuracy of ADC for IDH typing decreased Methodology with higher grade, which may reflect greater lesion heteroge- � retrospective neity [24]. It is probable that in such circumstances advanced � diagnostic study/observational diffusion acquisitions (e.g. DKI or multi-b-value imaging) � performed at one institution could provide greater tissue microstructural information. Open Access This article is distributed under the terms of the Creative The good performance of the single slice ROI technique in Commons Attribution 4.0 International License (http:// IDH typing of non-enhancing lower grade gliomas was unex- creativecommons.org/licenses/by/4.0/), which permits unrestricted use, pected, but is highly relevant. It implies that such easy-to- distribution, and reproduction in any medium, provided you give appro- priate credit to the original author(s) and the source, provide a link to the perform measurements could be incorporated into clinical re- Creative Commons license, and indicate if changes were made. ports, complementing advanced MR modalities such as perfu- sion and 2HG spectroscopy [48, 49] pending tissue diagnosis. The origin of data from 18 MRI systems could represent a lim- itation of this study, but reflects clinical reality. The fact that References significant separation of glioma subtypes could be obtained from this dataset further underscores the robustness of ADC. 1. Pierallini A, Bonamini M, Bozzao A et al (1997) Supratentorial It remains unknown why intermediate ADC values were diffuse astrocytic tumours: proposal of an MRI classification. Eur observed in the 1p19q co-deleted gliomas, despite their best Radiol 7:395–399 prognosis. This result is consistent with published data on 2. Watanabe M, Tanaka R, Takeda N (1992) Magnetic resonance im- aging and histopathology of cerebral gliomas. Neuroradiology 34: intermediate diffusivity in oligodendroglioma; interestingly 463–469 this tumour subtype may also mimic malignant gliomas on 3. Tervonen O, Forbes G, Scheithauer BW, Dietz MJ (1992) Diffuse MR perfusion studies [39, 50]. Bfibrillary^ astrocytomas: correlation of MRI features with histo- In summary, the results from this study suggest that for pathologic parameters and tumor grade. Neuroradiology 34:173– newly diagnosed non-enhancing gliomas with ADC ratio 4. Ideguchi M, Kajiwara K, Goto H et al (2015) MRI findings and values of 1.8 or less, further investigation with consideration pathological features in early-stage glioblastoma. J Neurooncol of early tissue diagnosis is advisable given an increased risk of 123:289–297 wt IDH molecular status. 5. Cuccarini V, Erbetta A, Farinotti M et al (2016) Advanced MRI may complement histological diagnosis of lower grade gliomas and help in predicting survival. J Neurooncol 126:279–288 6. Olar A, Raghunathan A, Albarracin CT et al (2012) Absence of Conclusions IDH1-R132H mutation predicts rapid progression of nonenhancing diffuse glioma in older adults. Ann Diagn Pathol 16:161–170 ADC measurement appears to be a simple and powerful meth- 7. Baehring JM, Bi WL, Bannykh S et al (2007) Diffusion MRI in the early diagnosis of malignant glioma. J Neurooncol 82:221–225 od for molecular subtyping of non-enhancing WHO II–III wt 8. Ginsberg LE, Fuller GN, Hashmi M et al (1998) The significance of gliomas, specifically to identify IDH neoplasms. In our pa- lack of MR contrast enhancement of supratentorial brain tumors in tient cohort, a two-dimensional ROI measurement in the larg- adults: histopathological evaluation of a series. Surg Neurol 49: est lesion cross-section appeared representative of the entire 436–440 9. Dang L, White DW, Gross S et al (2009) Cancer-associated IDH1 tumour with comparable results. mutations produce 2-hydroxyglutarate. Nature 462:739–744 10. Yan H, Parsons DW, Jin G et al (2009) IDH1 and IDH2 mutations Funding This study received no specific funding, but was supported by in gliomas. N Engl J Med 360:765–773 the National Institute for Health Research University College London 11. Cancer Genome Atlas Research Network, Brat DJ, RGW V et al Biomedical Research Centre. (2015) Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N Engl J Med 372:2481–2498 Compliance with ethical standards 12. Louis DN, Perry A, Burger P et al (2014) International Society of Neuropathology–Haarlem consensus guidelines for nervous system Guarantor The scientific guarantor of this publication is Stefanie Thust. tumor classification and grading. Brain Pathol Zurich Switz 24: 429–435 Conflict of interest The authors of this manuscript declare no relation- 13. Eckel-Passow JE, Lachance DH, Molinaro AM et al (2015) Glioma ships with any companies whose products or services may be related to groups based on 1p/19q, IDH, and TERT promoter mutations in tumors. N Engl J Med 372:2499–2508 the subject matter of the article. 3788 Eur Radiol (2018) 28:3779–3788 14. Abudumijiti A, Chan AK-Y, Shi Z et al (2017) Adult IDH wild-type 33. Karavaeva E, Harris RJ, Leu K et al (2015) Relationship between [18F]FDOPA PET uptake, apparent diffusion coefficient (ADC), lower-grade gliomas should be further stratified. Neuro Oncol. https://doi.org/10.1093/neuonc/nox078 and proliferation rate in recurrent malignant gliomas. Mol 15. Suzuki Y, Shirai K, Oka K et al (2010) Higher pAkt expression Imaging Biol 17:434–442 predicts a significant worse prognosis in glioblastomas. J Radiat 34. Fudaba H, Shimomura T, Abe T et al (2014) Comparison of multi- Res (Tokyo) 51:343–348 ple parameters obtained on 3T pulsed arterial spin-labeling, diffu- 16. Reuss DE, Sahm F, Schrimpf D et al (2015) ATRX and IDH1- sion tensor imaging, and MRS and the Ki-67 labeling index in R132H immunohistochemistry with subsequent copy number anal- evaluating glioma grading. AJNR Am J Neuroradiol 35:2091–2098 ysis and IDH sequencing as a basis for an Bintegrated^ diagnostic 35. Ellingson BM, Malkin MG, Rand SD et al (2010) Validation of approach for adult astrocytoma, oligodendroglioma and glioblasto- functional diffusion maps (fDMs) as a biomarker for human glioma ma. Acta Neuropathol (Berl) 129:133–146 cellularity. J Magn Reson Imaging JMRI 31:538–548 17. Hartmann C, Hentschel B, Wick W et al (2010) Patients with IDH1 36. Higano S, Yun X, Kumabe T et al (2006) Malignant astrocytic wild type anaplastic astrocytomas exhibit worse