Skip to main content
Download PDF

Exploring a recurrence model for atypical meningioma based on multiparametric MRI radiomic and clinical characteristics: a multicenter retrospective cohort study

Radiation Oncology volume 20, Article number: 30 (2025) Cite this article

Abstract

Objectives

Atypical meningioma (AM) has a high recurrence rate. This study explored the factors associated with recurrence and built a predictive model for AM by combining radiomic and clinical features.

Methods

This retrospective cohort study enrolled 451 adult AM patients who underwent surgical treatment at three institutions between May 2012 and April 2024. The patients in institution 1 were randomly assigned to the training dataset (n = 246) or internal validation dataset (n = 164) at a ratio of 6:4, and patients in institutions 2 and 3 composed the external validation dataset (n = 41). The clinical and pathological characteristics of the patients were collected, and radiomics technology was used to extract image features from preoperative multiparametric MR images. After feature screening, three types of AM recurrence prediction models were constructed: the radiomic model, the clinical model and the combined model.

Results

The median follow-up time was 27 months, and 22.8% (n = 103) of the patients relapsed after surgery. A total of 23 radiomic features were included in the model. Compared with the radiomic model and clinical model, the combined model performed better in predicting recurrence, with C-index values of 0.8453, 0.7867 and 0.8125 in the training, internal validation and external validation datasets, respectively, and the AUC value remained above 0.85 within 5 years. The radiomics score plays the most important role in predicting the recurrence of AM, with features such as t1c_original_shape_MajorAxisLength and t1c_original_shape_Flatness being of high importance. Among the clinical features, secondary tumors, subtotal resection and high Ki-67 levels contribute to AM recurrence.

Conclusions

Radiomics has additional value for predicting AM tumor recurrence and has favorable predictive performance when combined with clinical features.

Introduction

Meningioma is the most common primary intracranial tumor in adults, accounting for 36.7% of all intracranial tumors [1]. Histologically, meningiomas are classified into three grades: benign meningioma (grade I), atypical meningioma (AM) (grade II), and malignant meningioma (grade III) [2]. AM is a relatively rare subtype of meningioma, but with the update of the World Health Organization grading standard, its proportion exceeds 20% of all diagnoses [3]. AM has high mitotic activity, and its 5-year recurrence rate is as high as 40% [4]. Although total resection is currently recognized as the first choice for the treatment of AM, there are still limitations in postoperative patient management, especially in the prediction of recurrence.

Most of the previous models for predicting the recurrence of AM were based solely on some clinical features of patients, especially the degree of tumor resection and Ki-67 [5, 6]. Oversimplified clinical features make the prediction not always ideal. As the most important examination for patients with meningioma, MRI images have many hidden information that can be mined. As an emerging technology, radiomics has been widely used in clinical tumor research, including meningioma, in efforts to find meaningful information in radiological images that is visually inaccessible to humans [7]. Most such studies focus on exploring the relationship between preoperative radiomics and the differential diagnosis or predicted grade of tumors and lack a large sample multiparameter MRI radiomic model to help predict recurrence in patients with AM. It is also unknown how important the individual sequences of multi-parameter MRI images play in the AM prognostic model. Therefore, this study used the preoperative multiparametric MR images from five sequences of multicenter AM patients to explore their prognostic factors via radiomic feature extraction and modeling combined with clinical features.

Materials and methods

Patient selection

The study included data from three hospitals: 410 patients from the First Affiliated Hospital of Zhengzhou University (FAHZZU), 31 patients from the First Affiliated Hospital of Henan University of Science and Technology (FAHHAUST), and 10 patients from Zhumadian Central Hospital (ZMDCH). In this cohort, the pathologies of all patients were rereviewed by a neuropathologist based on the 2021 criteria, independent of the previous outcomes. The exclusion criteria were as follows: (1) age < 18 years; (2) multiple intracranial tumors; (3) death unrelated to disease progression within 3 months after the operation; (4) incomplete preoperative multiparametric MR (including T1, contrast-enhanced T1 (T1C), T2, T2 fluid-attenuated inversion recovery (T2FLAIR) and the corresponding apparent diffusion coefficient (ADC) maps from diffusion weighted imaging (DWI)) image information or clinical information or refusal to join the study; and (5) MRI data with intense and irreparable motion artifacts, as confirmed by a senior neuroradiologist. The patient screening process is described in Fig. 1. Patients from FAHZZU were randomly divided into a training dataset (n = 246) and an internal validation dataset (n = 164) at a ratio of 6:4, whereas patients from FAHHAUST and ZMDCH were jointly used as the external validation dataset (n = 41). This retrospective study adhered to the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Zhengzhou University (No. 2024-KY-0056-001). Informed consent was waived by the Committee due to the retrospective nature of this study.

