Background

Low-grade gliomas (LGG) represent 30 % of all gliomas and include tumors with astrocytic and/or oligodendroglial features [1]. Natural evolution is towards malignancy but survival rate varies from less than 2 years to more than 10 years. Clinical approaches range from “wait and see” to early diagnosis through biopsy or surgical resection [2]. Subsequent treatments must be carefully evaluated given the possibility of long-term adverse effects. For these reasons, non-invasive, pre-surgical diagnosis and grading are fundamental in the decision-making process and clinical management.

Although conventional Magnetic Resonance Imaging (cMRI) with contrast agents is an established and useful tool in the characterization of cerebral tumours, its sensitivity for classification and grading of gliomas ranges 55–83 % [3, 4]. High-grade glioma (HGG) may be misdiagnosed as LGG by cMRI in the absence of oedema, contrast enhancement (CE), or necrosis. In contrast, CE has been reported in up to 40 % of LGG [5]. Furthermore, cMRI doesn’t provide information about blood volume, angiogenesis, micronecrosis, cellularity, which are important features to define tumour grade [4].

Advanced MRI (aMRI), such as diffusion- and perfusion-weighted imaging (DWI and PWI, respectively) and magnetic resonance spectroscopy (MRS), have been widely used in pre-surgical assessment and follow-up of gliomas [6]. Unlike CE, increased perfusion is generally independent of blood–brain-barrier integrity and defines tumour neoangiogenesis [7]. On DWI, the apparent diffusion coefficient (ADC) inversely correlates to cellularity and reflects tumour infiltration before changes are visible on cMRI [8]. MRS provides information about the metabolites choline (Cho), N-acetyl aspartate (NAA), and lipids/lactate, useful for tumour grading [4].

Several studies focused on the usefulness of aMRI in gliomas [3, 4, 8–20], suggesting a relationship of cut-off values to histopathology [4, 21–26] and, more recently, to survival [10, 20, 25, 27–35].

Even though neuropathology is still considered the gold standard in glioma diagnosis, its limitations due to tumour heterogeneity and insufficiently representative specimens [27] are acknowledged.

Aims of the present study were to assess the possible pre-operative diagnostic advantage of associating aMRI (DWI, PWI, MRS) to cMRI in a group of selected patients suspected to have LGG based on cMRI and to investigate whether aMRI can be useful in the prediction of outcome (overall survival [OS] and progression-free survival [PFS]).

Methods

Patients

This prospective observational study of radiological diagnosis of gliomas was performed at the Neurological Institute Besta, Milan, Italy. Institution Ethics Committee was regularly informed about the study, and informed consent was obtained from all participants. Between 2006 and 2009, adult patients with suspected supratentorial glioma according to pre-surgical cMRI were considered for inclusion. Patients scheduled for biopsy, who had already undergone biopsy/surgery, or treated with chemo- or radio-therapy before surgery were excluded. LGG was suspected by cMRI when no necrosis or relevant oedema (i.e., absence of oedema or minimal oedema not distinguishable from tumour burden) was observed and when CE was absent or faint [5]. Patients who fulfilled these cMRI criteria underwent all aMRI techniques (PWI, DWI, MRS) and complete cMRI analysis that is routinely performed within the Institute. Clinical and cMRI follow-up was performed every six months.

Magnetic resonance imaging

MRI was performed with a 1.5 T unit (Siemens, Avanto).

cMRI protocol

Sagittal and bi-commissural T1, bi-commissural T2-proton density weighted images (w.i.), coronal Fluid Attenuated Inversion Recovery (FLAIR). Post-contrast volumetric T1 w.i. was performed after PWI. The following features were considered for complete cMRI analysis: homogeneity on T1/T2, presence of CE, mass effect, multifocality, leptomeningeal/subependymal seeding, contralateral. Details in Online Resource 1.

PWI

Dynamic susceptibility-weighted contrast-enhanced (DSC) PWI (Gadovist®, 0.1 cc/Kg, 5 ml/sec and fixed 3 cc pre-bolus. Maximum tumour cerebral blood volume (CBV) was obtained by identifying regions of maximum perfusion from colour maps. Three regions of interest (ROI) were placed on the highest colour areas of tumours. CBV values were normalized (rCBV) with an identical ROI positioned on the contralateral healthy white matter (CHWM). rCBVmax was obtained from the highest average value ROI.

