Existing preoperative meningioma grading methods are limited by insufficient utilization of single-modal information and redundant MRI features. To address these issues, we propose MICF-Net, a multimodal fusion network integrating T1WI, T2-FLAIR, and seven clinical-medical features. A multimodal fusion module (MICF-Block) and a cross-modal alignment loss are introduced to enhance feature fusion and semantic consistency. Experiments results on 168 meningioma cases showed that MICF-Net achieved an accuracy of 89.9%, sensitivity of 91.4%, specificity of 86.7%, and AUC of 88.9%, outperforming single-modal and representative multimodal methods. MICF-Net provides an effective approach for preoperative meningioma grading.