Multimodal Radiomics Combined with Habitat Analysis for Precise Preoperative Differentiation of WHO Grade I Meningioma Subtypes.

BACKGROUND: Preoperative differentiation of World Health Organization (WHO) Grade I meningioma subtypes is clinically needed but limited with conventional magnetic resonance imaging (MRI). METHODS: A retrospective multicenter cohort of 334 patients (162 fibrous, 91 transitional, 81 meningothelial) was enrolled. T1-weighted contrast-enhanced (T1CE) and T2-weighted (T2WI) MRI were analyzed. Radiomics features were extracted and habitat analysis performed using K-Means clustering. Logistic Regression (LR), Random Forest, and XGBoost models combining radiomics and habitat features were developed and validated in an independent test set (n=67). RESULTS: The optimal combined T1CE+T2WI+habitat model with LR achieved an area under the curve (AUC) of 0.863 (95% CI: 0.763-0.954) and accuracy of 0.716 in the test set, outperforming unimodal radiomics (T1CE LR AUC=0.770), multimodal radiomics (T1CE+T2WI LR AUC=0.823), and habitat-only (LR AUC=0.619) models. Habitat analysis identified distinct imaging subregions corresponding to histological characteristics. CONCLUSION: Integrating multimodal radiomics and habitat analysis enables accurate noninvasive preoperative discrimination of WHO Grade I meningioma subtypes, providing a valuable tool for neurosurgical decision-making.