MRI radiomics-based machine learning for preoperative prediction of Simpson grade in intracranial meningiomas: a pilot study

MRI radiomics-based machine learning for preoperative prediction of Simpson grade in intracranial meningiomas: a pilot study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article MRI radiomics-based machine learning for preoperative prediction of Simpson grade in intracranial meningiomas: a pilot study Mateo Moreno Gómez, Iván Mauricio Herrera Mora, Adriana Milena Páez Rodríguez This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8704777/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background: meningiomas are the most frequent primary central nervous system tumors, and extent of resection is a key determinant of long-term tumor control. Quantitative preoperative MRI radiomics may support surgical planning by estimating the likelihood of achieving Simpson grade I versus II resection. This study evaluated the feasibility of an MRI radiomics-based machine learning approach for this purpose. Methods: a retrospective pilot study (2018–2024) included adults with intracranial meningioma and adequate preoperative contrast-enhanced T1-weighted MRI. Tumors were segmented in 3D Slicer and radiomic features were extracted using PyRadiomics under IBSI-compliant settings. The endpoint was binary (Simpson grade I vs II). Model selection used a selector–classifier benchmark with internal validation via nested leave-one-out cross-validation to mitigate data leakage. Performance was assessed using balanced accuracy, AUROC, AUPRC, and calibration (intercept, slope, Brier score). Uncertainty was estimated using bootstrap on out-of-fold predictions. Results: out of approximately 70 screened cases, 12 were included for modeling (Simpson I=4; Simpson II=8) with 126 candidate predictors. The best-performing configuration was LASSO (C=0.2) combined with linear discriminant analysis, achieving balanced accuracy 0.812, AUROC 0.844, AUPRC 0.867, and Brier score 0.129. Calibration was close to ideal (intercept ≈0.00; slope ≈0.97). Bootstrap 95% confidence intervals reflected uncertainty consistent with the small sample size. Conclusions: this pilot study supports the feasibility of a standardized MRI radiomics-based model for preoperative estimation of surgical resectability (Simpson I vs II) in intracranial meningiomas. Multicenter external validation is required to confirm generalizability and clinical utility. Meningioma Radiomics Machine learning Magnetic resonance imaging Simpson grade Extent of resection Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 INTRODUCTION Meningiomas are the most frequent primary intracranial tumors in adults, with an estimated annual incidence of approximately 5.3 per 100,000 and a sustained increase with aging [15]. The 2021 World Health Organization (WHO) classification recognizes three histological grades with significant prognostic implications; specifically, grades II–III exhibit greater biological aggressiveness and higher recurrence rates compared with grade I [13]. Surgical resection remains the cornerstone of management, with adjuvant radiotherapy reserved for selected high-risk scenarios [2]. The extent of resection, traditionally quantified using the Simpson grading system, is consistently associated with long-term tumor control and risk of relapse [21]. In addition, anatomical and morphological factors, including tumor location and volume, condition resectability and preoperative planning [12, 24]. Limitations of conventional visual assessment on magnetic resonance imaging (MRI) have motivated the use of radiomics and machine learning (ML) to extract quantitative imaging phenotypes (e.g., intensity, texture, and shape) that are not readily appreciable to the naked eye [1]. In oncology, radiomics has been positioned as a quantitative frontier for outcome prediction and risk stratification [18]. In meningiomas specifically, radiomics and ML approaches have been explored for non-invasive grading and related clinically relevant endpoints, including texture-based multiparametric differentiation and enhanced T1-weighted imaging models [4, 5, 9, 11, 26], as well as prediction of invasive behavior and outcome-relevant endpoints such as local failure [14, 25]. These developments parallel advances in tumor biology, such as DNA methylation–based classifications, which refine prognostic heterogeneity and provide a framework for multimodal model design [20]. Furthermore, radiomic signatures have shown potential to outperform purely clinical models in other central nervous system and head-and-neck neoplasms, supporting the conceptual validity of these approaches [10, 17]. Despite these advances, published radiomics studies in meningioma present relevant methodological limitations that hinder clinical translation. A meta-analysis reported an alarmingly low average Radiomics Quality Score (RQS ≈19%) and high heterogeneity across acquisition and segmentation protocols [23]. Consistently, systematic reviews emphasize variability in methodology and endpoints across the meningioma radiomics studies [19]. While performance for non-invasive grading has been reported in the moderate-to-high range in selected studies [9, 11], most current work prioritizes biological grading or related surrogate endpoints, whereas preoperative prediction of surgical resectability remains comparatively underdeveloped. This oversight is clinically important because estimating the probability of achieving an optimal resection (Simpson grade I) directly informs individualized operative planning and patient counseling [3, 12, 24]. To address these limitations, rigorous methodological standards and transparent reporting frameworks are required, including the Image Biomarker Standardization Initiative (IBSI) and TRIPOD+AI guidelines [27, 6]. The aim of this pilot study was to adopt methodological standards to develop and internally validate a preoperative ML model integrating relevant clinical variables and standardized MRI radiomic features to estimate the probability of achieving Simpson grade I versus II resection in intracranial meningiomas, thereby providing a quantitative tool to support neurosurgical decision-making. MATERIAL AND METHODS Ethical considerations and data governance The study was conducted in accordance with the Declaration of Helsinki and its later amendments. Ethical approval was granted by the Institutional Review Board of Universidad Tecnológica de Pereira (Protocol No. 67-040725). All clinical and imaging data were anonymized prior to analysis; no direct identifiers or re-identification keys were retained. Data was managed in a controlled-access environment with encryption. Given the retrospective design and the exclusive use of de-identified data, the requirement for informed consent was waived by the ethics committee. Study design, setting, and participant selection A retrospective observational study was conducted including adult patients with histopathological confirmed intracranial meningioma who underwent surgical resection at Hospital Universitario San Jorge (Pereira, Colombia) between 2018 and 2024. Approximately 70 potentially eligible cases were identified from institutional surgical and pathology registries. Case ascertainment and cohort assembly followed a pre-specified screening workflow comprising imaging quality control and clinical/pathological verification. DICOM (Digital Imaging and Communications in Medicine) studies underwent quality assessment for suitability for volumetric segmentation and radiomic feature extraction, and clinical/histopathological records were reviewed for completeness and internal consistency. After this process and removal of duplicated/incongruent records, 13 cases met eligibility criteria. Eligibility required (i) technically adequate preoperative contrast-enhanced T1-weighted MRI in DICOM format suitable for volumetric segmentation and radiomic analysis, and (ii) complete clinical documentation for outcome ascertainment. Cases were excluded due to insufficient image quality or spatial resolution for reliable volumetric reconstruction and feature extraction (e.g., severe motion artifacts or slice thickness >3 mm), inconsistent histopathological reporting, or duplicated records. One eligible patient with a Simpson grade IV resection was excluded to ensure statistical stability under a binary endpoint in a small-sample, high-dimensional setting. The final analytical cohort comprised 12 patients. Outcome definition The primary endpoint was extent of resection, defined as a binary classification task: Simpson grade I versus Simpson grade II. This endpoint was selected as representing a clinically relevant decision threshold for operative planning and recurrence risk. Other Simpson grades were not modeled due to marginal frequency in the available cohort, which would have compromised estimation stability. Image acquisition and tumor segmentation Preoperative MRI included contrast-enhanced T1-weighted sequences for all patients. Volumetric tumor segmentation was performed using 3D Slicer (version 5.6.2). Segmentations were generated following a semi-automated workflow (Fig. 1). To minimize inter-observer variability, all masks were independently reviewed by trained neurosurgeons and finalized by consensus adjudication, yielding a single 3D tumor mask per case. Three-dimensional surface reconstructions were generated from the final segmentation for qualitative visualization and documentation of tumor morphology (Fig. 2). Figure 1Semi-automated meningioma segmentation on contrast-enhanced T1-weighted MRI using 3D Slicer a–c Axial, coronal, and sagittal MRI views showing a skull-base meningioma with homogeneous enhancement d–f Corresponding tumor mask (yellow) generated using semi-automated tools and finalized by consensus review. Figure 2 Three-dimensional reconstruction of the segmented meningioma a,c Axial and coronal MRI views showing the meningioma and the corresponding segmentation mask (yellow) b,d Three-dimensional surface reconstruction derived from the final segmentation Radiomic features extraction and standardization Radiomic feature extraction was performed using PyRadiomics version 3.1.0 under a single documented configuration file [8]. Preprocessing and feature computation were aligned with IBSI recommendations and included isotropic resampling to 1×1×1 mm³ using B-spline interpolation, z-score intensity normalization, and texture discretization with a fixed bin width of 31 [27]. A total of 126 preoperative numeric predictors (clinical and radiomic) were assembled, encompassing first-order statistics, shape descriptors, and texture features (GLCM, GLRLM, GLSZM, NGTDM, and GLDM). Machine learning model and statistical analysis Preprocessing and feature selection. Computational modeling was