Grading Meningiomas Utilizing Multiparametric MRI with Inclusion of Susceptibility Weighted Imaging and Quantitative Susceptibility Mapping

Abstract

Background and purpose:

The ability to predict high grade meningioma preoperatively is important for clinical surgical planning. The purpose of this study is to evaluate the performance of comprehensive multiparametric MRI, including susceptibility weighted imaging (SWI) and quantitative susceptibility mapping (QSM) in predicting high grade meningioma both qualitatively and quantitatively.

Methods:

92 low grade and 37 higher grade meningiomas in 129 patients were included in this study. Morphological characteristics, quantitative histogram analysis of QSM and ADC images, and tumor size were evaluated to predict high grade meningioma using univariate and multivariate analyses. Receiver operating characteristic (ROC) analyses were performed on the morphological characteristics. Associations between Ki-67 proliferative index (PI) and quantitative parameters were calculated using Pearson correlation analyses.

Results:

For predicting high grade meningiomas, the best predictive model in multivariate logistic regression analyses included calcification (β=0.874, p=0.110), peritumoral edema (β=0.554, p=0.042), tumor border (β=0.862, p=0.024), tumor location (β=0.545, p=0.039) for morphological characteristics, and tumor size (β=4×10⁻⁵, p=0.004), QSM kurtosis (β=−5×10⁻³, p=0.058), QSM entropy (β=−0.067, p=0.054), maximum ADC (β=−1.6×10⁻³, p=0.003), ADC kurtosis (β=−0.013, p=0.014) for quantitative characteristics. ROC analyses on morphological characteristics resulted in an area under the curve (AUC) of 0.71 (0.61–0.81) for a combination of them. There were significant correlations between Ki-67 PI and mean ADC (r=−0.277, p=0.031), 25^th percentile of ADC (r=−0.275, p=0.032), and 50^th percentile of ADC (r =−0.268, p=0.037).

Conclusions:

Although SWI and QSM did not improve differentiation between low and high grade meningiomas, combining morphological characteristics and quantitative metrics can help predict high grade meningioma.

Introduction

Meningiomas are the most frequently diagnosed primary brain tumors in adults. WHO Grade I meningiomas are the most common and are considered benign with a low risk of recurrence. Atypical meningiomas (WHO Grade II), which account for 20–35% of all meningiomas, have recurrence rates up to 50% and 10-year survival of less than 80%[1]. Anaplastic meningiomas (WHO Grade III) are uncommon, accounting for only 1–3% of all meningiomas with a recurrence rate of up to 94% and are associated with poor overall survival rates[2]. The ability to differentiate between low grade and high grade meningiomas prior to treatment has important clinical ramifications. Patients with more aggressive (WHO grade II/III) meningiomas could benefit from early and complete resection [3, 4]. Preoperative tumor grade information is beneficial for neurosurgeons, since they frequently have to weigh the risks and benefits of resecting tumor-adjacent tissue that could potentially be involved with microscopic tumor[3].

In the last decade, several imaging studies[5–13] have attempted to differentiate among meningiomas of various grades, but the results remain controversial. The development of susceptibility weighted images (SWI) allows detection of the presence of tissue susceptibility and improves contrast in visualization of the venous vasculature, hemorrhage and calcification within tumors[14]. While SWI generates contrast based on magnitude and filtered phase images, quantitative susceptibility mapping (QSM), by computing a dipole deconvolution, further enables quantitative investigation of local tissue susceptibility, also allows distinguishing calcification which is hypointense due to its diamagnetic properties from hyperintense paramagnetic product of hemorrhage. Hwang et al[15] demonstrated that absence of calcification is one of independent risk factors for a high proliferative potential and most strongly associated with high-grade histopathology. In a recent study[16], a strong association between meningiomas exhibiting necrosis and/or hemorrhage and histologic grade was observed. Since the internal vascular structures, calcification, necrosis and hemorrhage can be visualized conspicuously with SWI and QSM, we hypothesize that these imaging techniques may be useful in differentiating grades of meningiomas.

In this study, we applied a comprehensive approach using (1) qualitative morphological characteristics, evaluated on conventional T1w, T1w+Gd, T2w FLAIR images, SWI, QSM, and apparent diffusion coefficient (ADC) maps, (2) quantitative histogram analysis of QSM and ADC images, and (3) tumor size to discriminate between low-and high grade meningiomas. In addition, correlations between Ki67 PI and quantitative parameters were assessed.