Fig. 1
figure 1

Flow charts. A: Patient selection and enrollment. B: Radiomics processing

Full size image

MR image acquisition

Patients were scanned on either 1.5 T or 3.0 T clinical MR scanners. Detailed information on the MR machines and imaging parameters is available in the Supplementary Material. The MRI protocol used to acquire data included the following sequences: (1) axial T1 before and after intravenous injection of a dose of 0.1 mmol/kg gadolinium-based contrast agent (Magnevist, Bayer Healthcare, Berlin, Germany) followed by a 20-mL saline flush at a rate of 2 mL/sec; (2) axial T2; (3) axial T2FLAIR; and (4) axial DWI. DWI was acquired before injection of the contrast agent and was used as a monopolar spin‒echo echo‒planar sequence. After diffusion-sensitizing gradient encoding in the x, y, and z directions, the ADC maps were calculated from DWI with b = 0 and b = 1000 s/mm2 images [8].

Image preprocessing, segmentation and feature extraction

The procedures for radiomics studies follow our previous studies [8, 9]. The first step of image preprocessing utilized the SimpleITK package within the Python tool (version 3.7). This process aimed to standardize both the intensity and geometry. The N4ITK algorithm was employed to address bias field distortions, helping to eliminate intensity non-uniformities and enhance the consistency of image intensities across the entire brain [10]. Within-subject registration of the T1, T2, T2FLAIR and ADC sequences was accomplished based on the axial T1C through the general registration (ANTs, https://github.com/netstim/SlicerANTs) module of 3D slicer software (version 5.2.2, https://www.slicer.org) in rigid mode [11]. Following image registration, an assessment of anatomical structure congruence within the same patient across different sequences was carried out to ensure registration quality. Low-quality images were reregistered. Intensity normalization was performed via histogram matching [8].

To minimize any subjective error that might have occurred, a radiologist with 12 years of experience and a neurosurgeon with 25 years of experience performed slice-by-slice manual segmentation of the tumor region of interest (ROI) of all patients via 3D slicer software, whose discrepancies were resolved via consensus discussion. The investigators were blinded to the patients’ clinical information. The ROIs were drawn as close to the tumor edge as possible, from which areas with edema and peritumoral tissues were excluded. Given the characteristics of the MRI signals of meningiomas, the vast majority of meningiomas were most clearly shown in the enhanced T1 sequence. Therefore, to more accurately determine the tumor boundary, the tumors were outlined only in an enhanced T1 sequence [12]. A threshold tool was applied to set the appropriate threshold that covered the target area as much as possible, followed by manual segmentation via an intensity-based level tracing tool and paint tool. These tools were all implemented in the segment editor module of 3D slicer software [13]. Based on the image segmentations, isotropic voxel resampling was conducted, resulting in voxel dimensions of 1 × 1 × 1 mm3 via trilinear interpolation. The PyRadiomics package was used to extract radiomic features (including wavelet-based features), and it is incorporated into the SlicerRadiomics extension of the 3D Slicer software [8, 14].

For every tumor ROI, 864 radiomic features, including first-order statistics, shape, gray level cooccurrence matrix (GLCM), gray level dependence matrix (GLDM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), neighboring gray tone difference matrix (NGTDM), and wavelet-based features, were extracted from each sequence. A total of 2569 radiomic features were ultimately obtained.

Radiomic feature processing and screening

Interobserver reproducibility was used to assess the consistency of the extracted features. A second neurosurgeon with 14 years of experience, unaware of the patients’ clinical information, conducted segmentation on 50 randomly chosen patients via the same protocol. The data from this examiner were exclusively employed to assess interobserver agreement through the intraclass correlation coefficient (ICC). Features with an ICC value higher than 0.9 were selected. Patients were then randomly divided into a training dataset and a internal validation dataset at a 6:4 ratio. All features in the training and validation datasets were normalized to zero mean and unit variance via z score normalization. The correlation between features was then tested. For feature pairs with correlation coefficients greater than 0.75, features with better univariate predictive power (larger C-index values in the univariate Cox regression analysis) were retained. The least absolute shrinkage and selection operator (LASSO) regression algorithm was used for further feature screening. The LASSO regression algorithm is a method of high-dimensional data variable screening and is widely applied in radiomic analyses [15,16,17]. The feature screening process is performed using only the training dataset.