implemented in Python version 3.10 using scikit-learn. The analytical workflow was designed to prioritize parsimony and mitigate overfitting in a small-sample, high-dimensional setting, while enforcing strict separation between training and held-out data. Data-dependent transformations were performed strictly within each training split and then applied to the held-out case to prevent information leakage. Feature scaling used z-score standardization computed exclusively on the training data of each split. To reduce redundancy among predictors, a pairwise correlation-based pruning step was applied prior to feature selection. Three feature selection strategies were benchmarked: ANOVA-KBest, LASSO and Elastic Net. In line with the subject-to-predictor ratio of the analytical cohort, the maximum number of retained predictors was constrained during model screening, with ≤3 predictors as the primary cap and a sensitivity cap up to ≤5 predictors evaluated. Regularization strength was tuned through the inverse penalty parameter C; for LASSO, C=0.2 was assessed as a high-penalty configuration consistent with the final selected model. Classification models . Seven supervised classifiers were evaluated to compare linear, probabilistic, and ensemble-based approaches under controlled conditions: Linear Discriminant Analysis (LDA), Gaussian Naïve Bayes, Logistic Regression, Support Vector Machines (linear and radial basis function kernels), Random Forest, Nearest Centroid, and k-Nearest Neighbors using Mahalanobis distance. This model set was selected to reflect the classifiers reported in the comparative results and to assess performance under alternative inductive biases. Validation strategy. A two-stage internal validation framework was applied. First, candidate models (feature selection × classifier) were screened using leave-one-out cross-validation (LOOCV) to identify top-performing configurations. Second, the best-performing pipeline was evaluated using nested LOOCV, with an inner loop for hyperparameter tuning and feature selection and an outer loop for unbiased performance estimation on the held-out case, consistent with TRIPOD+AI reporting recommendations [6]. Performance metrics, calibration, and uncertainty. Balanced accuracy was pre-specified as the primary metric to account for class imbalance, complemented by accuracy and macro-averaged F1 score. Discrimination was further characterized using AUROC and AUPRC. Calibration was evaluated using the Brier score and calibration intercept and slope derived from out-of-fold predicted probabilities. Uncertainty was quantified using 1,000 bootstrap resamples of out-of-fold predictions to derive 95% confidence intervals for reported metrics. RESULTS Cohort assembly and baseline characteristics Approximately 70 patients with diagnosis of intracranial meningioma were initially identified from surgical and pathology registries. After DICOM imaging quality control, verification of clinical and histopathological completeness, 13 cases met the eligibility criteria. For model development and internal validation, the primary endpoint was defined as Simpson grade I versus Simpson grade II. One eligible patient with a Simpson grade IV resection (n=1) was excluded because a sparsely represented third class would yield unstable estimates The final analytical cohort comprised 12 patients (Simpson I, n=4; Simpson II, n=8) with 126 preoperative predictors. Regarding the study cohort, the median age was 61.5 years (IQR 54.8–65.8; range 34–80), with a slight female predominance (7/12, 58.3%). The most frequent presenting symptom was headache (10/12, 83.3%), followed by focal neurological deficits (8/12, 66.7%) and seizures (6/12, 50.0%). Most tumors were in the supratentorial compartment (10/12, 83.3%); frontoparietal and frontal (midline/basal) locations were the most common. All patients underwent gross-total tumor removal; Simpson grade I was achieved in 4 cases (33.3%) and Simpson grade II in 8 cases (66.7%). Baseline demographic, clinical, and topographic characteristics are summarized in Table 1 . Table 1 Baseline demographic, clinical, and topographic characteristics of the final cohort Characteristic Value (N=12) Age (years), median [IQR] 61.5 [54.8–65.8] Range 34–80 Sex, n (%) Male 5 (41.7%) Female 7 (58.3%) Clinical presentation, n (%) Headache 10 (83.3%) Focal neurological deficit 8 (66.7%) Seizures 6 (50.0%) Tumor topography, n (%) Supratentorial 10 (83.3%) Frontoparietal 3 (25.0%) Frontal (midline/basal) 2 (16.7%) Frontotemporal / sphenoid wing 2 (16.7%) Others (occipital, intraventricular, insular/orbital) 3 (25.0%) Infratentorial 2 (16.7%) Posterior fossa / cerebellopontine angle 2 (16.7%) Extent of resection (Simpson grade), n (%) Grade I 4 (33.3%) Grade II 8 (66.7%) Comparative model screening (LOOCV) Comparative screening was performed using leave-one-out cross-validation (LOOCV), combining three feature-selection strategies (ANOVA-KBest, LASSO, and Elastic Net) with correlation-based pruning and a prespecified cap on the number of retained predictors (≤3 as the primary cap; sensitivity analyses up to ≤5) across seven supervised classifiers. Balanced accuracy was used as the primary selection criterion to account for class imbalance. Across screened configurations, the best-performing pipeline corresponded to LASSO feature selection (C=0.2; cap ≤3–5 variables) combined with linear discriminant analysis (LDA), achieving accuracy=0.833, balanced accuracy=0.812, and macro-averaged F1=0.812 on out-of-fold predictions (Table 2). LASSO combined with Gaussian Naïve Bayes and LASSO combined with a Mah alanobis-distance k-nearest neighbor classifier showed comparable accuracy during screening but demonstrated lower discrimination metrics in the final evaluation. Support vector machine and random forest models showed consistently inferior performance under the present sample size (Table 2). Table 2 Comparative evaluation of feature-selection and classifier configurations during LOOCV. Feature selection method Classifier Regularization (C) Max. variables Accuracy Balanced accuracy Macro-F1 LASSO LDA 0.2 3 0.833 0.812 0.812 LASSO LDA 0.2 5 0.833 0.812 0.812 LASSO Gaussian Naïve Bayes 0.2 3 0.833 0.812 0.812 LASSO kNN (Mahalanobis) 0.2 3 0.833 0.812 0.812 LASSO kNN (Mahalanobis) 0.2 5 0.833 0.812 0.812 Elastic Net SVM (RBF) 0.5 5 0.583 0.556 0.556 LASSO Logistic Regression 0.7 3 0.667 0.562 0.562 LASSO SVM (linear) 0.2 3 0.667 0.562 0.562 ANOVA-KBest LDA — 3 0.583 0.496 0.496 Final model performance, calibration, and uncertainty The best-performing workflow (LASSO, C=0.2; ≤3 variables + LDA) was evaluated using nested LOOCV to minimize optimistic bias by re-optimizing feature selection and tuning within each iteration. On out-of-fold predicted probabilities, the final model achieved accuracy=0.833, balanced accuracy=0.812, macro-F1=0.812, AUROC=0.844, AUPRC=0.867, and Brier score=0.129 (Table 3; Fig. 3). The confusion matrix showed correct classification of 3/4 Simpson grade I cases (sensitivity 0.75) and 7/8 Simpson grade II cases (sensitivity 0.875) (Fig. 4a). Calibration was close to ideal, with a global intercept approximately 0.00 and slope approximately 0.97 (Table 3; Fig. 4b,c). Table 3 Final model performance (LASSO + LDA) after nested LOOCV. Metric Result Accuracy 0.833 Balanced accuracy 0.812 Macro-F1 0.812 AUROC 0.844 AUPRC 0.867 Brier score 0.129 Global calibration Intercept = 0.000; Slope = 0.973 Figure 3 Discrimination performance of the final model (LASSO + LDA) under nested LOOCV a,b ROC curves for Simpson grade I and Simpson grade II, respectively c,d Precision–recall curves for Simpson grade I and Simpson grade II, respectively Figure 4 Confusion matrix and calibration analysis of the final model (LASSO + LDA) A Confusion matrix for the binary classification task (Simpson grade I vs II) b,c Class-specific calibration curves comparing predicted probabilities with observed event fractions Uncertainty estimated by bootstrap resampling (1,000 iterations) of out-of-fold predictions yielded 95% confidence intervals of 0.583–1.000 for accuracy, 0.500–1.000 for balanced accuracy, 0.478–1.000 for macro-F1, 0.583–1.000 for AUPRC, and 0.026–0.272 for the Brier score. Feature association and correlation structure Point-biserial association analysis indicated that shape- and intensity-derived features concentrated most of the discriminative signal, with Mean.1, Flatness, Strength, Elongation, and Sphericity ranking among the most associated variables (Fig. 5). During nested validation, Strength was selected in 12/12 folds, followed by Flatness and Elongation. A Pearson correlation heatmap of the top-ranked variables highlighted clustering patterns primarily among shape descriptors (Fig. 6). Figure 5 Preoperative variables most associated with the surgical endpoint (Simpson grade I vs II) Bar plot of the top ten variables ranked by point-biserial correlation; positive values indicate association with Simpson grade I and negative values with Simpson grade II Figure 6 Fig. 6 Correlation structure of top-ranked predictors Pearson correlation heatmap of the variables ranked highest by point-biserial association with the endpoint. DISCUSSION This pilot study developed and internally validated a preoperative prediction machine learning model to estimate the likelihood of achieving a more radical extent of resection in intracranial meningiomas, using standardized radiomic descriptors extracted from contrast-enhanced T1-weighted MRI together with preoperative clinical variables. In the studied cohort, the selected configuration combining LASSO feature selection with Linear Discriminant Analysis showed good discrimination (AUROC 0.844; AUPRC 0.867) and calibration (slope ≈ 0.97). Given the important role of extent of resection in long-term tumor control and recurrence risk, quantitative tools that complement conventional imaging assessment may support individualized planning and preoperative counseling, particularly in scenarios where dural management and anatomical constraints are expected to influence the operative strategy [12, 21, 24]. From a surgical standpoint, achieving a more radical Simpson grade is not solely a matter of removing the intradural mass; it also depends on the treating of the dural attachment and on interface related factors such as adherence and cleavage planes, which are variably appreciated on routine visual inspection and are often clarified intraoperatively [12, 21]. Within this context, the pattern of radiomic features retained by the model is clinically informative. Strength emerged as the most stable predictor across all nested folds (12/12), indicating that a reproducible measure of intratumoral intensity contrast carries consistent signal for the surgical endpoint. Prior radiomics studies in meningiomas have shown that intensity