Material and methods

Patient selection

This retrospective study was approved by our local institutional review board. 192 consecutive patients with meningiomas who underwent MR imaging in our institution from January 2015 to December 2017 were identified. 129 of these were included in our study by meeting the following inclusion criteria: (1) pathologically-proven meningioma and (2) a preoperative MRI study that included QSM, SWI, ADC, T2w FLAIR, T1w, and T1w+Gd sequences. 63 patients were excluded either because they lacked MR images for analysis (n=41) or the available images were of poor quality (n=22). The Ki67 proliferative index (PI) for each patient was also recorded.

MRI protocol and image processing

All brain MRIs were performed on our clinical scanners (GE Signa HDxt 3.0T, GE Discovery MR750w 3.0T). Image parameters for multi echo GRE scans were: field of view = 24cm, TR = 49–70ms, TE1/ΔTE = 5.3–5.8/5.9–10.3ms, number of TEs = 9–12, acquisition matrix= 416×320, readout bandwidth = 195–244Hz/pixel, slice thickness = 3mm, flip angle = 15–20°. The SWI and QSM were generated from same GRE images. QSM was reconstructed with a fully automated zero-referenced Morphology Enabled Dipole Inversion (MEDI+0) method[17] that uses the ventricular cerebrospinal fluid (CSF) as a zero reference. ADC maps were reconstructed from the DWI sequence obtained with b-value 0 and 1000s/mm².

Semi-quantitative evaluation of the morphological characteristics

Two experienced neuroradiologists (SZ and IK, with 7 and 23 years of experience, respectively), who were blinded to the histopathological findings, reviewed all the images independently for vascularity index, hemorrhage, calcification, peritumoral edema, tumor borders, tumor location, ADC appearance, with discrepancies resolved by consensus. The scoring for each morphological characteristic was as follows:

Vascularity index, evaluated on SWI based on the percentage of vascularization within the tumor (1: none or minimal, 2: moderate; up to half of tumor showing vascularization, 3: extensive; more than half of tumor);

Hemorrhage, evaluated mainly on the QSM images based on the percentage of hemorrhage within the tumor (1=none (0%), 2= <5%, 3= 6–33%, 4= 34–67%, 5=>67%);

Calcification was identified as hypointense both on QSM and SWI[18, 19], and the examples are shown in Figs. 1 and 2 (1= more than half of tumor volume, 2= up to half of tumor volume, 3=none or minimal);

Fig.1.

WHO grade II meningioma was located along the high right frontal convexity (score, 3), had smooth borders (1), extensive perilesional edema (3), and showed isointense compared to normal-appearing white matter on ADC map (2). A focal coarse intratumoral calcification (3) was hypointense on both QSM and SWI images (arrows) and hyperdense on CT (arrowhead). No vascularization (1) and hemorrhage (1) was found within the lesion.

Fig.2.

WHO grade I meningioma was located along the right frontal parafalcine (score, 2), had smooth borders (1), and isointense to normal-appearing white matter on ADC map (2). Diffuse calcification (1) can be seen hypointense on both QSM and SWI images (arrows), and moderate vascular structure (2) on SWI image. No peritumoral edema (1) and hemorrhage (1) was found within the tumor.

Peritumoral edema (1: none or minimal, 2: moderate, 3: extensive with mass effect);

Tumor border (1: smooth border, 2: moderately irregular, 3: wildly irregular);

Tumor location (1: skull base, 2: parafalcine/parasagittal, 3: convexity);

ADC signal was evaluated visually by referencing the normal-appearing white matter (NAMW) [6] (1: hyperintense compared to NAWM, 2: isointense to NAWM, 3: focal areas hypointense to NAWM, 4: most of the lesion is hypointense to NAWM).

Quantitative histogram analysis of QSM, ADC and tumor size

For the QSM lesion segmentation, tumors were manually delineated on each image slice of the entire tumor on co-registered T1w+Gd images using ITKsnap (http://www.itksnap.org). Any associated trapped cyst or peritumoral edema around the tumor was excluded. Tumor size was calculated based on the segmented lesion mask. ADC maps were segmented separately on raw ADC images in order to mitigate geometric distortion.

Histogram analysis of QSM and ADC was performed using Matlab 2017a (IBM), with following histogram parameters: mean, standard deviation, minimum, maximum, kurtosis, skewness, entropy, 10^th percentile, 25^th percentile, 50^th percentile, 75^th percentile, 90^th percentile.

Statistical analysis

All data were analyzed using R statistical program (version 3.5.1), and statistical significance was set at p<0.05. Univariate and multivariate analyses were performed on morphological characteristics and quantitative histogram parameters to predict high grade meningiomas. Multiple logistic regression analyses with variable selection ways including the Akaike information criterion, deviance and Wald test were used to develop the best multivariate model for predicting high grade meningiomas. Receiver operating characteristic (ROC) analyses were performed on the 6 morphological characteristics separately and on a combination of them for predicting high grade meningioma. The associations between Ki67 PI and quantitative parameters were calculated using Pearson correlation analyses.

