Multisequence magnetic resonance imaging habitat analysis for pre-operative meningioma grade prediction

Introduction

Meningiomas are the most common primary intracranial and central nervous system tumor (1). Histological meningioma grading and adequacy of definitive treatment currently guide prognosis and overall patient outcome. While the majority of meningiomas are classified as low-grade [World Health Organization (WHO) grade 1] (2,3), and can be effectively treated with surgery or radiotherapy with minimal side effects (4), high-grade (WHO grades 2 and 3) meningiomas typically necessitate a combination of both therapies or more aggressive and intensive treatment planning (4,5). Magnetic resonance imaging (MRI) has been widely used in the diagnosis of meningiomas, and recent studies have suggested that MRI features of heterogeneous enhancement, indistinct tumor-brain interface, necrosis, peritumoral oedema, and hyperintensity on diffusion-weighted imaging (DWI) with decreased apparent diffusion coefficient (ADC) values to be factors predictive of advanced histopathological grade (6-9). In addition, MRI features have been shown to correlate well with structural and cellular features of tumors at the microscopic and macroscopic level (10), while sampling from different subregions within a single tumor could yield distinct information for grading (11). Therefore, evaluation of MRI features across different subregions of tumors may be able to reflect and provide an interpretation of the underlying intrinsic tumor heterogeneity and identify important markers in non-invasive tumor grading.

Habitat analysis is a computational technique that aims to spatially delineate imaging features reflective of the tumor microenvironment. It involves segmentation of the tumor into distinct “habitats” based on differences in imaging characteristics computed at the voxel level. Habitat analysis can highlight the spatial distribution and local patterns of tumor heterogeneity across the tumor volume. To date, the technique has been applied to computed tomography (CT) and MRI [including conventional sequences and parametric maps generated from dynamic susceptibility contrast (DSC) MRI] (12,13), and has been explored for the prediction of tumor aggressiveness and clinical or molecular tumor signatures. For instance, in patients with high-grade glioma, Bailo et al. (12) found a high reproducibility of habitat maps and correlations with aggressive histological tumor features of high cellularity and neovascularization. Recent studies have additionally attempted to relate tumor habitats to mutational status in glioma, reporting the significance of relevant habitats that correlate with cellular pathway alterations such as upregulation of STAT-1 and natural killer cell activity (14). However to date, heterogeneity in meningioma has not been well elucidated, and only prior one study has previously explored habitat analysis in meningioma (15).

This study sought to explore the use of habitat analysis with filtered multi-sequence MR images in evaluation of meningioma grade. Specifically, we sought to compare habitat measures between low-grade and high-grade meningiomas and combine selected habitat measures together with clinical factors to develop an integrated prediction model for meningioma grade prediction. While advanced MRI sequences have been included in prior habitat analysis studies for glioma (12,16,17), we propose the use of conventional MRI (14,15), due to its availability and standardized utilization across institutions for tumor diagnosis and follow up, increasing the relevance and potential broader applicability of our method. We evaluate the performance of our model with combined habitat and clinical data for meningioma grade prediction. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1041/rc).

Methods

Patient population

This retrospective study was approved by the Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethic Committee (No. 2022.382) and individual patient consent for this retrospective analysis was waived. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments (18). Sixty consecutive patients aged between 18 to 65 years, with 60 pathologically proven meningiomas who underwent preoperative MRI and surgical resection at the Prince of Wales Hospital, Shatin, Hong Kong SAR, China, between 1^st May 2016 and 1^st May 2023 were retrospectively recruited. The exclusion criteria were lack of histopathology details from resection; pre-operative MRI performed more than six months prior to surgical resection, to limit potential mismatch between imaging findings and histopathology at resection; and recurrent or partially treated meningiomas. Patients were divided into low-grade (WHO grade 1, n=30) and high-grade (WHO grade 2/3, n=30) cohorts based on final surgical pathology according to the updated 2021 WHO classification system (5).