and texture derived descriptors contribute meaningfully to clinically relevant classification tasks, supporting the broader relevance of heterogeneity metrics in this disease [4, 5, 9, 11]. Although radiologic–pathologic correlation was not available in the present cohort, a conservative biological rationale is plausible and consistent with the radiomics paradigm that images contain quantifiable information. Greater post contrast heterogeneity may reflect variation in vascular architecture and stromal microstructure, which could translate into differences in tumor consistency, cleavage planes, and the sharpness of the tumor–dura boundary that explain situations of technical difficulty and the feasibility of dural management [7, 12]. This interpretation should be tested in larger cohorts that prospectively capture standardized intraoperative descriptors (e.g., consistency, adherence, sinus proximity, dural involvement) and, when feasible, integrate matched histopathological correlates. Shape-derived features, particularly Elongation and Flatness, also showed association with the surgical endpoint but lower stability, consistent with redundancy and collinearity among geometric descriptors. Meningiomas that grow along curved dural surfaces or extend across skull-base planes may present elongated or flattened morphologies, which can translate into constrained operative corridors and increased proximity to critical neurovascular structures [24]. Importantly, location is known to modulate both the technical meaning and prognostic value of Simpson grading, reinforcing the plausibility that shape descriptors may partially encode topographic constraints relevant to dural treatment decisions and surgical aggressiveness [9, 12]. Enhancement-derived information also appeared relevant in this cohort, as reflected by the association of Mean.1 (post-contrast mean intensity). Enhancement patterns on T1-weighted imaging are influenced by vascularity and microenvironmental factors and have been leveraged in multiple radiomics studies for clinically meaningful tasks in meningiomas, including grading-related classification using conventional and multiparametric MRI [9, 5]. While these signals are biologically plausible, interpretation should remain cautious and in the search of prospective standardization across scanners and acquisition protocols [23, 27]. From a methodological standpoint, combining a penalized linear selector with a linear classifier provided a balance between parsimony, stability, and interpretability in a high-dimensional setting with limited sample size. LASSO reduces variance by shrinking weak predictors toward zero and retaining only the strongest signals, which is particularly relevant in radiomics where predictor redundancy and collinearity are common [22]. In contrast, more flexible non-linear approaches often require substantially larger datasets to avoid instability and optimistic performance estimates [16]. The adoption of nested leave-one-out cross-validation further strengthened internal validity by ensuring that feature selection and tuning were performed strictly within each training split, reducing information leakage and providing a less biased estimate of model performance [6, 16]. This design aligns with current expectations for transparent development and evaluation of prediction models in biomedical AI [6]. When placed in the broader meningioma radiomics literature, this work addresses concerns that have been raised in various reviews. Published evidence has pointed out that many studies show low methodological quality and substantial variability in acquisition, segmentation, and validation, which limits reproducibility and makes results difficult to compare across cohorts [23]. The methodology of the study aimed to reduce variability by using a standardized, fully documented preprocessing and feature-extraction workflow consistent with IBSI-oriented recommendations and by reporting the modeling process in line with current transparency standards for prediction models [6, 27]. Although the findings require cautious interpretation, it lays out a clear, reproducible workflow that can be straightforwardly evaluated in larger, multicenter cohorts and adapted to different clinical settings. Clinically, a preoperative estimate that a case is more likely to require less radical dural management may still be useful even when not used as a decision rule. It can help frame preoperative expectations, guide planning around surgical corridors and adjacent neurovascular structures, and support earlier discussion of surveillance and adjunctive strategies when anatomy is likely to limit dural treatment. In that sense, this pilot study demonstrates feasibility under a standardized workflow, showing that routine MRI-derived quantitative descriptors can be mapped to a surgically meaningful endpoint in a way that is reproducible. LIMITATIONS Various limitations should be considered. First, the sample size limits precision and increases uncertainty around performance estimates. This is consistent with early radiomics work in meningiomas and neuroimaging, where proof-of-concept models have often been developed in similarly sized cohorts, but it still requires cautious interpretation [11, 23, 26]. Second, the retrospective single-center design constrains generalizability, particularly because MRI acquisition parameters and scanner hardware can influence radiomic feature distributions and reproducibility. Third, the analysis relied on a single primary sequence, and other inputs may capture complementary biological and anatomical information. Future work should prioritize external validation in multicenter cohorts, ideally with homogeneous preprocessing and sensitivity analyses across acquisition settings. Expanding the imaging input to include additional MRI sequences may improve robustness and capture complementary features. In parallel, prospective data collection that includes structured intraoperative variables and, when feasible, matched histopathology would enable more direct assessment of why specific imaging descriptors relate to resectability. Finally, implementation studies should evaluate whether preoperative probabilistic estimates meaningfully support surgical planning, counseling, and follow-up decisions, in line with guidance on evaluating clinical performance and impact of AI tools. CONCLUSION A standardized radiomics-based modeling approach showed feasibility for preoperative estimation of surgical resectability in intracranial meningiomas. In this cohort, the LASSO + LDA configuration achieved favorable discrimination and calibration under nested internal validation, and feature patterns suggested that heterogeneity- and shape-related descriptors may encode aspects of operative complexity. These results support further testing in larger, multicenter datasets and provide a clear methodological basis for external validation and refinement toward clinically useful, reproducible decision-support tools. Declarations STATEMENTS AND DECLARATIONS Funding: the authors did not receive support from any organization for the submitted work . Competing Interests : the authors have no relevant financial or non-financial interests to disclose. Human Ethics and Consent to Participate declarations: this study involved human participant data. Ethical approval was granted by the Institutional Review Board of Universidad Tecnológica de Pereira (Protocol No. 67-040725). Given the retrospective design and the exclusive use of fully de-identified data, the requirement for informed consent to participate was waived by the ethics committee. Consent to Participate: Informed consent to participate was waived by the Institutional Review Board of Universidad Tecnológica de Pereira (Protocol No. 67-040725) due to the retrospective nature of the study and the use of fully anonymized, de-identified data. Consent for Publication: Not applicable. Data Availability: the datasets and models generated and/or analyzed during the current study are available upon request and subject to institutional approval. Data will be provided in de-identified form. Author Contributions MMG : Mateo Moreno Gómez IMHM : Iván Mauricio Herrera Mora AMPR : Adriana Milena Páez Rodríguez IMHM, AMPR, and MMG contributed to the conceptualization of the study. MMG and IMHM developed the methodology. MMG implemented the software and performed radiomics feature extraction and preprocessing. IMHM and MMG conducted the formal analysis, with input from AMPR. MMG, IMHM, and AMPR contributed to the investigation and verification of clinical/imaging data. IMHM and AMPR performed the 3D Slicer tumor segmentations and generated the 3D reconstructions (Figs. 1–2). MMG prepared the remaining figures and tables (Tables 1–3; Figs. 3–6). MMG and IMHM wrote the first draft of the manuscript. AMPR, MMG, and IMHM reviewed and edited the manuscript. AMPR supervised the study. All authors read and approved the final manuscript. References Aerts HJWL, Velazquez ER, Leijenaar RTH, et al (2014) Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 5(1):4006 Apra C, Peyre M, Kalamarides M (2018) Current treatment options for meningioma. Expert Rev Neurother 18(3):241–249 Cahill KS, Claus EB (2011) Treatment and survival of patients with nonmalignant intracranial meningioma: results from the Surveillance, Epidemiology, and End Results Program of the National Cancer Institute. Clinical article. J Neurosurg 115(2):259–267 Chen C, Guo X, Wang J, Guo W, Ma X, Xu J (2019) The Diagnostic Value of Radiomics-Based Machine Learning in Predicting the Grade of Meningiomas Using Conventional Magnetic Resonance Imaging: A Preliminary Study. Front Oncol. doi: 10.3389/fonc.2019.01338 Chu H, Lin X, He J, Pang P, Fan B, Lei P, Guo D, Ye C (2021) Value of MRI Radiomics Based on Enhanced T1WI Images in Prediction of Meningiomas Grade. Acad Radiol 28(5):687–693 Collins GS, Moons KGM, Dhiman P, et al (2024) TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 385:e078378 Gillies RJ, Kinahan PE, Hricak H (2016) Radiomics: Images Are More than Pictures, They Are Data. Radiology 278(2):563–577 van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, Beets-Tan RGH, Fillion-Robin J-C, Pieper S, Aerts HJWL (2017) Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 77(21):e104–e107 Ke C, Chen H, Lv X, et al (2020) Differentiation Between Benign and Nonbenign Meningiomas by Using Texture Analysis From Multiparametric MRI. J Magn Reson Imaging JMRI 51(6):1810–1820 Kickingereder P, Burth S, Wick A, et al (2016) Radiomic Profiling of Glioblastoma: Identifying an Imaging Predictor of Patient Survival with Improved Performance over Established Clinical and Radiologic Risk Models. Radiology 280(3):880–889 Laukamp KR, Shakirin G, Baeßler B, Thiele F, Zopfs D, Große Hokamp N, Timmer