prognosis than tumors: clinical importance of apparent diffusion coefficient in pre- IDH1-mutated glioblastomas, and IDH1 mutation status accounts diction of grade and prognosis. Radiology 241:839–846 for the unfavorable prognostic effect of higher age: implications for 37. Miloushev VZ, Chow DS, Filippi CG (2015) Meta-analysis of dif- classification of gliomas. Acta Neuropathol (Berl) 120:707–718 fusion metrics for the prediction of tumor grade in gliomas. AJNR 18. Darlix A, Deverdun J, Menjot de Champfleur N et al (2017) IDH Am J Neuroradiol 36:302–308 mutation and 1p19q codeletion distinguish two radiological pat- 38. Zulfiqar M, Yousem DM, Lai H (2013) ADC values and prognosis terns of diffuse low-grade gliomas. J Neurooncol 133:37–45 of malignant astrocytomas: does lower ADC predict a worse prog- 19. Metellus P, Coulibaly B, Colin C et al (2010) Absence of IDH nosis independent of grade of tumor? A meta-analysis. AJR Am J mutation identifies a novel radiologic and molecular subtype of Roentgenol 200:624–629 WHO grade II gliomas with dismal prognosis. Acta Neuropathol 39. Cui Y, Ma L, Chen X et al (2014) Lower apparent diffusion coef- (Berl) 120:719–729 ficients indicate distinct prognosis in low-grade and high-grade gli- 20. Patterson DM, Padhani AR, Collins DJ (2008) Technology insight: oma. J Neurooncol 119:377–385 water diffusion MRI–a potential new biomarker of response to can- 40. Khayal IS, Vandenberg SR, Smith KJ et al (2011) MRI apparent cer therapy. Nat Clin Pract Oncol 5:220–233 diffusion coefficient reflects histopathologic subtype, axonal dis- 21. Zhang L, Min Z, Tang M et al (2017) The utility of diffusion MRI ruption, and tumor fraction in diffuse-type grade II gliomas. with quantitative ADC measurements for differentiating high-grade Neuro Oncol 13:1192–1201 from low-grade cerebral gliomas: Evidence from a meta-analysis. J 41. Tozer DJ, Jäger HR, Danchaivijitr N et al (2007) Apparent diffusion Neurol Sci 373:9–15 coefficient histograms may predict low-grade glioma subtype. 22. Hempel J-M, Bisdas S, Schittenhelm J et al (2017) In vivo molec- NMR Biomed 20:49–57 ular profiling of human glioma using diffusion kurtosis imaging. J 42. Clark K, Voronovich Z, Horbinski C (2013) How molecular testing Neurooncol 131:93–101 can help (and hurt) in the workup of gliomas. Am J Clin Pathol 139: 23. Xiong J, Tan W, Wen J et al (2016) Combination of diffusion tensor 275–288 imaging and conventional MRI correlates with isocitrate dehydro- 43. Leu K, Ott GA, Lai A et al (2017) Perfusion and diffusion MRI genase 1/2 mutations but not 1p/19q genotyping in oligodendroglial signatures in histologic and genetic subtypes of WHO grade II-III tumours. Eur Radiol 26:1705–1715 diffuse gliomas. J Neurooncol. https://doi.org/10.1007/s11060- 24. Tan WL, Huang WY, Yin B et al (2014) Can diffusion tensor im- 0172506-9 aging noninvasively detect IDH1 gene mutations in astrogliomas? 44. Arponen O, Arponent O, Sudah M et al (2015) Diffusion-weighted A retrospective study of 112 cases. AJNR Am J Neuroradiol 35: imaging in 3.0 Tesla breast MRI: diagnostic performance and tumor 920–927 characterization using small subregions vs. whole tumor regions of 25. Han X, Suo S, Sun Y et al (2017) Apparent diffusion coefficient interest. PloS One 10:e0138702 measurement in glioma: Influence of region-of-interest determina- 45. Xing Z, Yang X, She D et al (2017) Noninvasive assessment of IDH tion methods on apparent diffusion coefficient values, interobserver mutational status in World Health Organization grade II and III variability, time efficiency, and diagnostic ability. J Magn Reson astrocytomas using DWI and DSC-PWI combined with conven- Imaging JMRI 45:722–730 tional MR imaging. AJNR Am J Neuroradiol 38:1138–1144 26. Pipe J (2009) Chapter 2 - pulse sequences for diffusion-weighted 46. Lee S, Choi SH, Ryoo I et al (2015) Evaluation of the microenvi- MRI. In: Johansen-Berg H, Behrens TEJ (eds) Diffusion MRI. ronmental heterogeneity in high-grade gliomas with IDH1/2 gene Academic Press, San Diego, pp 11–35 mutation using histogram analysis of diffusion-weighted imaging 27. Jenkinson M, Beckmann CF, Behrens TEJ et al (2012) FSL. and dynamic-susceptibility contrast perfusion imaging. J NeuroImage 62:782–790 Neurooncol 121:141–150 28. Yushkevich PA, Piven J, Hazlett HC et al (2006) User-guided 3D 47. Yamashita K, Hiwatashi A, Togao O et al (2016) MR imaging- active contour segmentation of anatomical structures: significantly based analysis of glioblastoma multiforme: estimation of IDH1 improved efficiency and reliability. NeuroImage 31:1116–1128 mutation status. AJNR Am J Neuroradiol 37:58–65 29. Jenkinson M, Bannister P, Brady M, Smith S (2002) Improved 48. Kickingereder P, Sahm F, Radbruch A et al (2015) IDH mutation optimization for the robust and accurate linear registration and mo- status is associated with a distinct hypoxia/angiogenesis tran- tion correction of brain images. NeuroImage 17:825–841 scriptome signature which is non-invasively predictable with 30. Jenkinson M, Smith S (2001) A global optimisation method for rCBV imaging in human glioma. Sci Rep 5:16238 robust affine registration of brain images. Med Image Anal 5: 49. Andronesi OC, Rapalino O, Gerstner E et al (2013) Detection of 143–156 oncogenic IDH1 mutations using magnetic resonance spectroscopy 31. Seiz M, Freyschlag CF, Schenkel S et al (2011) Management of of 2-hydroxyglutarate. J Clin Invest 123:3659–3663 patients with low-grade gliomas - a survey among German neuro- 50. Lev MH, Ozsunar Y, Henson JW et al (2004) Glial tumor grading surgical departments. Cent Eur Neurosurg 72:186–191 and outcome prediction using dynamic spin-echo MR susceptibility 32. Khan OH, Mason W, Kongkham PN et al (2016) Neurosurgical mapping compared with conventional contrast-enhanced MR: con- management of adult diffuse low grade gliomas in Canada: a founding effect of elevated rCBV of oligodendrogliomas multi-center survey. J Neurooncol 126:137–149 [corrected]. AJNR Am J Neuroradiol 25:214–221 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Radiology Springer Journals

Apparent diffusion coefficient for molecular subtyping of non-gadolinium-enhancing WHO grade II/III glioma: volumetric segmentation versus two-dimensional region of interest analysis

Download PDF
10 pages

Loading next page...