Clinical feature processing and screening

As in our previous study, baseline patient characteristics were sourced from our institutional records [6, 18,19,20]. All patients underwent surgery at our institution. Follow-up information was gathered through MRI data and telephone communications. Various parameters, including sex, age at surgery, tumor location, preoperative tumor size (maximum diameter), primary or secondary tumor status, peritumoral edema (graded according to Trittmacher criteria, degree 0) [21], presence of brain and bone involvement, initial symptoms, disease progression (from symptom onset to surgery), preoperative Karnofsky performance status (KPS), extent of resection, tumor blood supply (identified as a flow void at T2), tumor necrosis, S-100 protein levels, the Ki-67 index, and postoperative radiotherapy, were analyzed. Two experienced neurosurgeons independently summarized and evaluated all the data, resolving disagreements through discussion.

Patients were stratified based on age, tumor size, and disease progression using cutoff values of 55 years (median age), 46 mm (median tumor size), and 1 month, respectively. Preoperative KPS scores categorized patients into functional, near-functional (KPS 70–100) and severely impaired groups (KPS ≤ 70). Consistent with our previous large-scale meta-analysis and the majority of relevant studies by other scholars, the extent of resection was classified as gross total resection (GTR, Simpson grade I–II excision) or subtotal resection (STR, Simpson grade III–IV excision) [22]. The Ki-67 index was categorized as low (≤ 10%) or high (> 10%) [23, 24]. Tumor location was classified as convexity or nonconvexity. Tumor recurrence was defined as the presence of new lesions or significant growth (more than 25%) of residual tumor on MRI scans [19, 25]. Owing to duplication, tumor size (major axis length) was incorporated into the model as a radiomic feature. Progression-free survival (PFS) was defined as the time between surgery and tumor recurrence, recorded in monthly units [26]. The LASSO regression algorithm was subsequently used to select clinical features.

Development and validation of the models

The radiomic features and clinical features selected above were modeled via Cox regression to obtain the radiomic model and clinical model, respectively. In addition, the radiomics scores of each patient were calculated based on the radiomic model and combined with the clinical features to construct a combined model. The training dataset was used to implement the above three models, and then the internal and external validation datasets were used to evaluate the models. For each model, Harrell’s concordance index (C-index) was calculated for all datasets, and the model with the highest C-index was considered optimal.

Statistical analysis

The statistical analysis was performed via R 4.4.1 software. The balance of the distributions of the clinical features between the two datasets was verified. Categorical variables were assessed via the chi-square test, whereas continuous variables were analyzed via the Wilcoxon rank-sum test or Student’s t test. Characteristics of the correlations tested via the corrplot package. The receiver operating characteristic (ROC) area under the curve (AUC) and C-index were used to evaluate the classification performance of each model. Two-sided p < 0.05 was considered statistically significant.

Results

Demographic characteristics

A total of 451 patients were included in this study (Table 1), with 1.45 times more females than males, and the median age of onset was 55 years. The most common tumor site is convex, which often begins with headache. The median tumor size was 46 mm, and peritumoral edema was often present. The median follow-up was 27 months, and by the end of the study, 103 patients had relapsed, for a recurrence rate of 22.8%.

Table 1 Demographic characteristics
Full size table

Radiomic feature selection and model development

A total of 2569 features were extracted from multiparametric MRI sequences. After ICC selection, 1686 features remained. After z score normalization, 142 radiomic features were selected via feature correlation analysis (Fig. 2). The feature correlation heatmaps are shown in Supplementary Fig. 1. LASSO regression was used for further analysis, and 23 radiomic features were identified as highly correlated with recurrence, including 4 features from T1, 5 features from T1C, 7 features from T2 and 7 features from T2FLAIR. Using these features, Cox regression was used to establish a recurrence prediction model, and the radiomic feature model and radiomic feature score (radscore) of each sample were subsequently obtained (Fig. 2A-C).

Fig. 2
figure 2

Radiomic model and clinical model. A: LASSO regression was used to select radiomic features. B: Importance of radiomic model features. C: Nomogram based on the radiomic model. D: LASSO regression was used to select clinical features. E: Importance of clinical model features. F: Nomogram based on clinical model features

Full size image

Clinical feature screening and model establishment

A total of 15 clinical features were included in this study. The feature correlation heatmaps are shown in Supplementary Fig. 2. Similarly, via LASSO regression analysis, three clinical features that were highly associated with recurrence, namely, primary or secondary, extent of resection and Ki-67, were selected. Using these features, Cox regression was used to construct a recurrence prediction model, and finally, the clinical feature model was obtained (Fig. 2D-F).