M, Kabbasch C, Perkuhn M, Borggrefe J (2019) Accuracy of Radiomics-Based Feature Analysis on Multiparametric Magnetic Resonance Images for Noninvasive Meningioma Grading. World Neurosurg 132:e366–e390 Lemée J-M, Corniola MV, Da Broi M, Joswig H, Scheie D, Schaller K, Helseth E, Meling TR (2019) Extent of Resection in Meningioma: Predictive Factors and Clinical Implications. Sci Rep 9(1):5944 Louis DN, Perry A, Wesseling P, et al (2021) The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro-Oncol 23(8):1231–1251 Morin O, Chen WC, Nassiri F, et al (2019) Integrated models incorporating radiologic and radiomic features predict meningioma grade, local failure, and overall survival. Neuro-Oncol Adv 1(1):vdz011 Ostrom QT, Cioffi G, Gittleman H, Patil N, Waite K, Kruchko C, Barnholtz-Sloan JS (2019) CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2012-2016. Neuro-Oncol 21(Suppl 5):v1–v100 Park SH, Han K (2018) Methodologic Guide for Evaluating Clinical Performance and Effect of Artificial Intelligence Technology for Medical Diagnosis and Prediction. Radiology 286(3):800–809 Parmar C, Grossmann P, Rietbergen M, Lambin P, Aerts H (2015) Radiomic Machine Learning Classifiers for Prognostic Biomarkers of Head & Neck Cancer. Front Oncol. doi: 10.3389/fonc.2015.00272 Reginelli A, Nardone V, Giacobbe G, Belfiore MP, Grassi R, Schettino F, Del Canto M, Grassi R, Cappabianca S (2021) Radiomics as a New Frontier of Imaging for Cancer Prognosis: A Narrative Review. Diagnostics 11(10):1796 S S, Pendem S, K P, S SN, Menon GR, Priyanka -, B D (2025) Machine learning based radiomics approach for outcome prediction of meningioma – a systematic review. doi: 10.12688/f1000research.162306.1 Sahm F, Schrimpf D, Stichel D, et al (2017) DNA methylation-based classification and grading system for meningioma: a multicentre, retrospective analysis. Lancet Oncol 18(5):682–694 Simpson D (1957) The recurrence of intracranial meningiomas after surgical treatment. J Neurol Neurosurg Psychiatry 20(1):22–39 Tibshirani R (1996) Regression Shrinkage and Selection Via the Lasso. J R Stat Soc Ser B Methodol 58(1):267–288 Ugga L, Perillo T, Cuocolo R, Stanzione A, Romeo V, Green R, Cantoni V, Brunetti A (2021) Meningioma MRI radiomics and machine learning: systematic review, quality score assessment, and meta-analysis. Neuroradiology 63(8):1293–1304 Voß KM, Spille DC, Sauerland C, Suero Molina E, Brokinkel C, Paulus W, Stummer W, Holling M, Jeibmann A, Brokinkel B (2017) The Simpson grading in meningioma surgery: does the tumor location influence the prognostic value? J Neurooncol 133(3):641–651 Zhang J, Yao K, Liu P, et al (2020) A radiomics model for preoperative prediction of brain invasion in meningioma non-invasively based on MRI: A multicentre study. EBioMedicine 58:102933 Zhu Y, Man C, Gong L, et al (2019) A deep learning radiomics model for preoperative grading in meningioma. Eur J Radiol 116:128–134 Zwanenburg A, Vallières M, Abdalah MA, et al (2020) The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology 295(2):328–338 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 20 Mar, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviewers agreed at journal 22 Feb, 2026 Reviewers invited by journal 22 Feb, 2026 Editor assigned by journal 05 Feb, 2026 Submission checks completed at journal 04 Feb, 2026 First submitted to journal 26 Jan, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8704777","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":586270434,"identity":"abd61592-9a10-4332-a87c-01478b958df7","order_by":0,"name":"Mateo Moreno Gómez","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA40lEQVRIiWNgGAWjYNCDDxDKgIAyZgSTcQZENQlamHmI0aLbfv7ggw8M2+TM23sPPrZt+5PHwN68TYKhxganFrMzycyGMxhuG8ucOZdsnNtmUMzAc6xMguFYGm4tB5LZpHkYbifOkMgxkwZqSWwAMiQYGw7j1nL+MftvsBb5N2bSliAtQAZ+LTeS2ZghtvCYSTOCbeEhpOWxseQMg9vGEjw5xoY954yL2XjSii0S8PnlfOLDDx8qbstJsJ8xfPCjTC6Pn/3wxhsf8IQYBCBFRAIbmCSgAQWQpHgUjIJRMApGBgAAiqlK7YLluxQAAAAASUVORK5CYII=","orcid":"","institution":"Technological University of Pereira","correspondingAuthor":true,"prefix":"","firstName":"Mateo","middleName":"Moreno","lastName":"Gómez","suffix":""},{"id":586270435,"identity":"446248fc-9221-4967-993f-03696b8a2174","order_by":1,"name":"Iván Mauricio Herrera Mora","email":"","orcid":"","institution":"Technological University of Pereira","correspondingAuthor":false,"prefix":"","firstName":"Iván","middleName":"Mauricio Herrera","lastName":"Mora","suffix":""},{"id":586270436,"identity":"be17df08-2c35-4172-844b-b01e0e3e8277","order_by":2,"name":"Adriana Milena Páez Rodríguez","email":"","orcid":"","institution":"Technological University of Pereira","correspondingAuthor":false,"prefix":"","firstName":"Adriana","middleName":"Milena Páez","lastName":"Rodríguez","suffix":""}],"badges":[],"createdAt":"2026-01-27 02:08:26","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8704777/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8704777/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102377412,"identity":"f5f539cd-ee5f-439e-b758-0b4d2143eebf","added_by":"auto","created_at":"2026-02-11 05:48:23","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":8846335,"visible":true,"origin":"","legend":"\u003cp\u003e1Semi-automated meningioma segmentation on contrast-enhanced T1-weighted MRI using 3D Slicer\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/c9ebdbe79002d97d6d915954.jpg"},{"id":102398441,"identity":"4fb6e188-0f45-4b5d-8e7a-f088d7a5c36d","added_by":"auto","created_at":"2026-02-11 10:22:43","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":6031437,"visible":true,"origin":"","legend":"\u003cp\u003eThree-dimensional reconstruction of the segmented meningioma\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/a3c9f7580547889c4926cda2.jpg"},{"id":102377410,"identity":"91af9e3b-0049-425e-a915-509956917c81","added_by":"auto","created_at":"2026-02-11 05:48:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1642652,"visible":true,"origin":"","legend":"\u003cp\u003eDiscrimination performance of the final model (LASSO + LDA) under nested LOOCV\u003c/p\u003e","description":"","filename":"Fig3ROCPRPanels.png","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/4741effc560907c5e677d364.png"},{"id":102745225,"identity":"4c68f7d5-3f11-42b5-9571-568514fc382d","added_by":"auto","created_at":"2026-02-16 08:45:00","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":3624620,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix and calibration analysis of the final model (LASSO + LDA)\u003c/p\u003e","description":"","filename":"Fig4ConfusionCalibration.png","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/552c74e8c29edf3ee5095823.png"},{"id":102377414,"identity":"78608b52-93dd-40b4-b02a-b0da1c6ece1a","added_by":"auto","created_at":"2026-02-11 05:48:23","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":848518,"visible":true,"origin":"","legend":"\u003cp\u003ePreoperative variables most associated with the surgical endpoint (Simpson grade I vs II)\u003c/p\u003e","description":"","filename":"Fig5Top10PointBiserial.png","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/08766a918980eb8fdea814d6.png"},{"id":102377411,"identity":"875e8170-5df3-4399-ab1b-82ab4f0144ac","added_by":"auto","created_at":"2026-02-11 05:48:23","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":2703705,"visible":true,"origin":"","legend":"\u003cp\u003e6 Correlation structure of top-ranked predictors\u003c/p\u003e","description":"","filename":"Fig6CorrelationHeatmap.png","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/46768a258de4306b97ea4dc4.png"},{"id":102751353,"identity":"d5c0e2c5-647b-4f76-b329-9d72bfb0472e","added_by":"auto","created_at":"2026-02-16 09:25:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":27076555,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8704777/v1/d30e12f9-a1c8-4578-8513-81f6d30e6b86.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"MRI radiomics-based machine learning for preoperative prediction of Simpson grade in intracranial meningiomas: a pilot study","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eMeningiomas are the most frequent primary intracranial tumors in adults, with an estimated annual incidence of approximately 5.3 per 100,000 and a sustained increase with aging [15]. The 2021 World Health Organization (WHO) classification recognizes three histological grades with significant prognostic implications; specifically, grades II–III exhibit greater biological aggressiveness and higher recurrence rates compared with grade I [13]. Surgical resection remains the cornerstone of management, with adjuvant radiotherapy reserved for selected high-risk scenarios [2]. The extent of resection, traditionally quantified using the Simpson grading system, is consistently associated with long-term tumor control and risk of relapse [21]. In addition, anatomical and morphological factors, including tumor location and volume, condition resectability and preoperative planning [12, 24].\u003c/p\u003e\n\u003cp\u003eLimitations of conventional visual assessment on magnetic resonance imaging (MRI) have motivated the use of radiomics and machine learning (ML) to extract quantitative imaging phenotypes (e.g., intensity, texture, and shape) that are not readily appreciable to the naked eye [1]. In oncology, radiomics has been positioned as a quantitative frontier for outcome prediction and risk stratification [18]. In meningiomas specifically, radiomics and ML approaches have been explored for non-invasive grading and related clinically relevant endpoints, including texture-based multiparametric differentiation and enhanced T1-weighted imaging models [4, 5, 9, 11, 26], as well as prediction of invasive behavior and outcome-relevant endpoints such as local failure [14, 25]. These developments parallel advances in tumor biology, such as DNA methylation–based classifications, which refine prognostic heterogeneity and provide a framework for multimodal model design [20]. Furthermore, radiomic signatures have shown potential to outperform purely clinical models in other central nervous system and head-and-neck neoplasms, supporting the conceptual validity of these approaches [10, 17].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDespite these advances, published radiomics studies in meningioma present relevant methodological limitations that hinder clinical translation. A meta-analysis reported an alarmingly low average Radiomics Quality Score (RQS ≈19%) and high heterogeneity across acquisition and segmentation protocols [23]. Consistently, systematic reviews emphasize variability in methodology and endpoints across the meningioma radiomics studies [19]. While performance for non-invasive grading has been reported in the moderate-to-high range in selected studies [9, 11], most current work prioritizes biological grading or related surrogate endpoints, whereas preoperative prediction of surgical resectability remains comparatively underdeveloped. This oversight is clinically important because estimating the probability of achieving an optimal resection (Simpson grade I) directly informs individualized operative planning and patient counseling [3, 12, 24].