MRI protocol and image processing

Non-enhanced [T2-weighted (T2W)] and contrast-enhanced [T1-weighted with contrast (T1C)] sequences were included for analysis. The details of the sequence imaging parameters are shown in Table S1. To align the information from the two sequence MRIs, image registration was performed by registering T1C to T2W images using ITK-SNAP (V3.8.0) (19). The registration was implemented by a multi-resolution schedule (coarsest level, 4; finest level, 2) based on a rigid transformation model with mutual information metric and linear interpolation in ITK-SNAP (19). Subsequently, the tumor core was manually delineated on each slice of the registered T1C images by an experienced neuroradiologist (T.Y.S., more than 10 years of experience), including areas of enhancing tumor and tumor necrosis without perilesional oedema. Both T2W and T1C images were examined to delineate tumor margins in the data and to ensure that non-tumoral structures, such as sulci, subarachnoid cisterns, and vascular structures were appropriately excluded. The aforementioned segmentations thus defined the tumor volumes of interest (VOIs) for subsequent analyses.

Subsequently, the T2W and T1C images were then filtered using a three-dimensional (3D) Laplacian of Gaussian (LoG) filter with SimpleITK (V2.2.1) (19), which enhances boundaries between tissue types by emphasizing spatial intensity gradients. Furthermore, the filter assisted in reducing the impact of noise and scanner-related variations. The comparison between the original images and the filtered images is presented in Figures S1-S3. The width of the LoG filter is determined by the value of sigma (σ), with different values of σ resulting in the emphasis of varying intensity gradients. A LoG filter with σ=3 was selected following a series of tests conducted with a range of values (σ=1–5). The optimal balance between noise reduction and feature enhancement was achieved with σ=3 in our dataset. The filtered tumor regions were then extracted in the tumor VOIs. Otsu’s thresholding algorithm (20) was then applied to the filtered tumor regions to determine the optimal threshold for cluster generation from the T2W and T1C images respectively, based on the intensity distribution. For cluster generation, voxels with values above or below the threshold were extracted from the tumor VOI for each of the T2W and T1C images, and clustered within Otsu’s algorithm (20) to high- and low-intensity clusters. There were four clusters generated from the filtered MRI sequences, namely the T2W high-intensity cluster, the T2W low-intensity cluster, the T1C high-intensity cluster, and the T1C low-intensity cluster. Subsequently, the four clusters were combined one-to-one to take all possible intersection combinations as a different “habitat”, with two combinations containing clusters from the same MRI sequence (T2W high-intensity cluster ∩ T2W low-intensity cluster and T1C high-intensity cluster ∩ T1C low-intensity cluster) excluded (Figure 1). Habitat I represents subregions of high T2W and high T1C intensity, Habitat II represents subregions of high T2W and low T1C intensity, Habitat III represents subregions of low T2W and high T1C intensity, and Habitat IV represents subregions of low T2W and low T1C intensity.

Figure 1 Example T2W, T2W with habitats, T1C, and T1C with habitats images from a high-grade meningioma. Cluster results were derived using the automatic Otsu’s thresholding algorithm method and cluster intersection to generate habitats. The arrows refer to habitats generated by intersecting the two clusters from T2W and T1C images. The different colors of the tumor region indicate different habitats. Red areas indicate Habitat I (T2W hyperintensity and T1C hyperintensity); blue areas indicate Habitat II (T2W hyperintensity and T1C hypointensity); green areas indicate Habitat III (T2W hypointensity and T1C hyperintensity); yellow areas indicate Habitat IV (T2W hypointensity and T1C hypointensity). T1C, T1-weighted with contrast; T2W, T2-weighted.

Heterogeneity quantification

Three quantitative metrics, namely relative mean, relative standard deviation (SD), and entropy, were used to characterize the tumor habitat information. The relative mean and SD were obtained by regularizing to the whole tumor as follows:

RelativeX=|X(eachhabitat)−X(wholetumour)|[1]

where X indicates the mean or SD. Entropy was calculated as:

Entropy=−Sum(pi×log2(pi))[2]

where p_i indicates the number of intensities in the i-th bin; Sum(⋅) indicates the sum of the entropy from all bins. A total of twenty-four measures were extracted by calculating the three quantitative metrics (relative mean, SD, entropy) across the four habitats derived from the two MRI sequences (T1C and T2W) for each case (3 × 4 × 2 = 24).