 
/lp/springer-journals/apparent-diffusion-coefficient-for-molecular-subtyping-of-non-asZ0nLsZ0R

References (57)

Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Medicine & Public Health; Imaging / Radiology; Diagnostic Radiology; Interventional Radiology; Neuroradiology; Ultrasound; Internal Medicine
ISSN
0938-7994
eISSN
1432-1084
DOI
10.1007/s00330-018-5351-0
pmid
29572636
Publisher site
See Article on Publisher Site

Abstract

Objectives To investigate if quantitative apparent diffusion coefficient (ADC) measurements can predict genetic subtypes of non- gadolinium-enhancing gliomas, comparing whole tumour against single slice analysis. Methods Volumetric T2-derived masks of 44 gliomas were co-registered to ADC maps with ADC mean (ADC )calculated. mean For the slice analysis, two observers placed regions of interest in the largest tumour cross-section. The ratio (ADC )between ratio ADC in the tumour and normal appearing white matter was calculated for both methods. mean Results Isocitrate dehydrogenase (IDH) wild-type gliomas showed the lowest ADC values throughout (p < 0.001). ADC in mean the IDH-mutant 1p19q intact group was significantly higher than in the IDH-mutant 1p19q co-deleted group (p <0.01). A −6 2 volumetric ADC threshold of 1201 × 10 mm /s identified IDH wild-type with a sensitivity of 83% and a specificity of 86%; mean a volumetric ADC cut-off value of 1.65 provided a sensitivity of 80% and a specificity of 92% (area under the curve (AUC) ratio 0.9–0.94). A slice ADC threshold for observer 1 (observer 2) of 1.76 (1.83) provided a sensitivity of 80% (86%), specificity of ratio 91% (100%) and AUC of 0.95 (0.96). The intraclass correlation coefficient was excellent (0.98). Conclusions ADC measurements can support the distinction of glioma subtypes. Volumetric and two-dimensional measurements yielded similar results in this study. Key Points � Diffusion-weighted MRI aids the identification of non-gadolinium-enhancing malignant gliomas � ADC measurements may permit non-gadolinium-enhancing glioma molecular subtyping � IDH wild-type gliomas have lower ADC values than IDH-mutant tumours � Single cross-section and volumetric ADC measurements yielded comparable results in this study . . . . Keywords Brain Diffusion magnetic resonance imaging Isocitrate dehydrogenase Glioma Neuroimaging Abbreviations DKI Diffusion kurtosis imaging ADC Apparent diffusion coefficient DTI Diffusion tensor imaging AUC Area under the curve DWI Diffusion-weighted imaging CS Centrum semiovale IDH Isocitrate dehydrogenase * S. C. Thust Lysholm Department of Neuroradiology, National Hospital for [email protected] Neurology and Neurosurgery, Queen Square, London WC1N 3BG, UK 1 Department of Clinical Neurology, National Hospital for Neurology Neuroradiological Academic Unit, Department of Brain Repair and and Neurosurgery, London, UK Rehabilitation, UCL Institute of Neurology, London, UK Department of Neurodegenerative Disease, UCL Institute of 2 Neurology and Division of Neuropathology, London, UK Imaging Department, University College London Foundation Hospital, London, UK Queen Square MS Centre. Department of Neuroinflammation, UCL Institute of Neurology, University College London, London, UK 3780 Eur Radiol (2018) 28:3779–3788 wt mut del IDH Isocitrate dehydrogenase wild-type IDH 1p19q belong to the oligodendroglioma group mut int wt IDH 1p19 Isocitrate dehydrogenase-mutant 1p19q [13]. IDH gliomas probably constitute a genetically hetero- intact geneous category of lesions, but often exhibit aggressive be- mut del IDH 1p19 Isocitrate dehydrogenase-mutant 1p19q co- haviour and have been suspected to represent early glioblas- deleted toma [14–17]. In the emerging literature on MR imaging fea- wt ICC Intraclass correlation coefficient tures of IDH glioma, initial lack of enhancement has been LGG Low grade glioma reported in some of these tumours [6, 18, 19]. NAWM Normal appearing white matter Diffusion-weighted imaging (DWI) is a technique of great PACS Picture archiving and communications interest in cancer, because water diffusivity is impaired in system highly cellular tissues, which reflects tumour proliferative rate ROC Receiver operating characteristic and aggressiveness [20]. The phenomenon of reduced diffu- ROI Centrum semiovale region of interest sion preceding fulminant radiological progression of pre- CS ROI Tumour region of interest sumed LGG has been observed prior to molecular typing tum wt TE Echo time [7], evoking later descriptions of IDH glioma serial imaging TR Repetition time findings [4]. Quantitative apparent diffusion coefficient VOI Centrum semiovale volume of interest (ADC) values have demonstrated high accuracy for glioma CS VOI Tumour volume of interest grading through meta-analysis [21]. For the non-invasive tum wt WHO World Health Organization identification of low to intermediate IDH glioma, diffusion 2HG D2-hydroxyglutarate tensor imaging (DTI) and diffusion kurtosis imaging (DKI) have shown potential, suggesting that reduced and heteroge- wt nous diffusivity are IDH features [22–24]. However, ad- Introduction vanced diffusion techniques are not universally available out- side academic hospital institutions, may require longer scan Gadolinium contrast uptake was previously considered the times and dedicated post-processing. best MR imaging predictor of glioma histological grade and Mean ADC measurement could be a rapid and practicable malignancy[1–3]. On the basis of this, it has been common approach to assess glioma diffusivity, being computationally practice to interpret non-enhancing intrinsic tumours as prob- non-demanding compared to histograms or texture analysis. able low grade gliomas (LGG) [4]. But conventional MRI has Although theoretically superior, there is no conclusive evi- proven to be unreliable in predicting subsequent tumour be- dence that whole lesion analysis outperforms region-of- haviour, whereby a proportion of presumed LGG may rapidly interest placement for the identification of malignant gliomas progress with development of malignant features such as en- [25]. hancement and necrosis [4–8]. The study presented sought to (i) investigate whether ADC The discovery of several key genetic alterations as princi- measurements from routine clinical DWI were associated with pal determinants of glioma prognosis has challenged the ref- glioma molecular subtype and (ii) to compare the performance erence standard of glioma grouping by histology [9]. of volumetric whole tumour ADC with single slice ADC Mutations