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo address these limitations, rigorous methodological standards and transparent reporting frameworks are required, including the Image Biomarker Standardization Initiative (IBSI) and TRIPOD+AI guidelines [27, 6]. The aim of this pilot study was to adopt methodological standards to develop and internally validate a preoperative ML model integrating relevant clinical variables and standardized MRI radiomic features to estimate the probability of achieving Simpson grade I versus II resection in intracranial meningiomas, thereby providing a quantitative tool to support neurosurgical decision-making.\u003c/p\u003e"},{"header":"MATERIAL AND METHODS ","content":"\u003cp\u003e\u003cstrong\u003eEthical considerations and data governance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was conducted in accordance with the Declaration of Helsinki and its later amendments. Ethical approval was granted by the Institutional Review Board of Universidad Tecnol\u0026oacute;gica de Pereira (Protocol No. 67-040725). All clinical and imaging data were anonymized prior to analysis; no direct identifiers or re-identification keys were retained. Data was managed in a controlled-access environment with encryption. Given the retrospective design and the exclusive use of de-identified data, the requirement for informed consent was waived by the ethics committee.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy design, setting, and participant selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA retrospective observational study was conducted including adult patients with histopathological confirmed intracranial meningioma who underwent surgical resection at Hospital Universitario San Jorge (Pereira, Colombia) between 2018 and 2024. Approximately 70 potentially eligible cases were identified from institutional surgical and pathology registries.\u003c/p\u003e\n\u003cp\u003eCase ascertainment and cohort assembly followed a pre-specified screening workflow comprising imaging quality control and clinical/pathological verification. DICOM (Digital Imaging and Communications in Medicine) studies underwent quality assessment for suitability for volumetric segmentation and radiomic feature extraction, and clinical/histopathological records were reviewed for completeness and internal consistency. After this process and removal of duplicated/incongruent records, 13 cases met eligibility criteria.\u003c/p\u003e\n\u003cp\u003eEligibility required (i) technically adequate preoperative contrast-enhanced T1-weighted MRI in DICOM format suitable for volumetric segmentation and radiomic analysis, and (ii) complete clinical documentation for outcome ascertainment. Cases were excluded due to insufficient image quality or spatial resolution for reliable volumetric reconstruction and feature extraction (e.g., severe motion artifacts or slice thickness \u0026gt;3 mm), inconsistent histopathological reporting, or duplicated records. One eligible patient with a Simpson grade IV resection was excluded to ensure statistical stability under a binary endpoint in a small-sample, high-dimensional setting. The final analytical cohort comprised 12 patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOutcome definition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe primary endpoint was extent of resection, defined as a binary classification task: Simpson grade I versus Simpson grade II. This endpoint was selected as representing a clinically relevant decision threshold for operative planning and recurrence risk. Other Simpson grades were not modeled due to marginal frequency in the available cohort, which would have compromised estimation stability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImage acquisition and tumor segmentation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePreoperative MRI included contrast-enhanced T1-weighted sequences for all patients. Volumetric tumor segmentation was performed using 3D Slicer (version 5.6.2). Segmentations were generated following a semi-automated workflow (Fig. 1). To minimize inter-observer variability, all masks were independently reviewed by trained neurosurgeons and finalized by consensus adjudication, yielding a single 3D tumor mask per case. Three-dimensional surface reconstructions were generated from the final segmentation for qualitative visualization and documentation of tumor morphology (Fig. 2).\u003c/p\u003e\n\u003cp\u003eFigure 1Semi-automated meningioma segmentation on contrast-enhanced T1-weighted MRI using 3D Slicer\u003c/p\u003e\n\u003cp\u003ea\u0026ndash;c Axial, coronal, and sagittal MRI views showing a skull-base meningioma with homogeneous enhancement\u003cbr /\u003e d\u0026ndash;f Corresponding tumor mask (yellow) generated using semi-automated tools and finalized by consensus review.\u003c/p\u003e\n\u003cp\u003eFigure 2 Three-dimensional reconstruction of the segmented meningioma\u003c/p\u003e\n\u003cp\u003ea,c Axial and coronal MRI views showing the meningioma and the corresponding segmentation mask (yellow)\u003cbr /\u003e b,d Three-dimensional surface reconstruction derived from the final segmentation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRadiomic features extraction and standardization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRadiomic feature extraction was performed using PyRadiomics version 3.1.0 under a single documented configuration file [8]. Preprocessing and feature computation were aligned with IBSI recommendations and included isotropic resampling to 1\u0026times;1\u0026times;1 mm\u0026sup3; using B-spline interpolation, z-score intensity normalization, and texture discretization with a fixed bin width of 31 [27]. A total of 126 preoperative numeric predictors (clinical and radiomic) were assembled, encompassing first-order statistics, shape descriptors, and texture features (GLCM, GLRLM, GLSZM, NGTDM, and GLDM).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine learning model and statistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePreprocessing and feature selection.\u003c/strong\u003e Computational modeling was implemented in Python version 3.10 using scikit-learn. The analytical workflow was designed to prioritize parsimony and mitigate overfitting in a small-sample, high-dimensional setting, while enforcing strict separation between training and held-out data.\u003c/p\u003e\n\u003cp\u003eData-dependent transformations were performed strictly within each training split and then applied to the held-out case to prevent information leakage. Feature scaling used z-score standardization computed exclusively on the training data of each split. To reduce redundancy among predictors, a pairwise correlation-based pruning step was applied prior to feature selection. Three feature selection strategies were benchmarked: ANOVA-KBest, LASSO and Elastic Net. In line with the subject-to-predictor ratio of the analytical cohort, the maximum number of retained predictors was constrained during model screening, with \u0026le;3 predictors as the primary cap and a sensitivity cap up to \u0026le;5 predictors evaluated. Regularization strength was tuned through the inverse penalty parameter C; for LASSO, C=0.2 was assessed as a high-penalty configuration consistent with the final selected model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClassification models\u003c/strong\u003e. Seven supervised classifiers were evaluated to compare linear, probabilistic, and ensemble-based approaches under controlled conditions: Linear Discriminant Analysis (LDA), Gaussian Na\u0026iuml;ve Bayes, Logistic Regression, Support Vector Machines (linear and radial basis function kernels), Random Forest, Nearest Centroid, and k-Nearest Neighbors using Mahalanobis distance. This model set was selected to reflect the classifiers reported in the comparative results and to assess performance under alternative inductive biases.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eValidation strategy.\u003c/strong\u003e A two-stage internal validation framework was applied. First, candidate models (feature selection \u0026times; classifier) were screened using leave-one-out cross-validation (LOOCV) to identify top-performing configurations. Second, the best-performing pipeline was evaluated using nested LOOCV, with an inner loop for hyperparameter tuning and feature selection and an outer loop for unbiased performance estimation on the held-out case, consistent with TRIPOD+AI reporting recommendations [6].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePerformance metrics, calibration, and uncertainty.\u003c/strong\u003e Balanced accuracy was pre-specified as the primary metric to account for class imbalance, complemented by accuracy and macro-averaged F1 score. Discrimination was further characterized using AUROC and AUPRC. Calibration was evaluated using the Brier score and calibration intercept and slope derived from out-of-fold predicted probabilities. Uncertainty was quantified using 1,000 bootstrap resamples of out-of-fold predictions to derive 95% confidence intervals for reported metrics.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003e\u003cstrong\u003eCohort assembly and baseline characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eApproximately 70 patients with diagnosis of intracranial meningioma were initially identified from surgical and pathology registries. After DICOM imaging quality control, verification of clinical and histopathological completeness, 13 cases met the eligibility criteria. For model development and internal validation, the primary endpoint was defined as Simpson grade I versus Simpson grade II. One eligible patient with a Simpson grade IV resection (n=1) was excluded because a sparsely represented third class would yield unstable estimates The final analytical cohort comprised 12 patients (Simpson I, n=4; Simpson II, n=8) with 126 preoperative predictors.