Combined model Building

The radscore of each sample was calculated according to the above radiomic feature model. Using the three selected clinical features and the radscore, Cox regression was used to establish a new combined model. The nomogram revealed that the radscore was the most important contribution to the model, with a hazard ratio (HR) of 2.304 (95% CI: 1.805–2.942, p < 0.001). In addition, primary or secondary resection and the extent of resection significantly positively contribute to recurrence (Fig. 3).

Fig. 3
figure 3

Combined model. A: The forest map shows the HR of each feature in the combined model. B: Nomogram based on combined model features

Full size image

Model comparison

The ROC curves, AUC values and C-indexes of the radiomic model, clinical model and combined model in the training, internal validation and external validation datasets are shown in Fig. 4 and Supplementary Table 2. The prediction performance of the clinical feature model was poor, and the area under the curve (AUC) for predicting recurrence within 5 years was only approximately 0.75. The radiomic feature model performed slightly better than the clinical feature model. The combined model combining the radiomic feature score and clinical features showed the best performance in predicting recurrence. In the training dataset, the combined model had the largest AUC values for predicting recurrence at 1, 3, and 5 years, which were 0.88, 0.87 and 0.86, respectively. Similarly, it maintained good predictive performance in both the internal and external validation datasets, with AUC values of 0.82, 0.79, and 0.77 for the internal validation datasets at 1, 3, and 5 years, respectively, and 0.93, 0.81, and 0.89 for the external validation datasets at 1, 3, and 5 years, respectively. The C-index of the combined model was the highest, reaching 0.8453, 0.7867 and 0.8125 in the training, internal validation and external validation datasets, respectively, whereas the C-index of the clinical model was only 0.7642, 0.7355, and 0.7969, respectively. The KM curves of all the models in all the datasets were significantly different (p < 0.05) (Fig. 5).

Fig. 4
figure 4

ROC curves, AUC values and C-indexes of the radiomic model, clinical model and combined model. A-I: ROC curves of the three datasets of the three models. J: Comparison of the AUC values of the three models in training dataset within 5 years. K: C-index comparison of the three models in the three datasets

Full size image
Fig. 5
figure 5

Kaplan–Meier curves of PFS by the three models. A-C: Training dataset. D-F: Internal validation dataset. G-I: External validation dataset. (log-rank test: P < 0.01)

Full size image

Discussion

Radiomics, as a new low-cost and noninvasive tool, is a promising approach for individualized cancer management [27]. One of the most relevant advantages is that it can reveal intraregional heterogeneity by identifying different subregions, thus representing the spatial complexity of the disease [27, 28]. Radiomics has been used to predict the pathological features of meningiomas. Several studies have revealed the possible role of radiomics in the preoperative grading and prediction of the histological subtypes of meningiomas [29,30,31]. Some studies have focused on exploring the link between radiomics and important molecular pathologies, such as Ki-67 [32, 33]. Some studies have used radiomics to predict the aggressiveness of meningiomas as an indirect tool for predicting recurrence and poor prognosis [34, 35]. Zhang et al. attempted to use a radiomic approach in preoperative MRI to predict the recurrence of skull base meningiomas after incomplete resection [36]. Park et al. developed a radiomic model for the selection of patients for radiotherapy after AM surgery [7]. In this study, we used multiparametric MR radiomic features and/or clinical features to construct three AM recurrence prediction models.

In the radiomic model, 23 features were found to be closely related to recurrence. As one of the most commonly used enhanced MR image types, T1C images have been widely used in the diagnosis and research of intracranial tumors, providing detailed information on tissue structure and lesions to support clinical practice [37]. T1C scanning is recommended for highly suspected intracranial tumors, especially for meningiomas, as suggested by basic imaging, such as CT and nonenhanced MRI, to better determine the tumor scope, internal structure and adjacent tissues and ultimately provide a key reference for surgical resection. T2 is often used to observe the content of water molecules and the characteristics of tumor tissue, such as hemorrhage and cystic degeneration, and it can also clearly show peritumoral edema and anatomical structure. T2FLAIR imaging combines the characteristics of T2 and FLAIR sequences, and this modality is often used to study the development and progression of various nervous system diseases. The combination of the two can help to determine the internal structure of the tumor. In this study, T2 combined with T2FLAIR imaging provided a total of 14 features for predicting the recurrence of AM. Among all radiomic features, t1c_original_shape_MajorAxisLength and t1c_original_shape_Flatness are more important. Larger tumors generally imply more complex tumor internal structures, higher cell proliferation load, and more complex biological behavior, which may increase the risk of recurrence. Large tumors are more likely to invade surrounding tissues such as brain parenchyma or blood vessels, and the higher surgical difficulty also increases the risk of STR, which is closely associated with recurrence. Interestingly, t1c_original_shape_Flatness was negatively associated with recurrence, suggesting that the more spherical the tumor, the lower the likelihood of recurrence. This is a new discovery that we are trying to explain. A high Flatness value indicates a regular shape of the tumor, which may reflect a milder biological behavior and a lower aggressiveness of the tumor to the surrounding tissue. Often these tumors are removed more thoroughly, reducing the risk of recurrence. Irregularly shaped tumors may be associated with higher aggressiveness and a more complex microenvironment, such as increased angiogenesis and increased tissue heterogeneity, making their biological behavior more malignant.