DWI

Bi-commissural, single-shot echo-planar sequence, 3 orthogonal directions, b values 0–500–1000 s/mm2. Three ROIs were placed on different areas of the tumour on the basis of cMRI and ADC map appearance. ADC values were normalized (rADC) by those obtained from the CHWM.

MRS

Multivoxel, 2D Proton MRS, point resolved spectroscopic imaging (PRESS) (TR = 1200, TE = 135) fully enclosing the volume of interest. The following features were considered: NAA reduction, choline increase, Cho/Cr (Choline/Creatine) ratio, presence of lactate/lipids. The two highest Cho/Cr were considered for statistical analysis.

Details in Online Resource 2.

Imaging evaluation and histopathological diagnosis

Two experienced neuroradiologists independently reviewed pre-surgical cMRI and aMRI, separately evaluating each advanced technique in light of cMRI and providing tumour grade estimation (LGG vs HGG). In particular, the following parameters were considered suggestive of HGG: rCBV ≥1.5 (mean between literature cut-off values [4, 36]), Cho/Cr ≥2 [37], rADC <1 [3]. Furthermore, signal heterogeneity of T2/FLAIR, ADC and CBV maps within the whole tumour was taken into account. If consensus wasn’t reached, a third neuroradiologist reviewed the case.

Progression was initially assessed according to Macdonald [38]; after RANO criteria publication [39], all cases were re-evaluated accordingly.

Histopathological diagnosis was performed according to the WHO classification of brain tumours [40].

Statistical analysis

Means and standard deviations (sd) were calculated for continuous variables.

Descriptive statistics were obtained using Mann–Whitney U and X2 tests for continuous and dichotomous variables, respectively.

Sensitivity, specificity, negative and positive predictive values were calculated for cMRI and aMRI using histopathology as a reference standard. Ninety-five percent confidence intervals (95 % CI) were calculated according to the exact method (based on binomial distribution).

Inter-rater reliability between the two radiologists was determined by Cohen’s kappa coefficient.

Receiver Operating Characteristic (ROC) curves were estimated to determine rCBV, rADC, Cho/Cr values that achieve optimal sensitivity and specificity for differentiating: (i) LGG versus HGG; (ii) patients with progression versus progression-free patients; (iii) living patients versus dead patients.

PFS and OS were calculated from histopathological diagnosis until disease progression or death/last follow-up if censored. Kaplan–Meier analysis estimated PFS/OS. The log rank test assessed differences in PFS/OS among patients with various radiological or clinical parameters.

SPSS 22.0 for IBM (SPSS Inc., Chicago, IL, USA) was used for statistical analyses.

Results

Validation of criteria of suspected LGG based on cMRI

One-hundred-ninety-six patients were considered for inclusion. Mean interval between radiological examination and subsequent surgery was 18.6 days (range 0–45 days). A total of 89 patients (45.4 %) met the previously mentioned cMRI criteria for LGG (see "Patients"). Thirty (33.8 %) of the 89 suspected LGG were histologically diagnosed as HGG and 57 (64 %) as LGG. In one case histopathology was inconclusive because of inadequate specimen; one tumour was diagnosed as dysembryoplastic neuroepithelial tumour and excluded. Sensitivity of cMRI criteria for differentiation between LGG and HGG was 81.7 % (95 % CI 69.1–90.9) and specificity 100 %. Tumours not fulfilling cMRI criteria for suspected LGG were never histologically diagnosed as LGG. Fourteen (46.6 %) of the 30 tumours diagnosed as HGG at histopathology showed no CE at cMRI, whereas 12 (21.1 %) of the 57 tumours histologically diagnosed as LGG had faint CE.

Sixty-eight of the 89 tumours that met cMRI criteria for LGG were subjected to complete cMRI analysis and PWI, DWI, MRS. They also underwent subsequent statistical analysis, as described in the next paragraphs. The characteristics of these 68 patients (38 males; mean age 39.6 ± 12.6 years) are reported in Table 1.

Table 1 Clinical and radiological features of the 68 patients included in the study who met the cMRI criteria for suspected LGG
Full size table

Gross total resection was performed in 27 patients, partial resection in 41.

Conventional and advanced MRI

Of 42 histologically diagnosed LGG cases, 78.5 % were identified by complete cMRI analysis and 85.7 % by aMRI (69 % by PWI, 90.5 % by DWI, 64 % by MRS alone).