\u003c/p\u003e\n\u003cp\u003eRegarding the study cohort, the median age was 61.5 years (IQR 54.8\u0026ndash;65.8; range 34\u0026ndash;80), with a slight female predominance (7/12, 58.3%). The most frequent presenting symptom was headache (10/12, 83.3%), followed by focal neurological deficits (8/12, 66.7%) and seizures (6/12, 50.0%). Most tumors were in the supratentorial compartment (10/12, 83.3%); frontoparietal and frontal (midline/basal) locations were the most common. All patients underwent gross-total tumor removal; Simpson grade I was achieved in 4 cases (33.3%) and Simpson grade II in 8 cases (66.7%). Baseline demographic, clinical, and topographic characteristics are summarized in \u003cstrong\u003eTable 1\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eTable 1 Baseline demographic, clinical, and topographic characteristics of the final cohort\u003c/p\u003e\n\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eValue (N=12)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eAge (years), median [IQR]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e61.5 [54.8\u0026ndash;65.8]\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eRange\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e34\u0026ndash;80\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eSex, n (%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Male\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e5 (41.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Female\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e7 (58.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eClinical presentation, n (%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Headache\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e10 (83.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Focal neurological deficit\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e8 (66.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Seizures\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e6 (50.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eTumor topography, n (%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Supratentorial\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e10 (83.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Frontoparietal\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3 (25.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Frontal (midline/basal)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e2 (16.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Frontotemporal / sphenoid wing\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e2 (16.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Others (occipital, intraventricular, insular/orbital)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3 (25.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Infratentorial\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e2 (16.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Posterior fossa / cerebellopontine angle\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e2 (16.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eExtent of resection (Simpson grade), n (%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Grade I\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e4 (33.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026nbsp; Grade II\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e8 (66.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eComparative model screening (LOOCV)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eComparative screening was performed using leave-one-out cross-validation (LOOCV), combining three feature-selection strategies (ANOVA-KBest, LASSO, and Elastic Net) with correlation-based pruning and a prespecified cap on the number of retained predictors (\u0026le;3 as the primary cap; sensitivity analyses up to \u0026le;5) across seven supervised classifiers. Balanced accuracy was used as the primary selection criterion to account for class imbalance.\u003c/p\u003e\n\u003cp\u003eAcross screened configurations, the best-performing pipeline corresponded to LASSO feature selection (C=0.2; cap \u0026le;3\u0026ndash;5 variables) combined with linear discriminant analysis (LDA), achieving accuracy=0.833, balanced accuracy=0.812, and macro-averaged F1=0.812 on out-of-fold predictions (Table 2). LASSO combined with Gaussian Na\u0026iuml;ve Bayes and LASSO combined with a Mah alanobis-distance k-nearest neighbor classifier showed comparable accuracy during screening but demonstrated lower discrimination metrics in the final evaluation. Support vector machine and random forest models showed consistently inferior performance under the present sample size (Table 2).\u003c/p\u003e\n\u003cp\u003eTable 2 Comparative evaluation of feature-selection and classifier configurations during LOOCV.\u003c/p\u003e\n\u003ctable width=\"685\"\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eFeature selection method\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eClassifier\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eRegularization (C)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eMax. variables\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eBalanced accuracy\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eMacro-F1\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eLDA\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.833\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eLDA\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.833\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eGaussian Na\u0026iuml;ve Bayes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.833\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003ekNN (Mahalanobis)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.833\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003ekNN (Mahalanobis)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.833\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eElastic Net\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eSVM (RBF)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.583\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.556\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.556\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eLogistic Regression\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.667\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.562\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.562\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eLASSO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eSVM (linear)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.667\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.562\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.562\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eANOVA-KBest\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eLDA\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u0026mdash;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.583\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.496\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.496\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eFinal model performance, calibration, and uncertainty \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe best-performing workflow (LASSO, C=0.2; \u0026le;3 variables + LDA) was evaluated using nested LOOCV to minimize optimistic bias by re-optimizing feature selection and tuning within each iteration. On out-of-fold predicted probabilities, the final model achieved accuracy=0.833, balanced accuracy=0.812, macro-F1=0.812, AUROC=0.844, AUPRC=0.867, and Brier score=0.129 (Table 3; Fig. 3). The confusion matrix showed correct classification of 3/4 Simpson grade I cases (sensitivity 0.75) and 7/8 Simpson grade II cases (sensitivity 0.875) (Fig. 4a). Calibration was close to ideal, with a global intercept approximately 0.00 and slope approximately 0.97 (Table 3; Fig. 4b,c).\u003c/p\u003e\n\u003cp\u003eTable 3 Final model performance (LASSO + LDA) after nested LOOCV.\u003c/p\u003e\n\u003ctable width=\"456\"\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eMetric\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.833\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eBalanced accuracy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eMacro-F1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.812\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eAUROC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.844\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eAUPRC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.867\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eBrier score\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003e0.129\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\n\u003cp\u003eGlobal calibration\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd\u003e\n\u003cp\u003eIntercept = 0.000; Slope = 0.973\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFigure 3 Discrimination performance of the final model (LASSO + LDA) under nested LOOCV\u003c/p\u003e\n\u003cp\u003ea,b ROC curves for Simpson grade I and Simpson grade II, respectively\u003cbr /\u003e c,d Precision\u0026ndash;recall curves for Simpson grade I and Simpson grade II, respectively\u003c/p\u003e\n\u003cp\u003eFigure 4 Confusion matrix and calibration analysis of the final model (LASSO + LDA)\u003c/p\u003e\n\u003cp\u003eA Confusion matrix for the binary classification task (Simpson grade I vs II)\u003cbr /\u003e b,c Class-specific calibration curves comparing predicted probabilities with observed event fractions\u003c/p\u003e\n\u003cp\u003eUncertainty estimated by bootstrap resampling (1,000 iterations) of out-of-fold predictions yielded 95% confidence intervals of 0.583\u0026ndash;1.000 for accuracy, 0.500\u0026ndash;1.000 for balanced accuracy, 0.478\u0026ndash;1.000 for macro-F1, 0.583\u0026ndash;1.000 for AUPRC, and 0.026\u0026ndash;0.272 for the Brier score.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature association and correlation structure\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePoint-biserial association analysis indicated that shape- and intensity-derived features concentrated most of the discriminative signal, with Mean.1, Flatness, Strength, Elongation, and Sphericity ranking among the most associated variables (Fig. 5). During nested validation, Strength was selected in 12/12 folds, followed by Flatness and Elongation. A Pearson correlation heatmap of the top-ranked variables highlighted clustering patterns primarily among shape descriptors (Fig. 6).\u003c/p\u003e\n\u003cp\u003eFigure 5 Preoperative variables most associated with the surgical endpoint (Simpson grade I vs II)\u003c/p\u003e\n\u003cp\u003eBar plot of the top ten variables ranked by point-biserial correlation; positive values indicate association with Simpson grade I and negative values with Simpson grade II\u003c/p\u003e\n\u003cp\u003eFigure 6 Fig. 6 Correlation structure of top-ranked predictors\u003c/p\u003e\n\u003cp\u003ePearson correlation heatmap of the variables ranked highest by point-biserial association with the endpoint.