Many studies have used patient clinical characteristics to explore prognostic factors for AM. Most patients are affected by headache or dizziness, and their physical condition is generally good; thus, surgery is currently recognized as the main treatment for AM. The extent of tumor resection has a large impact on patient prognosis [37,38,39]. Subtotal resection is currently recognized as a positive factor for recurrence [22]. Although there are few studies on secondary AM, the possible high surgical difficulty, high invasiveness and unique pathological features of secondary AM may be the reason for its significantly higher recurrence rate than that of primary AM [19]. Ki-67 in meningiomas has been closely monitored because it is closely related to the biological behavior of tumors and the prognosis of patients. The role of postoperative radiotherapy has been vigorously debated. In this study, we found that, in order of importance, secondary tumors, subtotal resection, a high Ki-67 index and a large tumor size (major axis length) contributed to AM recurrence.

To our knowledge, this study is a radiomic study of AM recurrence with the largest number of MRI sequences and the largest sample size. The combined model including the radscore obtained from the multiparametric MR radiomic model and the three clinical features showed the best performance in predicting AM recurrence in both the training dataset, internal validation dataset and the external validation dataset. The C-index of the model in the training dataset was 0.8453, which was significantly greater than the 0.8161 of the radiomic model and the 0.7642 of the clinical model, and the AUC value remained above 0.85 within 5 years. Furthermore, the nomogram revealed that the radiomics score played the most important role in predicting the recurrence of AM. These findings indicate that the addition of multiparametric MR radiomics can improve the prognostic performance of traditional clinical feature models. The combination of the two can lead to more accurate and individualized predictions of PFS in AM patients to better support treatment and follow-up.

There are several limitations in this study. To ensure uniform and clear multiparametric MR images, we excluded some older data, which resulted in a short median follow-up time for our patients. Our inclusion and exclusion criteria may have introduced several other inevitable selection biases. This study focuses on the radiomics features of tumor tissue, and the radiomics data related to peritumoral edema needs to be further explored [40]. Prospective cohorts from different centers will be needed in future work.

Conclusion

In this study, based on the multiparametric MR radiomics data and clinical data of 451 patients with AM in multiple centers, through feature screening, modeling and verification, recurrence prediction models with C-indexes of 0.8453, 0.7867 and 0.8125 in the training dataset, internal validation dataset and external validation dataset, respectively, were finally obtained. When combined with the clinical characteristics of AM patients, the radiomics score added prognostic value for predicting tumor recurrence. This combination could identify a subset of patients at high risk for recurrence who may benefit from intensive treatment and close follow-up.

Data availability

No datasets were generated or analysed during the current study.

Abbreviations

ADC:

Apparent diffusion coefficient

AM:

Atypical meningioma

AUC:

Area under the curve

DWI:

Diffusion-weighted imaging

GLCM:

Gray-level co-occurrence matrix

GLDM:

Gray-level dependence matrix

GLRLM:

Gray-level run length matrix

GLSZM:

Gray level size zone matrix

GTR:

Gross total resection

HR:

Hazard ratio

ICC:

Intraclass correlation coefficient

KPS:

Karnofsky performance status

LASSO:

Least absolute shrinkage and selection operator

NGTDM:

Neighboring gray tone difference matrix

PFS:

Progression-free survival

ROC:

Receiver operating characteristic

ROI:

Region of interest

STR:

Subtotal resection

T1C:

Contrast-enhanced T1

T2FLAIR:

T2 fluid-attenuated inversion recovery

References

  1. Ostrom QT, Gittleman H, Liao P, Rouse C, Chen Y, Dowling J, et al. CBTRUS statistical report: primary brain and central nervous system tumors diagnosed in the united States in 2007–2011. Neuro Oncol. 2014;16(Suppl 4):iv1–63.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Louis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, et al. The 2021 WHO classification of tumors of the central nervous system: a summary. Neuro Oncol. 2021;23:1231–51.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  3. Ostrom QT, Gittleman H, Truitt G, Boscia A, Kruchko C, Barnholtz-Sloan JS. CBTRUS statistical report: primary brain and other central nervous system tumors diagnosed in the united States in 2011–2015. Neuro Oncol. 2018;20:iv1–86.

