Prediction of meningioma brain invasion based on preoperative MRI noninvasive deep transfer learning radiomics model

Motivation: Meningioma brain invasion is a negative prognostic factor closely associated with recurrence, but distinct imaging features on pathology are lacking. Goal(s): We aimed to uncover the biological behavioral characteristics of meningioma brain invasion embedded in tumor images. Approach: We applied a deep migratory learning strategy to construct a 2.5D radiomics model using preoperatively contrast-enhanced T1-weighted images to deeply excavate the biological behavioral characteristics of meningioma brain invasion embedded in tumor images. Results: The combined model incorporating clinical and deep-learning radiomics had the best prediction efficacy for meningioma brain invasion and showed clinical applicability for accurate preoperative predictions. Impact: Our model, which combines clinical and deep-transfer learning radiomics features, demonstrates high efficacy in predicting brain invasion in meningiomas and may contribute to improved prognoses for patients.