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThis pilot study developed and internally validated a preoperative prediction machine learning model to estimate the likelihood of achieving a more radical extent of resection in intracranial meningiomas, using standardized radiomic descriptors extracted from contrast-enhanced T1-weighted MRI together with preoperative clinical variables. In the studied cohort, the selected configuration combining LASSO feature selection with Linear Discriminant Analysis showed good discrimination (AUROC 0.844; AUPRC 0.867) and calibration (slope ≈ 0.97). Given the important role of extent of resection in long-term tumor control and recurrence risk, quantitative tools that complement conventional imaging assessment may support individualized planning and preoperative counseling, particularly in scenarios where dural management and anatomical constraints are expected to influence the operative strategy [12, 21, 24].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFrom a surgical standpoint, achieving a more radical Simpson grade is not solely a matter of removing the intradural mass; it also depends on the treating of the dural attachment and on interface related factors such as adherence and cleavage planes, which are variably appreciated on routine visual inspection and are often clarified intraoperatively [12, 21]. Within this context, the pattern of radiomic features retained by the model is clinically informative. Strength emerged as the most stable predictor across all nested folds (12/12), indicating that a reproducible measure of intratumoral intensity contrast carries consistent signal for the surgical endpoint. Prior radiomics studies in meningiomas have shown that intensity and texture derived descriptors contribute meaningfully to clinically relevant classification tasks, supporting the broader relevance of heterogeneity metrics in this disease [4, 5, 9, 11].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAlthough radiologic–pathologic correlation was not available in the present cohort, a conservative biological rationale is plausible and consistent with the radiomics paradigm that images contain quantifiable information. \u0026nbsp;Greater post contrast heterogeneity may reflect variation in vascular architecture and stromal microstructure, which could translate into differences in tumor consistency, cleavage planes, and the sharpness of the tumor–dura boundary that explain situations of technical difficulty and the feasibility of dural management [7, 12]. This interpretation should be tested in larger cohorts that prospectively capture standardized intraoperative descriptors (e.g., consistency, adherence, sinus proximity, dural involvement) and, when feasible, integrate matched histopathological correlates.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eShape-derived features, particularly Elongation and Flatness, also showed association with the surgical endpoint but lower stability, consistent with redundancy and collinearity among geometric descriptors. Meningiomas that grow along curved dural surfaces or extend across skull-base planes may present elongated or flattened morphologies, which can translate into constrained operative corridors and increased proximity to critical neurovascular structures [24]. Importantly, location is known to modulate both the technical meaning and prognostic value of Simpson grading, reinforcing the plausibility that shape descriptors may partially encode topographic constraints relevant to dural treatment decisions and surgical aggressiveness [9, 12].\u003c/p\u003e\n\u003cp\u003eEnhancement-derived information also appeared relevant in this cohort, as reflected by the association of Mean.1 (post-contrast mean intensity). Enhancement patterns on T1-weighted imaging are influenced by vascularity and microenvironmental factors and have been leveraged in multiple radiomics studies for clinically meaningful tasks in meningiomas, including grading-related classification using conventional and multiparametric MRI [9, 5]. While these signals are biologically plausible, interpretation should remain cautious and in the search of prospective standardization across scanners and acquisition protocols [23, 27].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFrom a methodological standpoint, combining a penalized linear selector with a linear classifier provided a balance between parsimony, stability, and interpretability in a high-dimensional setting with limited sample size. LASSO reduces variance by shrinking weak predictors toward zero and retaining only the strongest signals, which is particularly relevant in radiomics where predictor redundancy and collinearity are common \u0026nbsp;[22]. In contrast, more flexible non-linear approaches often require substantially larger datasets to avoid instability and optimistic performance estimates [16]. The adoption of nested leave-one-out cross-validation further strengthened internal validity by ensuring that feature selection and tuning were performed strictly within each training split, reducing information leakage and providing a less biased estimate of model performance [6, 16]. This design aligns with current expectations for transparent development and evaluation of prediction models in biomedical AI [6].\u003c/p\u003e\n\u003cp\u003eWhen placed in the broader meningioma radiomics literature, this work addresses concerns that have been raised in various reviews. Published evidence has pointed out that many studies show low methodological quality and substantial variability in acquisition, segmentation, and validation, which limits reproducibility and makes results difficult to compare across cohorts [23]. The methodology of the study aimed to reduce variability by using a standardized, fully documented preprocessing and feature-extraction workflow consistent with IBSI-oriented recommendations and by reporting the modeling process in line with current transparency standards for prediction models [6, 27]. Although the findings require cautious interpretation, it lays out a clear, reproducible workflow that can be straightforwardly evaluated in larger, multicenter cohorts and adapted to different clinical settings.\u003c/p\u003e\n\u003cp\u003eClinically, a preoperative estimate that a case is more likely to require less radical dural management may still be useful even when not used as a decision rule. It can help frame preoperative expectations, guide planning around surgical corridors and adjacent neurovascular structures, and support earlier discussion of surveillance and adjunctive strategies when anatomy is likely to limit dural treatment. In that sense, this pilot study demonstrates feasibility under a standardized workflow, showing that routine MRI-derived quantitative descriptors can be mapped to a surgically meaningful endpoint in a way that is reproducible.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLIMITATIONS\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eVarious limitations should be considered. First, the sample size limits precision and increases uncertainty around performance estimates. This is consistent with early radiomics work in meningiomas and neuroimaging, where proof-of-concept models have often been developed in similarly sized cohorts, but it still requires cautious interpretation [11, 23, 26]. Second, the retrospective single-center design constrains generalizability, particularly because MRI acquisition parameters and scanner hardware can influence radiomic feature distributions and reproducibility. Third, the analysis relied on a single primary sequence, and other inputs may capture complementary biological and anatomical information.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFuture work should prioritize external validation in multicenter cohorts, ideally with homogeneous preprocessing and sensitivity analyses across acquisition settings. Expanding the imaging input to include additional MRI sequences may improve robustness and capture complementary features. In parallel, prospective data collection that includes structured intraoperative variables and, when feasible, matched histopathology would enable more direct assessment of why specific imaging descriptors relate to resectability. Finally, implementation studies should evaluate whether preoperative probabilistic estimates meaningfully support surgical planning, counseling, and follow-up decisions, in line with guidance on evaluating clinical performance and impact of AI tools.\u0026nbsp;\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eA standardized radiomics-based modeling approach showed feasibility for preoperative estimation of surgical resectability in intracranial meningiomas. In this cohort, the LASSO + LDA configuration achieved favorable discrimination and calibration under nested internal validation, and feature patterns suggested that heterogeneity- and shape-related descriptors may encode aspects of operative complexity. These results support further testing in larger, multicenter datasets and provide a clear methodological basis for external validation and refinement toward clinically useful, reproducible decision-support tools.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eSTATEMENTS AND DECLARATIONS\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e the authors did not receive support from any organization for the submitted work\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e: the authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman Ethics and Consent to Participate declarations:\u003c/strong\u003e this study involved human participant data. Ethical approval was granted by the Institutional Review Board of Universidad Tecnológica de Pereira (Protocol No. 67-040725). Given the retrospective design and the exclusive use of fully de-identified data, the requirement for informed consent to participate was waived by the ethics committee.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate:\u0026nbsp;\u003c/strong\u003eInformed consent to participate was waived by the Institutional Review Board of Universidad Tecnológica de Pereira (Protocol No. 67-040725) due to the retrospective nature of the study and the use of fully anonymized, de-identified data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for Publication:\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability:\u0026nbsp;\u003c/strong\u003ethe datasets and models generated and/or analyzed during the current study are available upon request and subject to institutional approval. Data will be provided in de-identified form.