    Article  PubMed  PubMed Central  Google Scholar 

  4. Aghi MK, Carter BS, Cosgrove GR, Ojemann RG, Amin-Hanjani S, Martuza RL, et al. Long-term recurrence rates of atypical meningiomas after gross total resection with or without postoperative adjuvant radiation. Neurosurgery. 2009;64:56–60.

    Article  PubMed  Google Scholar 

  5. Chen XY, Chen JY, Huang YX, Xu JH, Sun WW, Chen Y, et al. Establishment and validation of an integrated model to predict postoperative recurrence in patients with atypical meningioma. Front Oncol. 2021;11:754937.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Song D, Zhang M, Duan C, Wei M, Xu D, An Y, et al. A machine learning-based integrated clinical model for predicting prognosis in atypical meningioma patients. Acta Neurochir (Wien). 2023;165:4191–201.

    Article  PubMed  Google Scholar 

  7. Park CJ, Choi SH, Eom J, Byun HK, Ahn SS, Chang JH, et al. An interpretable radiomics model to select patients for radiotherapy after surgery for WHO grade 2 meningiomas. Radiat Oncol. 2022;17:147.

    Article  PubMed  PubMed Central  Google Scholar 

  8. Guo Y, Ma Z, Pei D, Duan W, Guo Y, Liu Z, et al. Improving noninvasive classification of molecular subtypes of adult gliomas with Diffusion-Weighted MR imaging: an externally validated machine learning algorithm. J Magn Reson Imaging. 2023;58:1234–42.

    Article  PubMed  Google Scholar 

  9. Yan J, Zhang S, Li KK, Wang W, Li K, Duan W, et al. Incremental prognostic value and underlying biological pathways of radiomics patterns in Medulloblastoma. EBioMedicine. 2020;61:103093.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Tustison NJ, Avants BB, Cook PA, Zheng Y, Egan A, Yushkevich PA, et al. N4ITK: improved N3 bias correction. IEEE Trans Med Imaging. 2010;29:1310–20.

    Article  PubMed  PubMed Central  Google Scholar 

  11. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. A reproducible evaluation of ants similarity metric performance in brain image registration. NeuroImage. 2011;54:2033–44.

    Article  PubMed  Google Scholar 

  12. Chu H, Lin X, He J, Pang P, Fan B, Lei P, et al. Value of MRI radiomics based on enhanced T1WI images in prediction of meningiomas grade. Acad Radiol. 2021;28:687–93.

    Article  PubMed  Google Scholar 

  13. Guo Z, Tian Z, Shi F, Xu P, Zhang J, Ling C, et al. Radiomic features of the edema region May contribute to grading meningiomas with peritumoral edema. J Magn Reson Imaging. 2023;58:301–10.

    Article  PubMed  Google Scholar 

  14. van Griethuysen J, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational radiomics system to Decode the radiographic phenotype. Cancer Res. 2017;77:e104–7.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Qian Z, Li Y, Wang Y, Li L, Li R, Wang K, et al. Differentiation of glioblastoma from solitary brain metastases using radiomic machine-learning classifiers. Cancer Lett. 2019;451:128–35.

    Article  CAS  PubMed  Google Scholar 

  16. Kim Y, Cho HH, Kim ST, Park H, Nam D, Kong DS. Radiomics features to distinguish glioblastoma from primary central nervous system lymphoma on multi-parametric MRI. Neuroradiology. 2018;60:1297–305.

    Article  PubMed  Google Scholar 

  17. Cui Y, Yang X, Shi Z, Yang Z, Du X, Zhao Z, et al. Radiomics analysis of multiparametric MRI for prediction of pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. Eur Radiol. 2019;29:1211–20.

    Article  PubMed  Google Scholar 

  18. Lin Y, Gao Q, Jin H, Wang N, Xu D, Wang F, et al. Analysis of approaches in the microsurgical treatment of 102 cases of petroclival meningioma in a single center. Front Neurol. 2021;12:627736.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Wang F, Xu D, Liu Y, Lin Y, Wei Q, Gao Q, et al. Risk factors associated with postoperative recurrence in atypical intracranial meningioma: analysis of 263 cases at a single neurosurgical centre. Acta Neurochir (Wien). 2019;161:2563–70.