Statistical analysis

Statistical analysis was conducted using Medcalc (V20.1) (21). The Shapiro-Wilk test was used for testing normality. The Kruskal-Wallis test was performed to test for differences in the calculated metrics between different habitats within individual patients due to at least one of the habitat measures violating normality. To ascertain whether there were differences between low-grade and high-grade meningiomas, metrics were compared using the independent T-tests and Mann-Whitney tests. Then, we applied the Holm-Bonferroni correction on the comparison results of each metric to control the false discovery rate. Statistically significant measures in the grade comparisons, along with clinical factors (age and gender), were then entered into a stepwise multiple regression analysis for meningioma grade prediction. The stepwise method examines the contribution and significance of each variable to identify and potentially eliminate measures that render the model non-significant. A P value <0.05 was considered statistically significant. We performed bootstrapping 1,000 times to assess model stability. The area under the curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE) were calculated to evaluate the performance of the regression models.

Results

Patient characteristics

A total of 60 patients were included, consisting of 30 WHO grade 1, 28 WHO grade 2, and 2 WHO grade 3 meningiomas. There were no significant differences in age, gender and gap time between the MRI scan and the surgery in patients with low-grade and high-grade meningiomas. The characteristics of patients are summarized in Table 1.

Comparison of quantitative metrics between habitats within individual patients

A statistically significant difference (P<0.01) in the relative means of T2W and T1C intensities between different habitats was observed within patients with both low-grade and high-grade meningiomas, as illustrated in Table 2. No significant differences were identified in the relative SD of T2W and T1C intensity between habitats (P≥0.65), with the exception of the low-grade T1C group (P=0.02). Similarly, no significant differences in entropy of T2W and T1C intensities between habitats were observed (P≥0.42).

Comparison of quantitative metrics between low-grade and high-grade groups

Comparing between tumor grades, high-grade meningiomas showed significantly higher relative mean and SD of T2W and T1C intensities in all habitats compared to low-grade tumors (P≤0.03). This finding is illustrated in Table 3 and Figure 2A,2B. Furthermore, the entropy of T1C intensities was found to be significantly higher (P≤0.01) in all habitats of high-grade compared to low-grade tumors (Table 3, Figure 2C and Figure 3).

Figure 2 Distribution of quantitative metrics (base 10 logarithm of relative mean, base 10 logarithm of relative SD, and entropy) of T2W and T1C intensities in habitats. The lower and upper edges of each boxplot are the first and third quartiles, the mid line shows the median, and X indicates the average of the distribution. The dots out of the box indicate outliers. *, P<0.05; **, P<0.01; ***, P<0.001; NS, not significant, P>0.05. SD, standard deviation; T1C, T1-weighted with contrast; T2W, T2-weighted.

Figure 3 Example showing the entropy of T1C intensity within the whole tumour and each of the four habitats in high- and low-grade tumors. T1C, T1-weighted with contrast.

Prediction models for meningioma grading

The relative mean of intensities in Habitat III on T1C (H3Mean) and the entropy of intensities in Habitat IV on T1C (H4En) were selected by the stepwise method of multiple regression for meningiomas grading. The formula for predicting meningioma grade was derived as follows: Grade =0.00026 × H3Mean + 0.34 × H4En−1.42 (Table 4). The variance inflation factor of both selected measures was 1.

The performance (AUC, ACC, SEN, and SPE) of the integrated multiple regression model incorporating both selected habitat measures (H3Mean and H4En) and clinical factors (age and gender) (0.84, 0.78, 0.73, and 0.83) was generally higher than that of the model using selected habitat measures alone (0.82, 0.75, 0.80, 0.70). The AUC (ACC, SEN, SPE) of the univariate regressions of H3Mean and H4En were 0.77 (0.77, 0.67, 0.87) and 0.75 (0.75, 0.93, 0.43), respectively (Table 5 and Figure 4).

Figure 4 ROC curves of the variables included in the model. H3Mean and H4En were selected through stepwise regression. Cli fac, clinical factors; H3Mean, relative mean of intensities in Habitat III on T1C; H4En, entropy of intensities in Habitat IV on T1C; ROC, receiver operating characteristic; T1C, T1-weighted with contrast.