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMMG\u003c/strong\u003e: Mateo Moreno Gómez\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIMHM\u003c/strong\u003e: Iván Mauricio Herrera Mora\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAMPR\u003c/strong\u003e: Adriana Milena Páez Rodríguez\u003c/p\u003e\n\u003cp\u003eIMHM, AMPR, and MMG contributed to the conceptualization of the study. MMG and IMHM developed the methodology. MMG implemented the software and performed radiomics feature extraction and preprocessing. IMHM and MMG conducted the formal analysis, with input from AMPR. MMG, IMHM, and AMPR contributed to the investigation and verification of clinical/imaging data. IMHM and AMPR performed the 3D Slicer tumor segmentations and generated the 3D reconstructions (Figs. 1–2). MMG prepared the remaining figures and tables (Tables 1–3; Figs. 3–6). MMG and IMHM wrote the first draft of the manuscript. AMPR, MMG, and IMHM reviewed and edited the manuscript. AMPR supervised the study. All authors read and approved the final manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAerts HJWL, Velazquez ER, Leijenaar RTH, et al (2014) Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 5(1):4006\u003c/li\u003e\n\u003cli\u003eApra C, Peyre M, Kalamarides M (2018) Current treatment options for meningioma. Expert Rev Neurother 18(3):241\u0026ndash;249\u003c/li\u003e\n\u003cli\u003eCahill KS, Claus EB (2011) Treatment and survival of patients with nonmalignant intracranial meningioma: results from the Surveillance, Epidemiology, and End Results Program of the National Cancer Institute. Clinical article. J Neurosurg 115(2):259\u0026ndash;267\u003c/li\u003e\n\u003cli\u003eChen C, Guo X, Wang J, Guo W, Ma X, Xu J (2019) The Diagnostic Value of Radiomics-Based Machine Learning in Predicting the Grade of Meningiomas Using Conventional Magnetic Resonance Imaging: A Preliminary Study. Front Oncol. doi: 10.3389/fonc.2019.01338\u003c/li\u003e\n\u003cli\u003eChu H, Lin X, He J, Pang P, Fan B, Lei P, Guo D, Ye C (2021) Value of MRI Radiomics Based on Enhanced T1WI Images in Prediction of Meningiomas Grade. Acad Radiol 28(5):687\u0026ndash;693\u003c/li\u003e\n\u003cli\u003eCollins GS, Moons KGM, Dhiman P, et al (2024) TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 385:e078378\u003c/li\u003e\n\u003cli\u003eGillies RJ, Kinahan PE, Hricak H (2016) Radiomics: Images Are More than Pictures, They Are Data. Radiology 278(2):563\u0026ndash;577\u003c/li\u003e\n\u003cli\u003evan Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, Beets-Tan RGH, Fillion-Robin J-C, Pieper S, Aerts HJWL (2017) Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 77(21):e104\u0026ndash;e107\u003c/li\u003e\n\u003cli\u003eKe C, Chen H, Lv X, et al (2020) Differentiation Between Benign and Nonbenign Meningiomas by Using Texture Analysis From Multiparametric MRI. J Magn Reson Imaging JMRI 51(6):1810\u0026ndash;1820\u003c/li\u003e\n\u003cli\u003eKickingereder P, Burth S, Wick A, et al (2016) Radiomic Profiling of Glioblastoma: Identifying an Imaging Predictor of Patient Survival with Improved Performance over Established Clinical and Radiologic Risk Models. Radiology 280(3):880\u0026ndash;889\u003c/li\u003e\n\u003cli\u003eLaukamp KR, Shakirin G, Bae\u0026szlig;ler B, Thiele F, Zopfs D, Gro\u0026szlig;e Hokamp N, Timmer M, Kabbasch C, Perkuhn M, Borggrefe J (2019) Accuracy of Radiomics-Based Feature Analysis on Multiparametric Magnetic Resonance Images for Noninvasive Meningioma Grading. World Neurosurg 132:e366\u0026ndash;e390\u003c/li\u003e\n\u003cli\u003eLem\u0026eacute;e J-M, Corniola MV, Da Broi M, Joswig H, Scheie D, Schaller K, Helseth E, Meling TR (2019) Extent of Resection in Meningioma: Predictive Factors and Clinical Implications. Sci Rep 9(1):5944\u003c/li\u003e\n\u003cli\u003eLouis DN, Perry A, Wesseling P, et al (2021) The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro-Oncol 23(8):1231\u0026ndash;1251\u003c/li\u003e\n\u003cli\u003eMorin O, Chen WC, Nassiri F, et al (2019) Integrated models incorporating radiologic and radiomic features predict meningioma grade, local failure, and overall survival. Neuro-Oncol Adv 1(1):vdz011\u003c/li\u003e\n\u003cli\u003eOstrom QT, Cioffi G, Gittleman H, Patil N, Waite K, Kruchko C, Barnholtz-Sloan JS (2019) CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2012-2016. Neuro-Oncol 21(Suppl 5):v1\u0026ndash;v100\u003c/li\u003e\n\u003cli\u003ePark SH, Han K (2018) Methodologic Guide for Evaluating Clinical Performance and Effect of Artificial Intelligence Technology for Medical Diagnosis and Prediction. Radiology 286(3):800\u0026ndash;809\u003c/li\u003e\n\u003cli\u003eParmar C, Grossmann P, Rietbergen M, Lambin P, Aerts H (2015) Radiomic Machine Learning Classifiers for Prognostic Biomarkers of Head \u0026amp; Neck Cancer. Front Oncol. doi: 10.3389/fonc.2015.00272\u003c/li\u003e\n\u003cli\u003eReginelli A, Nardone V, Giacobbe G, Belfiore MP, Grassi R, Schettino F, Del Canto M, Grassi R, Cappabianca S (2021) Radiomics as a New Frontier of Imaging for Cancer Prognosis: A Narrative Review. Diagnostics 11(10):1796\u003c/li\u003e\n\u003cli\u003eS S, Pendem S, K P, S SN, Menon GR, Priyanka -, B D (2025) Machine learning based radiomics approach for outcome prediction of meningioma \u0026ndash; a systematic review. doi: 10.12688/f1000research.162306.1\u003c/li\u003e\n\u003cli\u003eSahm F, Schrimpf D, Stichel D, et al (2017) DNA methylation-based classification and grading system for meningioma: a multicentre, retrospective analysis. Lancet Oncol 18(5):682\u0026ndash;694\u003c/li\u003e\n\u003cli\u003eSimpson D (1957) The recurrence of intracranial meningiomas after surgical treatment. J Neurol Neurosurg Psychiatry 20(1):22\u0026ndash;39\u003c/li\u003e\n\u003cli\u003eTibshirani R (1996) Regression Shrinkage and Selection Via the Lasso. J R Stat Soc Ser B Methodol 58(1):267\u0026ndash;288\u003c/li\u003e\n\u003cli\u003eUgga L, Perillo T, Cuocolo R, Stanzione A, Romeo V, Green R, Cantoni V, Brunetti A (2021) Meningioma MRI radiomics and machine learning: systematic review, quality score assessment, and meta-analysis. Neuroradiology 63(8):1293\u0026ndash;1304\u003c/li\u003e\n\u003cli\u003eVo\u0026szlig; KM, Spille DC, Sauerland C, Suero Molina E, Brokinkel C, Paulus W, Stummer W, Holling M, Jeibmann A, Brokinkel B (2017) The Simpson grading in meningioma surgery: does the tumor location influence the prognostic value? J Neurooncol 133(3):641\u0026ndash;651\u003c/li\u003e\n\u003cli\u003eZhang J, Yao K, Liu P, et al (2020) A radiomics model for preoperative prediction of brain invasion in meningioma non-invasively based on MRI: A multicentre study. EBioMedicine 58:102933\u003c/li\u003e\n\u003cli\u003eZhu Y, Man C, Gong L, et al (2019) A deep learning radiomics model for preoperative grading in meningioma. Eur J Radiol 116:128\u0026ndash;134\u003c/li\u003e\n\u003cli\u003eZwanenburg A, Valli\u0026egrave;res M, Abdalah MA, et al (2020) The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology 295(2):328\u0026ndash;338\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"acta-neurochirurgica","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"anch","sideBox":"Learn more about [Acta Neurochirurgica](http://link.springer.com/journal/701)","snPcode":"701","submissionUrl":"https://submission.springernature.com/new-submission/701/3","title":"Acta Neurochirurgica","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Meningioma, Radiomics, Machine learning, Magnetic resonance imaging, Simpson grade, Extent of resection","lastPublishedDoi":"10.21203/rs.3.rs-8704777/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8704777/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Background: meningiomas are the most frequent primary central nervous system tumors, and extent of resection is a key determinant of long-term tumor control. Quantitative preoperative MRI radiomics may support surgical planning by estimating the likelihood of achieving Simpson grade I versus II resection. This study evaluated the feasibility of an MRI radiomics-based machine learning approach for this purpose.\nMethods: a retrospective pilot study (2018–2024) included adults with intracranial meningioma and adequate preoperative contrast-enhanced T1-weighted MRI. Tumors were segmented in 3D Slicer and radiomic features were extracted using PyRadiomics under IBSI-compliant settings. The endpoint was binary (Simpson grade I vs II). Model selection used a selector–classifier benchmark with internal validation via nested leave-one-out cross-validation to mitigate data leakage. Performance was assessed using balanced accuracy, AUROC, AUPRC, and calibration (intercept, slope, Brier score). Uncertainty was estimated using bootstrap on out-of-fold predictions.\nResults: out of approximately 70 screened cases, 12 were included for modeling (Simpson I=4; Simpson II=8) with 126 candidate predictors. The best-performing configuration was LASSO (C=0.2) combined with linear discriminant analysis, achieving balanced accuracy 0.812, AUROC 0.844, AUPRC 0.867, and Brier score 0.129. Calibration was close to ideal (intercept ≈0.00; slope ≈0.97). Bootstrap 95% confidence intervals reflected uncertainty consistent with the small sample size.\nConclusions: this pilot study supports the feasibility of a standardized MRI radiomics-based model for preoperative estimation of surgical resectability (Simpson I vs II) in intracranial meningiomas. Multicenter external validation is required to confirm generalizability and clinical utility.","manuscriptTitle":"MRI radiomics-based machine learning for preoperative prediction of Simpson grade in intracranial meningiomas: a pilot study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-11 05:48:18","doi":"10.21203/rs.3.rs-8704777/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-03-20T13:46:39+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"225495884201511046131057582499628623115","date":"2026-03-20T12:49:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"228828023142787900219324470822527566750","date":"2026-02-22T08:10:25+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-22T05:51:34+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-05T10:14:04+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-05T00:46:59+00:00","index":"","fulltext":""},{"type":"submitted","content":"Acta Neurochirurgica","date":"2026-01-27T01:55:48+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"acta-neurochirurgica","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"anch","sideBox":"Learn more about [Acta Neurochirurgica](http://link.springer.com/journal/701)","snPcode":"701","submissionUrl":"https://submission.springernature.com/new-submission/701/3","title":"Acta Neurochirurgica","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"8cc11e13-aca0-46b9-ac25-9d3b18cf8889","owner":[],"postedDate":"February 11th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-02-22T05:53:14+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-11 05:48:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8704777","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8704777","identity":"rs-8704777","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}