    Article  PubMed  Google Scholar 

  20. Xu D, Wei Q, Li Z, Hu Y, Hu P, Zhao S, et al. Development and validation of predicting nomograms for craniopharyngioma: A retrospective, Multiple-Center, cohort study. Front Oncol. 2021;11:691288.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Trittmacher S, Traupe H, Schmid A. Pre- and postoperative changes in brain tissue surrounding a meningioma. Neurosurgery. 1988;22:882–5.

    Article  CAS  PubMed  Google Scholar 

  22. Song D, Xu D, Han H, Gao Q, Zhang M, Wang F, et al. Postoperative adjuvant radiotherapy in atypical meningioma patients: A Meta-Analysis study. Front Oncol. 2021;11:787962.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  23. Ho DM, Hsu CY, Ting LT, Chiang H. Histopathology and MIB-1 labeling index predicted recurrence of meningiomas: a proposal of diagnostic criteria for patients with atypical meningioma. Cancer. 2002;94:1538–47.

    Article  PubMed  Google Scholar 

  24. Barrett OC, Hackney JR, McDonald AM, Willey CD, Bredel M, Fiveash JB. Pathologic predictors of local recurrence in atypical meningiomas following gross total resection. Int J Radiat Oncol Biol Phys. 2019;103:453–9.

    Article  PubMed  Google Scholar 

  25. Weber DC, Ares C, Villa S, Peerdeman SM, Renard L, Baumert BG, et al. Adjuvant postoperative high-dose radiotherapy for atypical and malignant meningioma: A phase-II parallel non-randomized and observation study (EORTC 22042–26042). Radiother Oncol. 2018;128:260–5.

    Article  PubMed  Google Scholar 

  26. Palmieri M, Armocida D, De Pietro R, Chiarello G, Rizzo F, Garbossa D, et al. Clinical efficacy of adjuvant radiotherapy for world health organization grade II intracranial meningioma. World Neurosurg. 2023;175:e1117–23.

    Article  PubMed  Google Scholar 

  27. Xiao B, Fan Y, Zhang Z, Tan Z, Yang H, Tu W, et al. Three-Dimensional radiomics features from Multi-Parameter MRI combined with clinical characteristics predict postoperative cerebral edema exacerbation in patients with meningioma. Front Oncol. 2021;11:625220.

    Article  PubMed  PubMed Central  Google Scholar 

  28. Park JE, Kim HS. Radiomics as a quantitative imaging biomarker: practical considerations and the current standpoint in Neuro-oncologic studies. Nucl Med Mol Imaging. 2018;52:99–108.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Han Y, Wang T, Wu P, Zhang H, Chen H, Yang C. Meningiomas: preoperative predictive histopathological grading based on radiomics of MRI. Magn Reson Imaging. 2021;77:36–43.

    Article  PubMed  Google Scholar 

  30. Zhu Y, Man C, Gong L, Dong D, Yu X, Wang S, et al. A deep learning radiomics model for preoperative grading in meningioma. Eur J Radiol. 2019;116:128–34.

    Article  PubMed  Google Scholar 

  31. Park YW, Oh J, You SC, Han K, Ahn SS, Choi YS, et al. Radiomics and machine learning May accurately predict the grade and histological subtype in meningiomas using conventional and diffusion tensor imaging. Eur Radiol. 2019;29:4068–76.

    Article  PubMed  Google Scholar 

  32. Zhao Y, Xu J, Chen B, Cao L, Chen C. Efficient prediction of Ki-67 proliferation index in meningiomas on MRI: from traditional radiological findings to a machine learning approach. Cancers (Basel) 2022;14.

  33. Ouyang ZQ, He SN, Zeng YZ, Zhu Y, Ling BB, Sun XJ, et al. Contrast enhanced magnetic resonance imaging-based radiomics nomogram for preoperatively predicting expression status of Ki-67 in meningioma: a two-center study. Quant Imaging Med Surg. 2023;13:1100–14.

    Article  PubMed  PubMed Central  Google Scholar 

  34. Joo L, Park JE, Park SY, Nam SJ, Kim YH, Kim JH, et al. Extensive peritumoral edema and brain-to-tumor interface MRI features enable prediction of brain invasion in meningioma: development and validation. Neuro Oncol. 2021;23:324–33.

    Article  PubMed  Google Scholar 

  35. Xiao D, Zhao Z, Liu J, Wang X, Fu P, Le Grange JM, et al. Diagnosis of invasive meningioma based on brain-Tumor interface radiomics features on brain MR images: A multicenter study. Front Oncol. 2021;11:708040.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Zhang Y, Chen JH, Chen TY, Lim SW, Wu TC, Kuo YT, et al. Radiomics approach for prediction of recurrence in skull base meningiomas. Neuroradiology. 2019;61:1355–64.