Of 26 histologically diagnosed HGG cases, 38.4 % were identified by complete cMRI analysis and 53.8 % by aMRI (53.8 % by PWI, 46.1 % by DWI, 57.7 % by MRS alone).

Diagnostic accuracy of cMRI and aMRI relative to pathology is shown in Table 2.

Table 2 Diagnostic accuracy of cMRI and aMRI related to histopathological diagnosis in 68 patients meeting the cMRI criteria for suspected LGG
Full size table

Diagnostic concordance rate of two blinded neuroradiologists in grading of HGG versus LGG was 95 % (inter-rater agreement K = 0.94).

Mean rCBV values were higher in HGG than in LGG (p = 0.01). Statistical analysis showed significant differences between grade II/III and IV gliomas (p = 0.005 and p = 0.003, respectively), not between grade II and III.

Mean rADC values were higher in LGG than in HGG. Statistical significance was reached when grade II/III and IV gliomas were compared (p = 0.008 and p = 0.05, respectively).

Cho/Cr ratios were similar in grade II and III gliomas and slightly higher in grade IV and in the nodular portions of grade I gliomas. NAA was decreased but present in all lesions. Lactate was detected in one of two grade I lesions, in five (12.5 %) grade II, in one (14.2 %) GBM, whereas lipids were found in two (10.5 %) grade III gliomas. Mean rCBV, rADC, Cho/Cr values are shown in Table 3.

Table 3 Mean values of aMRI parameters: PWI (rCBV), DWI (rADC), and MRS (Cho/Cr)
Full size table

Because LGG are a non-homogeneous group of tumours arising from astrocytic/oligodendroglial lineages, we compared aMRI data of the most representative LGG types (i.e., fibrillary astrocytoma (FA), oligodendroglioma (ODG), oligoastrocytoma (OA)). ODG had the highest rCBV (p = 0.04) and Cho/Cr (p = 0.05) and the lowest rADC values (not statistically significant). FA had the lowest rCBV and Cho/Cr and the highest rADC values. Intermediate values were found in OA (Online Resource 3).

Both cMRI and aMRI were able to diagnose pilocytic astrocytomas. All GBM but one were identified as HGG by aMRI. All grade II ODGs but one were correctly diagnosed using aMRI. cMRI and aMRI recognized 26.3 % and 42.8 % of grade III gliomas, respectively (Fig. 1).

Fig. 1
figure 1

Examples of cMRI and aMRI glioma assessment. In each row, from left to right, T2 and T1 images with contrast agent, ADC, CBV and MRS are shown, respectively. a Glioblastoma. cMRI suggests LGG. Low ADC and high rCBV were most pronounced in the antero-lateral portion of the tumour in a ring-like shape (arrow in the fourth panel). High Cho/Cr and lactate peaks indicate HGG. The PFS and OS were 8 and 26 months, respectively. b Glioblastoma. cMRI with high ADC and low rCBV that suggests LGG; only the high Cho/Cr indicates HGG. The PFS and OS were 33 and >49 months, respectively, and the patient in currently alive. c Oligodendroglioma. From left to right dubious HGG by cMRI due to heterogeneously and poorly defined signal intensity on T2 w.i. and faint CE. Low rADC in many regions of the tumour, very high homogeneous rCBV, and high Cho/Cr are typical findings of oligodendrogliomas. Both the PFS and OS were >84 months, and the patient is currently free from disease. d Oligodendroglioma. Heterogeneous abnormal signal intensity on T2-weighted images of cMRI without CE and with low rADC, low rCBV and high Cho/Cr. The atypical finding of this oligodendroglioma is low rCBV. The PFS and OS were 33 and >65 months, respectively, and the patient is currently alive. e Anaplastic astrocytoma. cMRI showing a well-defined and homogenous lesion without CE and with low rCBV that suggests LGG. Low ADC and high Cho/Cr and lactate peaks indicate HGG. The PFS and OS were 43 and 80 months, respectively. f Anaplastic astrocytoma. cMRI and aMRI suggest LGG. Both the PFS and OS were >57 months, and the patient is currently free from disease

Full size image

rCBV, rADC, rCho/Cr values of optimal sensitivity and specificity in differentiating between LGG and HGG were: 1.29 (sensitivity 61.5 %, specificity 54.7 %), 1.69 (sensitivity 53.8 %, specificity 43 %), 2.1 (sensitivity 53.8 %, specificity 64.3 %), respectively.