in isocitrate dehydrogenase (IDH) represent a com- measurements. mon (> 70%) defining event in the development of LGG, conversely more than 90% of glioblastomas belong to the IDH wild-type group [10, 11]. Despite its oncogenic effect Materials and methods through production of a toxic metabolite D2- hydroxyglutarate (2HG), the presence of an IDH mutation is Patients associated with a favourable prognosis. The revised 2016 World Health Organization (WHO) clas- Following institutional board approval for a retrospective sification of brain tumours for the first time incorporates mo- study, we searched the neuropathology records revealing 37 wt lecular data to augment the diagnosis [12]. For WHO grade II/ patients with WHO grade II/III IDH glioma between 2009 III gliomas, three molecular subgroups have been defined: and 2016. For comparison of the molecular groups, control wt IDH wild-type glioma (IDH ) with survival similar to that samples of IDH (IDH1-R132H) mutant gliomas (34 mut int mut del of glioblastoma, IDH-mutant glioma with intact 1p19q IDH 1p19q and 32 IDH 1p19q )wererandomlyse- mut int (IDH 1p19 ) and an intermediate prognosis, and IDH- lected. We sought to evaluate ADC for suspected LGG prior mut del mutant 1p19q co-deleted glioma (IDH 1p19q )with the to tissue diagnosis. To replicate the clinical situation, only best prognosis and greatest chemosensitivity [11]. There is gliomas without gadolinium enhancement were included (2 partial overlap with histomorphology, whereby many non-enhancing gliomas were excluded because of missing mut int IDH 1p19 are astrocytic and the majority of images and degraded DWI, respectively). The study sample Eur Radiol (2018) 28:3779–3788 3781 wt consisted of 14 IDH (7 WHO II and 7 WHO III), 16 90.5 [69.5, 137] ms; TR = 4000 [2837, 10,000] ms, in-plane mut int 2 IDH 1p19q (8 WHO II and 8 WHO III) and 14 resolution = 1.25 × 1.25 [0.5 × 0.5, 2.5 × 2.5] mm ; slice mut del IDH 1p19q (11 WHO II and 3 WHO III), amounting to thickness = 5 [4, 6] mm; gap between slices = 1.5 [0, 2] 44 non-enhancing gliomas for the three molecular groups mm. For each patient, the imaging study was performed on (patient selection diagram shown in Fig. 1). No haemorrhagic average (standard deviation, sd) 2.3 (2.8) months prior to the or necrotic gliomas were featured in the study. tissue diagnosis. Image examples for the glioma molecular subgroups are shown in Fig. 2. MRI acquisition Post-processing and ADC analysis Ours is a quaternary neurosurgical centre; therefore the stan- dard (structural and DWI) MRI sequences in this study orig- ADC map calculation inated from 10 different referring institutions (institution 1 to institution 10): 29 from our own institutions, 4 from institution In a spin echo diffusion-weighted sequence, the signal S [S = b b 2, 3 from institution 3, 2 from institution 4, and one each from (−b ADC) S e ] from each pixel in an image is formed of a first the remaining six institutions. The studies were acquired on 18 component (S ) dependent on tissue properties (i.e. ‘spin den- different scanners (31 at 1.5 Tesla, and 13 at 3 Tesla) from all sity’,T and T relaxation times) and sequence properties (e.g. 1 2 major vendors: four General Electric scanners [Discovery −b ADC repetition time, TR); and a second component (e )de- MR450 (number of patients n = 5), 2× Signa Excite (n =1 pendent on the diffusion gradients (b, in units of s/mm )and each), Genesis Signa (n = 2)], seven Siemens scanners [3× the apparent diffusion coefficient (ADC, in units of mm /s). Avanto (n =7, n =2, n =1), aTrio (n = 9), Symphony (n = The ADC is obtained by dividing the image acquired with- 4), Skyra (n =3), Espree(n = 1)], six Philips scanners [Ingenia out diffusion gradients (S =S ) by the image acquired with b =0 0 (n =2), 5× Achieva (n = 1 each)] and one Toshiba scanner (n = diffusion gradients (S ): 1). All acquisitions included axial T2-weighted images, and axial standard 3-directional whole brain DWI. The median ADC ¼ðÞ 1=b lnðÞ S =S ð1Þ 0 b [min, max] values of the parameters of the T2-weighted im- In this division, the dependence of ADC from S (and ages were echo time (TE) = 99.5 [80, 141] ms; repetition time 0 therefore from T ,T and TR) is eliminated [26]. The ADC (TR) = 4610 [2500, 7480] ms, in-plane resolution = 0.5 × 0.5 1 2 maps were calculated using Eq. 1 and the utility fslmaths from [0.3 × 0.3, 0.9 × 0.9] mm ; slice thickness = 5 [1, 6] mm; gap the software library fsl (version 5.0) [27]. Offline whole tu- between slices = 1.5 [0, 2] mm. All DWI acquisitions included mour analysis and single slice analysis were subsequently diffusion gradient weighting values b = 0 s/mm and b =1000 performed. s/mm ; the median [min, max] of other parameters were TE = Fig. 1 Flow diagram of patients included and excluded from the analyses 3782 Eur Radiol (2018) 28:3779–3788 Fig. 2 WHO II/III molecular subgroup examples showing T2- weighted images, b1000, ADC maps and T1-weighted post gadolinium images of non- wt enhancing a IDH , b mut int IDH 1p19q and c mut del IDH 1p19q glioma Whole tumour (volumetric) ADC analysis Single slice ADC analysis Tumour volumes of interest (VOI ) were outlined by a Standard picture archiving and communication systems tum neuroradiology resident (S.H.) using ITK snap Toolbox (PACS) software (IMPAX 6.5.1.1008, Agfa-Gevaert, Mortsel, version 3.6 (www.itksnap.org [28]), covering the entire Belgium) was used to exploit tools routinely available for T2 signal abnormality with each segmentation optimised reporting of MR images. Two observers blinded to by a board-certified neuroradiologist specialised in brain histomolecular results (J.A.M. general radiology trainee = ob- tumour imaging (S.C.T.). For multicentric gliomas, the server 1 and S.C.T. = observer 2) located the tumour on the T2- total volume of signal abnormality was treated as one weighted sequence, selecting two round regions of interest on lesion. ADC maps were co-registered to T2 imaging using the ADC map viewed side-by-side: The first region of interest the FLIRT toolbox [29, 30]performingarigidbodytrans- (ROI ) was drawn in the largest lesion cross-section sparing tum formation with a six-parameter model and ‘Normalised the tumour margin to avoid partial volume effects. The second Mutual Information’ as cost function. Subsequently, round ROI aiming for a similar size to ROI was placed in CS tum ADC measurements were obtained for each tumour, contralateral centrum semiovale NAWM, taking care to exclude mean using the fslstats utility from