Discussion

This study explores the application of habitat analysis on LoG-filtered multi-sequence MRI for characterization of meningiomas and prediction of tumor grade. The application of LoG filtering enabled the identification of distinct habitats based on their spatial intensity patterns. The quantitative evaluation revealed a clear distinction between different habitats within individual patients, and significant differences in quantitative metrics results in habitats of different grades.

Significant differences in the relative mean of T2W and T1C intensities between habitats within individual patients were observed in both low-grade and high-grade meningiomas, while the entropy and relative SD of habitats were mostly not significantly different, suggesting that each habitat is distinctly delineated and well-defined (as shown in Table 2). Although the relative SD of T1C intensities between habitats was found to be significantly different in the low-grade meningioma group, no significant entropy difference was observed in this group. The significant difference in relative SD might be attributed to outlier intensities within the habitat, as entropy is typically less susceptible to the influence of outliers than relative SD.

Differences in both T2W and T1C sequences in habitats may reflect the underlying characteristics and tissue composition of the tumor. Habitat I, characterized by hyperintensity on both T2W and T1C sequences, may suggest the presence of less cellular tumor with a higher water or foamy interstitial cell (angioblastic) content (22,23). Habitat II, characterized by high T2W and low T1C intensity, may represent the presence of necrosis, oedema, and/or cystic formation in this subregion (24,25). Habitat III, distinguished by T2W hypointensity and T1C hyperintensity, may indicate the presence of high cellularity or firmness with more collagen component (23). Habitat IV, characterized by T2W and T1C hypointensity, may indicate the presence of hemosiderin, calcifications or fibrosis and increased tumor stroma (24).

Notably, high-grade meningiomas exhibited greater heterogeneity in all habitats in comparison to low-grade tumors. This was evidenced by the presence of significant differences in the relative mean of T2W and T1C intensities between tumor grades. Furthermore, an examination of the entropy and relative SD in each habitat revealed that high-grade tumor T1C images exhibited greater dispersion. However, an absence of significant entropy differences in T2W habitats between grades was noted, which may suggest that T2W signal changes may not fully capture the complexity differences between tumor grades compared to T1C.

The H3Mean and H4En were selected as significant measures in the stepwise multiple regression analysis, demonstrating statistically significant correlations with meningioma grade (P<0.01). Although high-grade and low-grade meningiomas showed no significant differences in age and gender, these factors were incorporated into the model due to their biological relevance and frequent association with tumour behavior in the literature. Specifically, age is a commonly considered prognostic factor in neuro-oncology, and gender differences in high-grade meningiomas have been previously reported (24,26). Compared to the multiple regression model that exclusively utilized the selected habitat measures alone, the incorporation of clinical factors into the model enhanced the AUC and ACC although through a trade-off between sensitivity and specificity.

Our integrated model achieved an AUC of 0.84 [95% bootstrap confidence interval (CI): 0.72–0.92], which was comparably higher than that of a prior study utilizing habitat analysis in meningioma grading (AUC of 0.838 for the training set and 0.73 for the test set). The AUC results of the prior study were obtained from naïve split on the dataset. The lower bound of our confidence interval of the final model is close to the test set AUC in the prior study, while the upper bound is much higher than the prior study’s training set AUC which may indicate overfitting in the previous model. Our model using selected habitat measures alone also achieved an AUC of 0.82 (95% bootstrap CI: 0.70–0.90), which was comparable to the prior study. Differences in performance between the previous study and our study may be a result of differences in our habitat generation method, which may translate to more accurate tumor grading for treatment planning.