Clinical-radiological outcome

During follow-up (median 60 months, range 4–84), 47 of the 68 LGG patients showed radiological progression according to RANO [24] (Table 1). The patients histologically diagnosed with HGG were treated with radiochemotherapy (temozolomide) after surgery, whereas LGG patients received radiotherapy and/or chemotherapy at progression. Twenty-five patients died due to progression, three for other reasons (melanoma, suicide, unknown cause).

Median PFS and OS of all patients were 51 and 80 months, respectively. PFS at 6, 12, 36 months was 95, 92, 75 % for LGG patients and 88, 61, 45 %, for HGG patients, respectively (not significant). OS at 12, 36, 60 months was 92, 65, 53 % for HGG patients and 97, 88, 80 %, for LGG patients, respectively (p = 0.01).

Maximum rCBV, minimum rADC, maximum rCho/Cr values that achieved optimal sensitivity and specificity for survival correlations were 1.46 (sensitivity 50 %, specificity 59.9 %), 1.69 (sensitivity 72 %, specificity 72 %), and 2 (sensitivity 55.8 %, specificity 56 %), respectively.

Significantly longer OS was observed in patients with maximum rCBV <1.46 and minimum rADC >1.69 (80 vs 55 months, p = 0.01; 80 vs 51 months, p = 0.002, respectively, Fig 2). This result was also confirmed when cases were stratified according to pathology (LGG vs HGG. Figure 2) and presence/absence of CE in cMRI. A trend of longer PFS was observed for rCBV <1.46 and rADC >1.69.

Fig. 2
figure 2

Kaplan–Meier curves showing the OS of the cohort of 68 patients when rADC (a and b) and rCBV (c and d) cut-off values are 1.69 and 1.46, respectively. b and d show results after stratification by tumour grade

Full size image

Furthermore, the ability of aMRI to predict OS increased when an increasing number of aMRI techniques were considered. Longer OS was observed in patients with no abnormal parameters compared to patients with one or with two-or-more abnormal parameters: eighty-month survival was achieved by patients without abnormal parameters, not reached in patients with one abnormal parameter, 55-month in the others (p = 0.033). This was also confirmed when cases were stratified according to pathology (LGG vs HGG, p = 0.04).

No significant association between age, gender, presence of CE at cMRI and prognosis was recorded.

For survival of patients with mismatched radiologic versus histopathological diagnoses, see Online Resource 4.

Discussion

In a cohort of patients with suspected supratentorial glioma on cMRI who were candidates for surgery, a group of 68 that satisfied all three cMRI criteria for LGG (no necrosis, no relevant oedema, absent or faint CE) was subjected to aMRI. Compared to histology, cMRI underestimated the fraction of HGG. Sensitivity of cMRI in differentiating HGG from LGG, which ranges 55–83 % in the literature [3, 4], reached 78.5 % in the present study. aMRI increased sensitivity to 85.7 %, but specificity was 53.8 %, indicating that aMRI just slightly increases diagnostic accuracy compared to histopathology.

Interestingly, 37.5 % of gliomas that were suggested to be high grade by one aMRI technique were histologically diagnosed as HGG. This percentage increased to 61.5 % with two techniques (in accordance with previous multiparametric analysis [4, 41, 42]) and to 66.7 % with three techniques.

Anaplastic gliomas may mimic the appearance of LGG, lacking CE and having minimal oedema. The tumours in our population of radiologically suspected LGG that were histologically diagnosed as HGG were non-enhancing or faintly enhancing with no oedema and no necrosis. All histological GBM that were suspected of being LGG had MRI features of secondary GBM, lacking ring CE and cystic-necrotic areas [43]. ROI positioning (especially in partially resected LGG) might explain discrepancies with histopathology.

Though tumour-growth-rate was correlated with survival in LGG [44], as long as stratification of LGG is based on morphology, the clinical management of these tumours often remains challenging. Currently, important decisions impacting patient therapy should be supported by complementary technologies, such as aMRI and molecular [45] analysis to improve neuropathological diagnosis.

Previously published studies showed differences in aMRI parameters for LGG versus HGG: rCBV [4, 9, 10, 28, 46] and Cho/Cr [4, 12, 36, 47] were higher in HGG and rADC was lower [3, 9, 11, 13, 14].