fsl [25–27]. images with visible ventricular surfaces, cortex and/or sulcal To consider possible interindividual variations in brain spaces at measurement level. Three patients were excluded diffusivity, we assessed the ADC in normal appearing from the single slice analysis because of non-availability of an mean white matter (NAWM). For each patient, a standardised ADC map on PACS. The ratio between the ADC in the mean second volume of interest (VOI ) was drawn in the con- tumour and CS was calculated [PACS_ADC = CS ratio tralateral centrum semiovale (CS). This VOI was used ADC (ROI )/ADC (ROI )]. No absolute ADC CS mean tum mean CS to calculate the ADC =ADC (VOI )/ values were measured by the single slice method, as their work- ratio mean tum wt ADC (VOI )(Fig. 2). For two IDH tumours, the station display can vary depending on the referring institution. mean CS NAWM analysis was omitted because of bilateral tumour An example of the volumetric segmentation and single slice infiltration. ADC measurement is demonstrated in Fig. 3. Eur Radiol (2018) 28:3779–3788 3783 Fig. 3 Image examples demonstrating the whole lesion volumetric segmentation (mask overlaid on right frontal mut int IDH 1p19q glioma), single slice largest tumour cross-section ROI and comparative tum contralateral NAWM ROI CS placements Histopathology and molecular analysis Results wt Paraffin blocks containing tissue were analysed at our institu- The mean age was greater in the IDH group than in the mut mut int tion’s neuropathology department according to WHO 2016 IDH groups (p = 0.0001 for IDH 1p19q , p = 0.005 mut del guidance and previously published data [16]. IDH R132H for IDH 1p19q ). The larger proportion of WHO II glio- mut del immuno-negative tumours underwent multiple gene Sanger mas in the IDH 1p19q was not statistically significant wt sequencing. A quantitative polymerase chain reaction-based (Pearson chi-square test p = 0.115 for IDH and p =0.105 mut int copy number assay was used to determine 1p/19q status. for IDH 1p19q ). The patient demographic data and tu- mour volumes are reported in Table 1. Statistical analysis Association between molecular subtype and ADC For the volumetric and single slice data, the statistical analysis values consisted of two steps each: (i) linear regression to assess the wt mut int wt association between the tumour type (IDH ,IDH 1p19q , In the volumetric analysis, IDH tumours showed significantly mut del IDH 1p19q ) and ADC values, followed by (ii) logistic lower whole tumour volume ADC (VOI )than mean tum mut int mut del regression to determine if ADC values can differentiate IDH 1p19q (p < 0.0005) and IDH 1p19q (p =0.001). wt mut mut int IDH from IDH gliomas. A receiver operating character- The ADC (VOI )intheIDH 1p19q group was signif- mean tum mut del istic (ROC) analysis was used to quantify the performance of icantlyhigherthanintheIDH 1p19q group (p = 0.0047). wt the logistic regression. For the identification of a cut-off point IDH gliomas had a significantly lower whole tumour mut int for the logistic regression the ‘nearest to (0,1)’ method was ADC than IDH 1p19q (p < 0.0005) and ratio mut del performed. Statistical significance was set at 5%. The inter- IDH 1p19q (p =0.019). TheADC in the ratio mut int rater agreement was expressed as an intraclass correlation co- IDH 1p19q group was significantly higher than in the mut del efficient (ICC) using a two-way random effects model. All IDH 1p19q group (p =0.0054). statistical analyses were performed using Stata version 14 On single slice assessment, a significantly lower mean wt (College Station, TX: StataCorp LP). PACS_ADC was observed for IDH than for ratio Table 1 Patient demographic data and tumour volumes Whole tumour ADC (VOI ) mean tum Patient group Nr of patients Age in years Tumour volume CS NAWM volume Tumour volume for patients 3 3 total (male) (mean ± sd) (years) (mean ± sd) (cm ) (mean ± sd) (cm ) with bilateral infiltration (mean ± sd) (cm ) wt IDH 14 (9) 53 (± 14) 64 (± 68) (n = 12) 11.6 (± 2.5) (n = 12) 366 (± 46) (n =2) mut int IDH 1p19 16 (6) 33.9 (± 8.6) 60 (± 44) 10.9 (± 2.3) N/A mut del IDH 1p19 14 (7) 38.9 (± 8.3) 48 (± 50) 10.8 (± 2.5) N/A 3784 Eur Radiol (2018) 28:3779–3788 mut int mut int Table 3 F test for the difference between IDH 1p19 and IDH 1p19q (p <0.0005 observer 1; p <0.0005 observer mut del IDH 1p19 mut del 2) and for IDH 1p19q (p = 0.001 observer 1; p =0.001 mut int observer 2). The PACS_ADC in the IDH 1p19q group ratio Analysis type p mut del was higher than in the IDH 1p19q group (p =0.0008 for ADC (VOI)0.0047 mean tum observer 1 and p = 0.0025 for observer 2). No statistical asso- Whole tumour ADC 0.0054 ratio ciations were demonstrated between the NAWM ADC mean PACS_ADC 1st observer 0.0008 ratio values and molecular subtype. PACS_ADC 2nd observer 0.0025 ratio The intra-rater agreement for the PACS_ADC mea- ratio surements was very high: the correlation of measurements made on the same individual was 0.96, while the correla- Diagnostic performance of ADC values tion between mean observer ratings was 0.98. The correla- tion of measurements equaled the consistency agreement, For ADC (VOI ), a ROC analysis quantified the accuracy mean tum indicating no systematic difference between the two ob- of correctly classifying tumour type to an area under the curve servers. The single slice ADC values were slightly but ratio (AUC) of 0.94. The cut-off point for the ADC (VOI )was mean tum systematically higher than the volumetric ADC .The −6 2 ratio 1201 × 10 mm /s, with a sensitivity of 0.83 and a specificity numerical results of the association between tumour type of 0.86. For a decrease in the ADC (VOI ) value by 1.0 × mean tum and ADC values for the volumetric and single slice analy- −5 2 wt 10 mm /s, the odds of IDH increasedby78% (p = 0.003). ses are reported in Table 2.InTable 3, the difference be- For the volumetric ADC , the ROC analysis yielded an ratio mut int tween the ADC values in IDH 1p19q and in AUC of 0.90 with a sensitivity of 0.80 and a specificity of mut del IDH 1p19q is shown. In Table 4 the ICC values are 0.92 for a threshold ADC of 1.65. For a decrease in the ratio detailed. The boxplots of the ADC and ADC values wt mean ratio volumetric ADC value by 0.1, the odds of IDH increased ratio are depicted in Fig. 4. by 46% (p = 0.004). wt Table 2 Results of the linear regression between ADC and tumour type (IDH is the reference group) Whole tumour ADC (VOI ) mean tum Patient group ADC (VOI ) Regression 95% CI of the p mean tum mean (sd) coefficient regr. coeff. −6 2 −6 2 −6 2 (10 mm /s) (10 mm /s) (10 mm /s) wt IDH 1032 (168) 1032 922–1141 0.0005 mut int IDH 1p19 1543 (254) 511 361–661 0.0005 mut del IDH 1p19 1321 (162) 289 134–444 0.001 Whole tumour ADC ratio Patient group ADC Regression 95% CI of the p ratio mean (sd) coefficient regr. coeff. wt IDH 1.49 (0.32) 1.49 1.32–1.66 0.0005 mut int IDH 1p19 2.09 (0.34) 0.59 0.37–0.82 0.0005 mut del IDH 1p19 1.77 (0.20) 0.28 0.05–0.51 0.019 Single slice PACS_ADC first observer ratio Patient group PACS_ADC Regression 95% CI of the p ratio mean (sd) coefficient regr. coeff. wt IDH 1.50 (0.21) 1.50 1.33–1.68 0.0005 mut int IDH 1p19 2.37 (0.35) 0.87 0.63–1.10 0.0005 mut del IDH 1p19 1.96 (0.27) 0.45 0.20–0.70 0.001 Single slice PACS_ADC second observer ratio Patient group PACS_ADC Regression 95% CI of the p ratio mean (sd) coefficient regr. coeff. wt IDH 1.48 (0.19) 1.48 1.28–1.68 0.0005 mut int IDH 1p19 2.37 (0.38) 0.88 0.62–1.14 0.0005 mut del IDH 1p19 1.96 (0.36) 0.47 0.20–0.75 0.001 mut Regression coefficient represents the difference in the dependent variable (ADC) between each of the two IDH groups and the reference group wt (IDH ) Eur Radiol (2018) 28:3779–3788 3785 Table 4 Inter-rater agreement expressed as intraclass correlation Discussion coefficient (ICC) In this analysis, we observed that ADC values obtained from Correlation ICC Consistency ICC (95% CI) (95% CI) standard clinical DWI are a highly significant predictor of non-enhancing glioma IDH status and may permit non- Observer 1 vs observer 2 - PACS_ADC ratio invasive molecular subtyping in accordance with the 2016 Individual ICC 0.96 (0.92–0.98) 0.96 (0.92–0.98) WHO classification. Average ICC 0.98 (0.96–0.99) 0.98 (0.96–0.99) Two recent surveys highlighted clinical practices in caring Observer 1 PACS_ADC vs volumetric ADC ratio ratio for patients with presumed LGG, with approximately 50% of Individual ICC 0.80 (0.35–0.92) 0.87 (0.77–0.93) neurosurgeons adopting a ‘wait and see’ approach balanced Average ICC 0.89 (0.52–0.96) 0.93 (0.87–0.96) against surgical risk [31], and only 21% performing an upfront Observer 2 PACS_ADC vs volumetric ADC wt ratio ratio biopsy [32]. Consequently, innocuous appearing IDH glio- Individual ICC 0.79 (0.43–0.91) 0.85 (0.74–0.92) mas may reveal their aggressive nature through progression Average ICC 0.88 (0.60–0.95) 0.92 (0.85–0.96) and receive treatment with a delay. Low ADC values are associated with increased glioma cel- lularity and worse prognosis, supported by comparisons of dif- fusivity, histological specimens and clinical data in multiple A ROC analysis quantified the accuracy of the studies [5, 33–37]. Low diffusivity predicts poor astrocytoma survival independent from WHO grade [38], although no linear PACS_ADC logistic regression in correctly classifying tu- ratio mour type to an AUC of 0.96 for observer 1 and 0.95 for relation exists between ADC and glioma prognosis [39]. Past studies to distinguish astrocytoma and oligodendroglioma observer 2. The cut-off point for the PACS_ADC for ob- ratio using ADC values yielded variable success [40, 41],andinret- server 1 (observer 2) was 1.83 (1.76) with a sensitivity of 0.80 rospect may have been influenced by the incomplete overlap (0.86) and a specificity of 1.00 (0.91) at the cut-off point. For a between histological and molecular groups. Diagnostic focus decrease in the single slice ADC value by 0.1, the odds of ratio wt has shifted to genetic typing, yet immunohistochemistry tests IDH increased by 62% (p = 0.005) for observer 1 and 57% are complex and not infallible, requiring interpretation in in the (p = 0.004) for observer 2. The numerical results for glioma subtype prediction are reported in Table 5. The ROC curves context of morphological criteria and test type performed to avoid interpretational errors [42]. are depicted in Fig. 5. Fig. 4 Boxplot of the values of the a whole tumour ADC (VOI ), b whole mean tum tumour ADC , c single slice ratio PACS_ADC first observer and ratio d single slice PACS_ADC ratio first observer 3786 Eur Radiol (2018) 28:3779–3788 Table 5 Cut-off point estimation Method Cut-off point Sensitivity at cut-off point Specificity at cut-off point AUC at cut-off point ADC (VOI)1201 0.83 0.86 0.85 mean tum −6 2 (10 mm /s) Whole tumour ADC 1.65 0.80 0.92 0.86 ratio PACS_ADC 1st observer 1.83 0.86 1.00 0.93 ratio PACS_ADC 2nd observer 1.76 0.86 0.91 0.88 ratio Recently, Leu et al. were able to assign gliomas to the WHO type. The quicker and easier single slice analysis even performed 2016 molecular groups using ADC; however, their method dif- marginally better. This is in line with results of previous imaging fered from ours by including enhancing lesions and ADC median research, which suggested that whole lesion diffusivity measure- values derived from b700–1000 gradients with DTI analysed for ment is not always superior to ROI analysis [25, 44]. some patients [43]. To our best knowledge, this is the first IDH The ability of ADC to predict glioma subtypes and optimum typing study to focus on non-enhancing gliomas, using b1000 thresholds may be subject to ROI placement technique with pre- values derived from 3-directional DWI. This is particularly im- vious research focusing on minimum ADC value analysis: Xing portant, as such tumours are usually assumed to be less aggres- et al. showed a statistical correlation between ADC and IDH sive in common clinical practice. status using a multiple (≥ 5) ROI technique with the mean of We found ADC values to be closely reproducible when the lowest ADC measurement chosen as minimum ADC in con- ratio comparing whole lesion measurements against single slice region sensus [45]. In a similar fashion, a previous DTI study for IDH of interest placements, for which there was near complete inter- typing used multiple ROI placements and a two-reader consen- observer agreement. The similarity of our volumetric and single sus method to obtain minimum ADC values [24]. slice results could be explained by a relative homogeneity of As a reference ROI, we chose the centrum semiovale for its these non-enhancing, non-necrotic gliomas. Both the absolute potentially greater reproducibility compared to a ‘mirror’ ROI ADC values and ADC appear valuable for this lesion [45], because this could be influenced by tumour location. We mean ratio Fig. 5 ROC curves for the a whole tumour ADC (VOI ), b whole tumour ADC , c single slice PACS_ADC observer 1 and d observer 2 mean tum ratio ratio Eur Radiol (2018) 28:3779–3788 3787 Statistics and