MRI-based radiomics and deep learning model for preoperative prediction of brain invasion in meningiomas

This study aims to develop and validate a predictive model combining radiomics and deep learning (DL) based on magnetic resonance imaging (MRI) for preoperative prediction of brain invasion (BI) in meningioma patients. A total of 438 patients with pathologically confirmed meningiomas from two hospitals were retrospectively included. The preoperative MRI data acquired for all individuals comprised diffusion-weighted imaging (DWI) and T2-weighted imaging (T2WI), in addition to contrast-enhanced T1-weighted imaging (T1C). Feature dimensionality was reduced using univariate logistic regression, correlation analysis, and the Boruta algorithm. A predictive model was constructed using the random forest (RF) algorithm. Receiver operating characteristic (ROC) curves were used to evaluate the predictive performance of different models, and decision curve analysis (DCA) was employed to evaluate the net benefit of the integrated model. Based on these features and routine clinical variables, seven predictive models were constructed. Among the six predictive models built from these selected features, the highest area under the curve (AUC) values were observed in the DL-radiomics (DLR) model, reaching 0.805, 0.795, and 0.793 in the training, internal validation, and external validation sets, respectively. DCA demonstrated a positive net benefit over default strategies within a limited threshold range. In this retrospective study, the DLR model showed moderate discrimination for preoperative prediction of meningioma BI. External calibration was weaker than training set calibration. Prospective validation and clinical impact evaluation are required.