A few authors set different cut-off values between LGG and HGG, ranging 0.94–3.34 [4, 11, 21, 36, 41, 42] for rCBV, 0.31–1.31 [3, 22–24, 36, 42, 48] for rADC, 1.3–2.04 [4, 20, 25, 26, 37, 42] for Cho/Cr. These thresholds were obtained mainly in populations that included a majority of GBM and few LGG patients, and overlapping values were observed within grade II and III lesions. In the largest cohort of unselected gliomas (120 HGG, 40 LGG), Law [4] identified a threshold 1.75 for rCBV and 1.56 for Cho/Cr. A cut-off of 1.07 for DWI was found in another group of 74 gliomas (37 GBM, 22 anaplastic astrocytomas, 15 LGG) [22].

In our population, cut-off values for HGG versus LGG were 1.29 for rCBV and 1.69 for rADC. The best cut-off value for Cho/Cr was 2.1, ensuring both sensitivity and specificity higher than 50 %.

PWI was considered predictive of tumour grade in previous studies because it shows diverse rCBV values in LGG, HGG [4, 9, 10, 36] and ODG [13]. In our cohort, the differences in rCBV significantly distinguished LGG from HGG and GBM from grade II-III gliomas, but not grade II from grade III gliomas. Mean rCBV values in our HGG patients were lower than those previously reported, which is probably due to patient selection, while rCBV in our LGG patients were similar to literature [9, 25].

In our cohort, DWI showed lower rADC in HGG than in LGG due to increased cellularity, one of the earliest features of anaplastic progression [12]. Lower ADC and higher CBV in HGG than in LGG patients were previously detected by Fan in 22 non-enhancing gliomas [9] and by Murakami and Khalid, who also published ADC cut-off values to distinguish grade II from grade III gliomas [23, 24]. In contrast, other authors [29, 48, 49] didn’t find significant differences in ADC values when comparing grade II versus III gliomas, which is consistent with our results. For more details, see Online Resource 5.

In our MRS analysis, we found that Cho/Cr ratio improved diagnostic accuracy compared to cMRI. Values were similar in grades I to III and slightly higher in grade IV, so that the cut-off value that optimally stratified LGG and HGG wasn’t statistically significant [41, 47]. However, previous studies reported that MRS is able to detect initial shifts to anaplasia sooner than other MRI techniques [20], identifying choline level as a useful parameter in determining glioma grade [4, 25, 26, 37, 42].

In agreement with the literature [14, 32, 50], our cohort of ODG had rCBV and Cho/Cr values higher than those of other LGG and similar to those of GBM because of high neovascularization and vessel hyperplasia [43]. OA appeared more heterogeneous, and recent developments in molecular genetics have cast doubt on its relevance as a distinct entity [45].

Apart from their use in stratification of LGG and in distinguishing LGG from HGG, radiological markers can be useful for survival prediction. A few recent studies correlated PWI, DWI and MRS measurements with time to progression or death [10, 31] and suggested cut-off values to distinguish early progressive from less aggressive, histologically diagnosed LGG [20, 25, 27, 29, 31–35] (Online Resource 6). However, many of these publications were limited to small cohorts of LGG or had short follow-up; moreover, their analysis mainly focused on CBV values and on PFS as an outcome parameter. A rCBV threshold of 1.75 was described to discriminate long from short survival in grade II gliomas [27, 31, 33]. Furthermore, a meta-analysis aimed at determining whether ADC values predict prognosis confirmed the predictive value of ADC on survival in HGG independently of histopathological grade [30].

In an outcome-based analysis, our rCBV and rADC cut-off values were able to stratify tumours, leading to different OS. Patients with radiological progression and shorter OS had lower rADC and higher rCBV values when compared to subjects with longer survival. In addition to predicting histological grade, aMRI was able to predict survival with an accuracy that improved as the number of aMRI parameters that meet the identified cutoffs increased, as significantly longer OS was observed in patients with no abnormal parameters compared to patients with one or two-or-more abnormal parameters. This trend is relatively independent of histological grade, which is in agreement with recent data that defines lower-grade gliomas [45].

In conclusion, our results suggest that aMRI slightly improves the accuracy of cMRI in distinguishing LGG from HGG; in particular, aMRI allowed the detection of malignant features in GBM that appear as LGG on cMRI and confirmed specific characteristics of ODG helping their differentiation from other gliomas. Notably, DWI and PWI can be of value in predicting OS in lower-grade gliomas independent of their grading. These observations support the integration of imaging with pathology and genetics to provide more robust classification of lower-grade gliomas and strengthen the rationale for treatment decisions.