A Novel Nomogram for Predicting Meningioma Grade Based on Radiomics Features and Clinical Characteristics.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Accurate preoperative grading of meningiomas is crucial for effective treatment planning and prognosis.
• Differentiating low-grade (WHO grade I) from high-grade (WHO grade II/III) meningiomas preoperatively remains a challenge.

Purpose of the Study:

• To develop and validate a predictive model using radiomics features and clinical data for preoperative meningioma grading.
• To compare the performance of different machine learning models in differentiating meningioma grades.

Main Methods:

• Retrospective analysis of 288 meningioma cases (191 low-grade, 97 high-grade) with histopathological confirmation.
• Extraction of radiomics features from contrast-enhanced T1-weighted MRI (CE-T1WI) using pyradiomics.
• Feature selection via LASSO regression, followed by evaluation of logistic regression, decision tree, SVM, and adaptive boosting models, with integration of clinical variables (peritumoral edema index, monocyte count).

Main Results:

• Four key radiomics features significantly discriminated between meningioma grades.
• The logistic regression model showed superior performance among the evaluated machine learning algorithms.
• The combined model (radiomics + clinical variables) achieved an AUC of 0.801 in the training set, while the radiomics model demonstrated the best performance in the validation set with an AUC of 0.770.

Conclusions:

• A radiomics-based model effectively predicts high-grade meningioma, outperforming clinical and combined models.
• This approach enhances the precision of preoperative meningioma grading.
• The findings have significant implications for optimizing treatment strategies and patient management.