In this study, we demonstrate utility of a habitat analysis framework for meningioma for probing tumor characteristics and tumor heterogeneity based on LoG-filtered conventional multisequence MRIs. Habitat mapping successfully delineated regional variations within meningiomas and discerned significant differences between low and high-grade tumors. High-grade meningiomas demonstrated increased heterogeneity between constituent habitats compared to lower-grade tumors, which may mirror increased variable biological changes. Further, the integrated model combining habitat and clinical features achieved a high performance in meningioma grading. The ability to noninvasively distinguish increased intra-tumor variability may provide additional computational information to aid tumor grading and risk assessment. This study has several limitations, including its retrospective design and relatively small sample sizes from a single center. We also acknowledge that we did not evaluate higher-order or deep learning based features within habitats, which may offer additional insights into tumour heterogeneity. Furthermore, the majority of high-grade tumours in this study were WHO grade 2 tumours, reflecting the typical distribution of cases encountered in practice, as grade 3 tumours typically constitute only a small minority of all cases. Therefore, the observed findings may primarily reflect characteristics of grade 2 tumours. It may need to be explored whether grade 3 meningiomas, due to their higher aggressiveness and histopathological differences, may exhibit different, or more pronounced habitat-related variations than observed in this study.

Conclusions

In conclusion, this study demonstrates that habitat analysis using conventional multisequence MRI offers a promising non-invasive approach to capture tumor heterogeneity for meningioma grading. Future work should focus on validating these findings in larger independent cohorts and on further understanding the biological or histopathological mechanisms correlating with different imaging habitats.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1041/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1041/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1041/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethic Committee (No. 2022.382) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Ostrom QT, Cioffi G, Waite K, Kruchko C, Barnholtz-Sloan JS. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2014-2018. Neuro Oncol 2021;23:iii1-iii105. [Crossref] [PubMed]
2. Goldbrunner R, Stavrinou P, Jenkinson MD, Sahm F, Mawrin C, Weber DC, Preusser M, Minniti G, Lund-Johansen M, Lefranc F, Houdart E, Sallabanda K, Le Rhun E, Nieuwenhuizen D, Tabatabai G, Soffietti R, Weller M. EANO guideline on the diagnosis and management of meningiomas. Neuro Oncol 2021;23:1821-34. [Crossref] [PubMed]
3. Kshettry VR, Ostrom QT, Kruchko C, Al-Mefty O, Barnett GH, Barnholtz-Sloan JS. Descriptive epidemiology of World Health Organization grades II and III intracranial meningiomas in the United States. Neuro Oncol 2015;17:1166-73. [Crossref] [PubMed]
4. Ugga L, Spadarella G, Pinto L, Cuocolo R, Brunetti A. Meningioma Radiomics: At the Nexus of Imaging, Pathology and Biomolecular Characterization. Cancers (Basel) 2022;14:2605. [Crossref] [PubMed]
5. Louis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, Hawkins C, Ng HK, Pfister SM, Reifenberger G, Soffietti R, von Deimling A, Ellison DW. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro Oncol 2021;23:1231-51. [Crossref] [PubMed]
6. Hale AT, Wang L, Strother MK, Chambless LB. Differentiating meningioma grade by imaging features on magnetic resonance imaging. J Clin Neurosci 2018;48:71-5. [Crossref] [PubMed]
7. Lin BJ, Chou KN, Kao HW, Lin C, Tsai WC, Feng SW, Lee MS, Hueng DY. Correlation between magnetic resonance imaging grading and pathological grading in meningioma. J Neurosurg 2014;121:1201-8. [Crossref] [PubMed]
8. Liu Y, Chotai S, Chen M, Jin S, Qi ST, Pan J. Preoperative radiologic classification of convexity meningioma to predict the survival and aggressive meningioma behavior. PLoS One 2015;10:e0118908. [Crossref] [PubMed]