biometry One of the authors (C.T. MD PhD MSc Medical avoided the internal capsule [24], which is a smaller structure Statistics) has significant statistics expertise. and more difficult to locate by an untrained rater. Lee et al. found ADC mean and ADC histograms useful for Informed consent Written informed consent was waived by the institu- IDH typing of WHO grade III and IV gliomas [46]. However, tional review board. for glioblastoma IDH typing alone, a recent study identified no difference in ADC values [47]. In Tan et al.’s study of grade Ethical approval Institutional review board approval was obtained. II–IV gliomas, the accuracy of ADC for IDH typing decreased Methodology with higher grade, which may reflect greater lesion heteroge- � retrospective neity [24]. It is probable that in such circumstances advanced � diagnostic study/observational diffusion acquisitions (e.g. DKI or multi-b-value imaging) � performed at one institution could provide greater tissue microstructural information. Open Access This article is distributed under the terms of the Creative The good performance of the single slice ROI technique in Commons Attribution 4.0 International License (http:// IDH typing of non-enhancing lower grade gliomas was unex- creativecommons.org/licenses/by/4.0/), which permits unrestricted use, pected, but is highly relevant. It implies that such easy-to- distribution, and reproduction in any medium, provided you give appro- priate credit to the original author(s) and the source, provide a link to the perform measurements could be incorporated into clinical re- Creative Commons license, and indicate if changes were made. ports, complementing advanced MR modalities such as perfu- sion and 2HG spectroscopy [48, 49] pending tissue diagnosis. The origin of data from 18 MRI systems could represent a lim- itation of this study, but reflects clinical reality. The fact that References significant separation of glioma subtypes could be obtained from this dataset further underscores the robustness of ADC. 1. Pierallini A, Bonamini M, Bozzao A et al (1997) Supratentorial It remains unknown why intermediate ADC values were diffuse astrocytic tumours: proposal of an MRI classification. Eur observed in the 1p19q co-deleted gliomas, despite their best Radiol 7:395–399 prognosis. This result is consistent with published data on 2. Watanabe M, Tanaka R, Takeda N (1992) Magnetic resonance im- aging and histopathology of cerebral gliomas. Neuroradiology 34: intermediate diffusivity in oligodendroglioma; interestingly 463–469 this tumour subtype may also mimic malignant gliomas on 3. Tervonen O, Forbes G, Scheithauer BW, Dietz MJ (1992) Diffuse MR perfusion studies [39, 50]. Bfibrillary^ astrocytomas: correlation of MRI features with histo- In summary, the results from this study suggest that for pathologic parameters and tumor grade. Neuroradiology 34:173– newly diagnosed non-enhancing gliomas with ADC ratio 4. Ideguchi M, Kajiwara K, Goto H et al (2015) MRI findings and values of 1.8 or less, further investigation with consideration pathological features in early-stage glioblastoma. J Neurooncol of early tissue diagnosis is advisable given an increased risk of 123:289–297 wt IDH molecular status. 5. Cuccarini V, Erbetta A, Farinotti M et al (2016) Advanced MRI may complement histological diagnosis of lower grade gliomas and help in predicting survival. J Neurooncol 126:279–288 6. Olar A, Raghunathan A, Albarracin CT et al (2012) Absence of Conclusions IDH1-R132H mutation predicts rapid progression of nonenhancing diffuse glioma in older adults. Ann Diagn Pathol 16:161–170 ADC measurement appears to be a simple and powerful meth- 7. Baehring JM, Bi WL, Bannykh S et al (2007) Diffusion MRI in the early diagnosis of malignant glioma. J Neurooncol 82:221–225 od for molecular subtyping of non-enhancing WHO II–III wt 8. Ginsberg LE, Fuller GN, Hashmi M et al (1998) The significance of gliomas, specifically to identify IDH neoplasms. In our pa- lack of MR contrast enhancement of supratentorial brain tumors in tient cohort, a two-dimensional ROI measurement in the larg- adults: histopathological evaluation of a series. Surg Neurol 49: est lesion cross-section appeared representative of the entire 436–440 9. Dang L, White DW, Gross S et al (2009) Cancer-associated IDH1 tumour with comparable results. mutations produce 2-hydroxyglutarate. Nature 462:739–744 10. Yan H, Parsons DW, Jin G et al (2009) IDH1 and IDH2 mutations Funding This study received no specific funding, but was supported by in gliomas. N Engl J Med 360:765–773 the National Institute for Health Research University College London 11. Cancer Genome Atlas Research Network, Brat DJ, RGW V et al Biomedical Research Centre. (2015) Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N Engl J Med 372:2481–2498 Compliance with ethical standards 12. Louis DN, Perry A, Burger P et al (2014) International Society of Neuropathology–Haarlem consensus guidelines for nervous system Guarantor The scientific guarantor of this publication is Stefanie Thust. tumor classification and grading. Brain Pathol Zurich Switz 24: 429–435 Conflict of interest The authors of this manuscript declare no relation- 13. Eckel-Passow JE, Lachance DH, Molinaro AM et al (2015) Glioma ships with any companies whose products or services may be related to groups based on 1p/19q, IDH, and TERT promoter mutations in tumors. N Engl J Med 372:2499–2508 the subject matter of the article. 3788 Eur Radiol (2018) 28:3779–3788 14. Abudumijiti A, Chan AK-Y, Shi Z et al (2017) Adult IDH wild-type 33. Karavaeva E, Harris RJ, Leu K et al (2015) Relationship between [18F]FDOPA PET uptake, apparent diffusion coefficient (ADC), lower-grade gliomas should be further stratified. Neuro Oncol. https://doi.org/10.1093/neuonc/nox078 and proliferation rate in recurrent malignant gliomas. Mol 15. Suzuki Y, Shirai K, Oka K et al (2010) Higher pAkt expression Imaging Biol 17:434–442 predicts a significant worse prognosis in glioblastomas. J Radiat 34. Fudaba H, Shimomura T, Abe T et al (2014) Comparison of multi- Res (Tokyo) 51:343–348 ple parameters obtained on 3T pulsed arterial spin-labeling, diffu- 16. Reuss DE, Sahm F, Schrimpf D et al (2015) ATRX and IDH1- sion tensor imaging, and MRS and the Ki-67 labeling index in R132H immunohistochemistry with subsequent copy number anal- evaluating glioma grading. AJNR Am J Neuroradiol 35:2091–2098 ysis and IDH sequencing as a basis for an Bintegrated^ diagnostic 35. Ellingson BM, Malkin MG, Rand SD et al (2010) Validation of approach for adult astrocytoma, oligodendroglioma and glioblasto- functional diffusion maps (fDMs) as a biomarker for human glioma ma. Acta Neuropathol (Berl) 129:133–146 cellularity. J Magn Reson Imaging JMRI 31:538–548 17. Hartmann C, Hentschel B, Wick W et al (2010) Patients with IDH1 36. Higano S, Yun X, Kumabe T et al (2006) Malignant astrocytic wild type anaplastic astrocytomas exhibit worse prognosis than tumors: clinical importance of apparent diffusion coefficient in pre- IDH1-mutated glioblastomas, and IDH1 mutation status accounts diction of grade and prognosis. Radiology 241:839–846 for the unfavorable prognostic effect of higher age: implications for 37. Miloushev VZ, Chow DS, Filippi CG (2015) Meta-analysis of dif- classification of gliomas. Acta Neuropathol (Berl) 120:707–718 fusion metrics for the prediction of tumor grade in gliomas. AJNR 18. Darlix A, Deverdun J, Menjot de Champfleur N et al (2017) IDH Am J Neuroradiol 36:302–308 mutation and 1p19q codeletion distinguish two radiological pat- 38. Zulfiqar M, Yousem DM, Lai H (2013) ADC values and prognosis terns of diffuse low-grade gliomas. J Neurooncol 133:37–45 of malignant astrocytomas: does lower ADC predict a worse prog- 19. Metellus P, Coulibaly B, Colin C et al (2010) Absence of IDH nosis independent of grade of tumor? A meta-analysis. AJR Am J mutation identifies a novel radiologic and molecular subtype of Roentgenol 200:624–629 WHO grade II gliomas with dismal prognosis. Acta Neuropathol 39. Cui Y, Ma L, Chen X et al (2014) Lower apparent diffusion coef- (Berl) 120:719–729 ficients indicate distinct prognosis in low-grade and high-grade gli- 20. Patterson DM, Padhani AR, Collins DJ (2008) Technology insight: oma. J Neurooncol 119:377–385 water diffusion MRI–a potential new biomarker of response to can- 40. Khayal IS, Vandenberg SR, Smith KJ et al (2011) MRI apparent cer therapy. Nat Clin Pract Oncol 5:220–233 diffusion coefficient reflects histopathologic subtype, axonal dis- 21. Zhang L, Min Z, Tang M et al (2017) The utility of diffusion MRI ruption, and tumor fraction in diffuse-type grade II gliomas. with quantitative ADC measurements for differentiating high-grade Neuro Oncol 13:1192–1201 from low-grade cerebral gliomas: Evidence from a meta-analysis. J 41. Tozer DJ, Jäger HR, Danchaivijitr N et al (2007) Apparent diffusion Neurol Sci 373:9–15 coefficient histograms may predict low-grade glioma subtype. 22. Hempel J-M, Bisdas S, Schittenhelm J et al (2017) In vivo molec- NMR Biomed 20:49–57 ular profiling of human glioma using diffusion kurtosis imaging. J 42. Clark K, Voronovich Z, Horbinski C (2013) How molecular testing Neurooncol 131:93–101 can help (and hurt) in the workup of gliomas. Am J Clin Pathol 139: 23. Xiong J, Tan W, Wen J et al (2016) Combination of diffusion tensor 275–288 imaging and conventional MRI correlates with isocitrate dehydro- 43. Leu K, Ott GA, Lai A et al (2017) Perfusion and diffusion MRI genase 1/2 mutations but not 1p/19q genotyping in oligodendroglial signatures in histologic and genetic subtypes of WHO grade II-III tumours. Eur Radiol 26:1705–1715 diffuse gliomas. J Neurooncol. https://doi.org/10.1007/s11060- 24. Tan WL, Huang WY, Yin B et al (2014) Can diffusion tensor im- 0172506-9 aging noninvasively detect IDH1 gene mutations in astrogliomas? 44. Arponen O, Arponent O, Sudah M et al (2015) Diffusion-weighted A retrospective study of 112 cases. AJNR Am J Neuroradiol 35: imaging in 3.0 Tesla breast MRI: diagnostic performance and tumor 920–927 characterization using small subregions vs. whole tumor regions of 25. Han X, Suo S, Sun Y et al (2017) Apparent diffusion coefficient interest. PloS One 10:e0138702 measurement in glioma: Influence of region-of-interest determina- 45. Xing Z, Yang X, She D et al (2017) Noninvasive assessment of IDH tion methods on apparent diffusion coefficient values, interobserver mutational status in World Health Organization grade II and III variability, time efficiency, and diagnostic ability. J Magn Reson astrocytomas using DWI and DSC-PWI combined with conven- Imaging JMRI 45:722–730 tional MR imaging. AJNR Am J Neuroradiol 38:1138–1144 26. Pipe J (2009) Chapter 2 - pulse sequences for diffusion-weighted 46. Lee S, Choi SH, Ryoo I et al (2015) Evaluation of the microenvi- MRI. In: Johansen-Berg H, Behrens TEJ (eds) Diffusion MRI. ronmental heterogeneity in high-grade gliomas with IDH1/2 gene Academic Press, San Diego, pp 11–35 mutation using histogram analysis of diffusion-weighted imaging 27. Jenkinson M, Beckmann CF, Behrens TEJ et al (2012) FSL. and dynamic-susceptibility contrast perfusion imaging. J NeuroImage 62:782–790 Neurooncol 121:141–150 28. Yushkevich PA, Piven J, Hazlett HC et al (2006) User-guided 3D 47. Yamashita K, Hiwatashi A, Togao O et al (2016) MR imaging- active contour segmentation of anatomical structures: significantly based analysis of glioblastoma multiforme: estimation of IDH1 improved efficiency and reliability. NeuroImage 31:1116–1128 mutation status. AJNR Am J Neuroradiol 37:58–65 29. Jenkinson M, Bannister P, Brady M, Smith S (2002) Improved 48. Kickingereder P, Sahm F, Radbruch A et al (2015) IDH mutation optimization for the robust and accurate linear registration and mo- status is associated with a distinct hypoxia/angiogenesis tran- tion correction of brain images. NeuroImage 17:825–841 scriptome signature which is non-invasively predictable with 30. Jenkinson M, Smith S (2001) A global optimisation method for rCBV imaging in human glioma. Sci Rep 5:16238 robust affine registration of brain images. Med Image Anal 5: 49. Andronesi OC, Rapalino O, Gerstner E et al (2013) Detection of 143–156 oncogenic IDH1 mutations using magnetic resonance spectroscopy 31. Seiz M, Freyschlag CF, Schenkel S et al (2011) Management of of 2-hydroxyglutarate. J Clin Invest 123:3659–3663 patients with low-grade gliomas - a survey among German neuro- 50. Lev MH, Ozsunar Y, Henson JW et al (2004) Glial tumor grading surgical departments. Cent Eur Neurosurg 72:186–191 and outcome prediction using dynamic spin-echo MR susceptibility 32. Khan OH, Mason W, Kongkham PN et al (2016) Neurosurgical mapping compared with conventional contrast-enhanced MR: con- management of adult diffuse low grade gliomas in Canada: a founding effect of elevated rCBV of oligodendrogliomas multi-center survey. J Neurooncol 126:137–149 [corrected]. AJNR Am J Neuroradiol 25:214–221

Journal

European Radiology – Springer Journals

Published: Mar 23, 2018

There are no references for this article.