Results

129 patients (40 male, 89 female) diagnosed with a meningioma who met the inclusion criteria were included in this study: 92 (71.3%) low grade (WHO grade I), 37 (28.7%) high grade, of whom 32 were atypical and 5 were anaplastic. Among them, 33 cases showed Ki67 PI<5%, 32 cases showed Ki-67 PI ≥5%, and 64 cases were not reported. (Table 1)

Table 1.

Patient demographics and radiological characteristics

	Low grade	High grade	p value
No. of patients	92	37	N/A
Gender (M/F)	24/68	16/21	0.057
Age (years, mean ± SD)	63.61±12.13	63.14±17.02	0.873
Ki67 PI (<5%/>5%/not reported)	31/6/55	2/26/9	N/A
Morphological characteristics
Vascularity index
1: none or minimal	30	7	0.280
2: moderate	52	26
3: extensive	10	4
Hemorrhage
1=none (0%)	79	31	0.438
2= <5%	9	4
3= 6–33%	4	1
4= 34–67%	0	1
5= >67%	0	0
Calcification
3=none or minimal	71	35	0.019
2= up to half of tumor	17	0
1= more than half of tumor	4	2
Peritumoral edema
1: none or minimal	59	12	0.005
2: moderate	20	15
3: extensive	13	10
Tumor border
1: smooth border	73	21	0.006
2: moderately irregular	18	12
3: wildly irregular	1	4
Tumor location
1: skull base	34	5	0.021
2: parafalcine/parasagittal	15	11
3: convexity	43	21
ADC feature
1: hyperintense than NAWM	23	7	0.405
2: isointense to NAWM	59	22
3: focal areas hypointense than NAWM	6	4
4: most of the lesion hypointense than NAWM	4	4

N/A, not applicable.

The scores of each individual morphological characteristic are summarized in Table 1. According to the univariate analysis, there were significant differences between low and high grade meningiomas in peritumoral edema (β=0.732, p=0.004) and tumor border (β=1.032, p=0.004) (supplemental Table S1). For quantitative variables, tumor size (β=2×10⁻⁵, p=0.005), histogram analysis of maximum ADC (β=−8×10⁻⁴, p=0.05), and ADC kurtosis (β=−0.088, p=0.038) showed significantly differences.

In multivariate logistic regression analyses, the final model for predicting high grade meningioma included calcification (β=0.874, p=0.110), peritumoral edema (β=0.554, p=0.042), tumor border (β=0.862, p=0.024), and tumor location (β=0.545, p=0.039) (supplemental Table S2). For quantitative variables, the final model included tumor size (β=4×10⁻⁵, p=0.004), QSM kurtosis (β=−5×10⁻³, p=0.058), QSM entropy (β=−0.067, p=0.054), maximum ADC (β=−1.6×10⁻³, p=0.003), and ADC kurtosis (β=−0.013, p=0.014) (supplemental Table S3).

ROC analyses on morphological characteristic resulted in an area under the curve (AUC) of 0.56 (0.45–0.66) for vascularity index, 0.52 (0.40–0.63) for hemorrhage, 0.58 (0.47–0.68) for calcification, 0.66 (0.55–0.76) for peritumoral edema, 0.62 (0.50–0.73) for tumor border, 0.60 (0.50–0.70) for tumor location, 0.58 (0.47–0.69) for ADC, and 0.71 (0.61–0.81) for a combination of them.

For 65 patients with available Ki67 PI, there were significant correlations between Ki67 PI and mean ADC (r=−0.277, p=0.031), 25^th percentile of ADC (r=−0.275, p=0.032), and 50^th percentile of ADC (r =−0.268, p=0.037) (Fig.3), but not with the other quantitative parameters.

Fig.3.

Correlations between Ki-67 proliferative index (PI) and mean ADC, 25^th percentile and 50^th percentile of ADC.

Discussion

Our study demonstrated that increased peritumoral edema, irregular tumor border and tumor location along the cerebral convexity are important predictors of high grade meningioma, consistent with the prior studies[3, 9, 19–22]. Controversially, some reports[23–25] indicated that peritumoral edema may depend on the expression of aquaporin-4 (AQP4), a small membrane protein with a crucial role in water transport and maintenance of fluid balance, but not on tumor grade, tumor volume and Ki67 expression. However, these studies either had rather small sample size of high grade meningioma[23] or were focused on low grade alone[24, 25] and further investigations are needed for conforming their results.