9. Huang X, Cao Y, Zhang G, Tang F, Sun D, Ren J, Li W, Zhou J, Zhang J. MRI morphological features combined with apparent diffusion coefficient can predict brain invasion in meningioma. Comput Biol Med 2025;187:109763. [Crossref] [PubMed]
10. Szczepankiewicz F, van Westen D, Englund E, Westin CF, Ståhlberg F, Lätt J, Sundgren PC, Nilsson M. The link between diffusion MRI and tumor heterogeneity: Mapping cell eccentricity and density by diffusional variance decomposition (DIVIDE). Neuroimage 2016;142:522-32. [Crossref] [PubMed]
11. Vasudevan HN, Choudhury A, Hilz S, Villanueva-Meyer JE, Chen WC, Lucas CG, Braunstein SE, Oberheim Bush NA, Butowski N, Pekmezci M, McDermott MW, Perry A, Solomon DA, Magill ST, Raleigh DR. Intratumor and informatic heterogeneity influence meningioma molecular classification. Acta Neuropathol 2022;144:579-83. [Crossref] [PubMed]
12. Bailo M, Pecco N, Callea M, Scifo P, Gagliardi F, Presotto L, Bettinardi V, Fallanca F, Mapelli P, Gianolli L, Doglioni C, Anzalone N, Picchio M, Mortini P, Falini A, Castellano A. Decoding the Heterogeneity of Malignant Gliomas by PET and MRI for Spatial Habitat Analysis of Hypoxia, Perfusion, and Diffusion Imaging: A Preliminary Study. Front Neurosci 2022;16:885291. [Crossref] [PubMed]
13. You D, Kim MM, Aryal MP, Parmar H, Piert M, Lawrence TS, Cao Y. Tumor image signatures and habitats: a processing pipeline of multimodality metabolic and physiological images. J Med Imaging (Bellingham) 2018;5:011009. [Crossref] [PubMed]
14. Dextraze K, Saha A, Kim D, Narang S, Lehrer M, Rao A, Narang S, Rao D, Ahmed S, Madhugiri V, Fuller CD, Kim MM, Krishnan S, Rao G, Rao A. Spatial habitats from multiparametric MR imaging are associated with signaling pathway activities and survival in glioblastoma. Oncotarget 2017;8:112992-3001. [Crossref] [PubMed]
15. Wang X, Li J, Sun J, Liu W, Cai L, Zhao P, Yang Z, Lv H, Wang Z. A Machine Learning Model Based on Unsupervised Clustering Multihabitat to Predict the Pathological Grading of Meningiomas. Biomed Res Int 2022;2022:8955227. [Crossref] [PubMed]
16. Yin L, Liu Y, Zhang X, Lu H, Liu Y. The Effect of Heterogenous Subregions in Glioblastomas on Survival Stratification: A Radiomics Analysis Using the Multimodality MRI. Technol Cancer Res Treat 2021;20:15330338211033059. [Crossref] [PubMed]
17. Qiao J, Kang H, Ran Q, Tong H, Ma Q, Wang S, Zhang W, Wu H. Metabolic habitat imaging with hemodynamic heterogeneity predicts individual progression-free survival in high-grade glioma. Clin Radiol 2024;79:e842-53. [Crossref] [PubMed]
18. World Medical Association Declaration of Helsinki. ethical principles for medical research involving human subjects. JAMA 2013;310:2191-4. [Crossref] [PubMed]
19. Lowekamp BC, Chen DT, Ibáñez L, Blezek D. The Design of SimpleITK. Front Neuroinform 2013;7:45. [Crossref] [PubMed]
20. Ostu N. A threshold selection method from gray-level histograms. IEEE Transactions on Systems, Man, and Cybernetics 1979;9:62-6.
21. Schoonjans F, Zalata A, Depuydt CE, Comhaire FH. MedCalc: a new computer program for medical statistics. Comput Methods Programs Biomed 1995;48:257-62. [Crossref] [PubMed]
22. Yu J, Chen FF, Zhang HW, Zhang H, Luo SP, Huang GD, Lin F, Lei Y, Luo L. Comparative Analysis of the MRI Characteristics of Meningiomas According to the 2016 WHO Pathological Classification. Technol Cancer Res Treat 2020;19:1533033820983287. [Crossref] [PubMed]
23. Smith KA, Leever JD, Chamoun RB. Predicting Consistency of Meningioma by Magnetic Resonance Imaging. J Neurol Surg B Skull Base 2015;76:225-9. [Crossref] [PubMed]
24. Perry A. Meningiomas. Practical surgical neuropathology: a diagnostic approach: Elsevier; 2018:259-98.
25. Villanueva-Meyer JE, Mabray MC, Cha S. Current Clinical Brain Tumor Imaging. Neurosurgery 2017;81:397-415. [Crossref] [PubMed]
26. Magill ST, Young JS, Chae R, Aghi MK, Theodosopoulos PV, McDermott MW. Relationship between tumor location, size, and WHO grade in meningioma. Neurosurg Focus 2018;44:E4. [Crossref] [PubMed]