    Article  PubMed  PubMed Central  Google Scholar 

  37. Loewenstern J, Shuman W, Rutland JW, Kessler RA, Kohli KM, Umphlett M, et al. Preoperative and histological predictors of recurrence and survival in atypical meningioma after initial gross total resection. World Neurosurg. 2019;128:e148–56.

    Article  PubMed  Google Scholar 

  38. Wiemels J, Wrensch M, Claus EB. Epidemiology and etiology of meningioma. J Neurooncol. 2010;99:307–14.

    Article  PubMed  PubMed Central  Google Scholar 

  39. Anand SK, Sardari H, Sadeghsalehi A, Bagheri SR, Eden SV, Lawton MT, et al. Radiopathologic predictors of recurrence in patients with a gross totally resected atypical meningioma. Neurol Res. 2022;44:468–74.

    Article  PubMed  Google Scholar 

  40. Frati A, Armocida D, Arcidiacono UA, Pesce A, D’Andrea G, Cofano F et al. Peritumoral Brain Edema in Relation to Tumor Size Is a Variable That Influences the Risk of Recurrence in Intracranial Meningiomas. Tomography. 2022;8: 1987-96.

Download references

Acknowledgements

We would like to thank our research team.

Funding

This study has received funding from the National Key Research and Development Program of China (2021YFE0204700), the Key Project of International Science and Technology Cooperation (241111520600), the Henan Province Science and Technology Research Project (232102310127), the National Natural Science Foundation of China (82102149), the Excellent Youth Talent Cultivation Program of Innovation in Health Science and Technology of Henan Province (YXKC2022061), the Key Program of Medical Science and Technique Foundation of Henan Province (SBGJ202002062), the Henan Province Science and Technology Research Project (242102311220), the Key Program of Medical Science and Technique Foundation of Henan Province (SBGJ202102128), and the Henan Province Science and Technology Research Project (242102310268).

Author information

Author notes

  1. Dengpan Song, Qingjie Wei and Shengqi Zhao contributed equally to this work.

Authors and Affiliations

  1. Department of Neurosurgery, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan Province, China

    Dengpan Song, Yuanhao Lou, Kaiyuan Zhang, Chengcheng Duan, Fang Wang, Dongming Yan, Yuchao Zuo & Fuyou Guo

  2. Department of Neurosurgery, The First Affiliated Hospital of Henan University of Science and Technology, Luoyang, Henan Province, China

    Qingjie Wei

  3. Department of Neurosurgery, Zhumadian Central Hospital, Zhumadian, Henan Province, China

    Shengqi Zhao

  4. Department of Neurosurgery, Beijing Children’s Hospital, Capital Medical University, Beijing, China

    Qiang Gao

  5. Department of MRI, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan Province, China

    Jing Yan

Authors
  1. Dengpan Song
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Qingjie Wei
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Shengqi Zhao
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Yuanhao Lou
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Kaiyuan Zhang
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Chengcheng Duan
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Fang Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Qiang Gao
    View author publications

    You can also search for this author inPubMed Google Scholar

  9. Jing Yan
    View author publications

    You can also search for this author inPubMed Google Scholar

  10. Dongming Yan
    View author publications

    You can also search for this author inPubMed Google Scholar

  11. Yuchao Zuo
    View author publications

    You can also search for this author inPubMed Google Scholar

  12. Fuyou Guo
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

DS conducted the statistical analysis and manuscript writing. QW, SZ, YL, KZ, CD, FW, JY, YZ and FG sifted, collected and processed the data and images. FW, QG, JY, YZ and FG evaluated the data and results. DS conceptualized the manuscript outline. JY, DY and FG revised the manuscript. All the authors read and approved the final manuscript.

Corresponding authors

Correspondence to Yuchao Zuo or Fuyou Guo.

Ethics declarations

Ethics approval and consent to participate

This retrospective study adhered to the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Zhengzhou University (No. 2024-KY-0056-001). The consent process was omitted due to the retrospective nature of our study.

Consent for publication

The authors affirm that the human research participants provided informed consent for the publication of the images in Fig. 1.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Song, D., Wei, Q., Zhao, S. et al. Exploring a recurrence model for atypical meningioma based on multiparametric MRI radiomic and clinical characteristics: a multicenter retrospective cohort study. Radiat Oncol 20, 30 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13014-025-02613-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13014-025-02613-y

Keywords

Radiation Oncology

ISSN: 1748-717X

Contact us