To our knowledge, this is the first study using QSM and SWI in the prediction of high-grade meningiomas. QSM and SWI are the best MRI techniques to identify calcification, hemorrhage and vascular structures within the tumor. In this study, we found a higher incidence of calcification in low grade meningiomas (23% in low grade vs 5% in high grade), suggesting benignity. A similar result was found in another study which concluded that absence of calcification was one of the imaging features most strongly associated with high-grade histopathology[15]. While Oiao et al[26] reported that high grade meningiomas had more perfusion[26], we found no significant difference for vascularization features between low and high grade meningiomas, signifying that the quantity of vascular structures within the tumor was less important for predicting high grade meningiomas. The semi-quantitative morphologic characteristics and quantitative histogram features of both SWI and QSM did not show a significant difference between low and high grades contradicting our hypothesis. One of the reasons may be related to extreme variability of the 15 histological subtypes of meningiomas in the WHO classification. Depending on subtype of meningioma, vascular, hemorrhagic, calcific and fibrotic components of tumor may vary and show high heterogeneity regardless of its grade. For example, angiomatous meningioma is characterized by predominance of hypervascularity, in spite of corresponding to WHO grade I. These wide varieties of pathological changes and their reflective imaging features may cause preoperative unpredictability in the grading of meningiomas.

Conventional MRI usually have limited usage in-vivo differentiation of low grade and high grade meningiomas, while combining with quantitative approaches such as histogram analysis enhance the prediction of growth kinetics and prognosis in meningiomas [12]. Recently, the utility of advanced MR imaging metrics, including ADC maps, in predicting high grade meningioma has been studied with mixed results[5, 10, 11, 27–29]. Some studies revealed that decreased ADC values due to significantly reduced water diffusibility in the extracellular compartment, predominantly as a consequence of more densely packed tumor cells and increased fibrosis or widespread collagen formation can predict the malignancy of meningiomas[6, 11], while others contradict that claim[7, 10, 29, 30]. In our study, the qualitative ADC hypointensity did not show a significant difference for predicting high grade meningioma. For the quantitative histogram parameters, maximum ADC and ADC kurtosis which probably reflects a higher degree of microstructural complexity within the tumor, significantly correlated with meningioma grade. Our results differed from the studies which reported that the entire ADC profile (mean, median, and all ADC percentiles) and diffusional kurtosis imaging metrics were significantly lower in higher grade meningioma[6, 11].Other studies found that ADC alone cannot distinguish meningioma grades, in which ADC values obtained by manually drawn regions of interest instead of whole tumor or analyzing only average ADC values[7, 10, 29, 30]. In addition to the methodological differences, these conflicting results may also be due to inherent tumor heterogeneity, seen in higher grade meningiomas containing micro-cystic, necrotic changes which causes high ADC. Calcification and hemorrhage can also affect ADC values regardless of tumor grade. Therefore, caution is recommended when using ADC analysis to predict high grade meningiomas.

We also found that tumor size is an important variable to predict high grade meningioma in accordance with prior studies[31, 32]. 89% of high-grade tumors (33/37) were larger than 20 cm³, whereas only 49% of low grade meningiomas (45/92) were larger than 20 cm³. Thus, malignant meningiomas typically have larger size, while low grade meningiomas can be any size.

Although Ki67 proliferation index is not considered one of the clear criterion for atypical or anaplastic meningiomas in the current WHO classification[33], high Ki67 PI usually indicates higher grade of meningioma[34, 35]. In our study, the mean ADC value, 25^th percentile and 50^th percentile of ADC, showed significant correlations with the proliferation index. While the correlations were found a little weaker than previous studies[5, 6, 11], which used a different algorithm for ADC calculation, ADC can help as an additional predictor to quantify cellular proliferation of meningiomas.

There are a few limitations in our study. First, this is a retrospective study; future research using morphological characteristics on MRI prospectively for predicting high grade meningioma would be necessary to validate our results. Second, the scores of qualitative morphological characteristics were subjective; there may exist some variability among different observers.

In conclusion, although the inclusion of SWI and QSM did not improve differentiation between low and high grade meningiomas, we found that peritumoral edema, tumor border irregularities, tumor size, and some ADC histogram parameters are helpful in predicting high grade meningioma, which can change clinical management.

Highlights.

• A precise preoperative prediction of high grade meningioma on MRI images brings benefits for clinical surgical planning.

• QSM and SWI are excellent MRI techniques to identify calcification, hemorrhage and vascular structures within the tumor.

• Combining morphological characteristics and quantitative metrics can help predict high grade meningioma.

Abbreviations:

SWI

susceptibility weighted imaging

QSM

quantitative susceptibility mapping

ADC

apparent diffusion coefficient