Motivation: Meningioma is the most common intracranial tumor, and resection is the standard treatment. Preoperative prediction of tumor consistency is important for surgical plan and outcome. Goal(s): We established novel radiomics model based on habitat-based features to predict meningioma consistency, which was compared to conventional radiomics analysis. Approach: We used a semi-automated method to segment tumor regions on T1CE images, extracted radiomics features, and applied K-means clustering to analyze voxel heterogeneity, followed by statistical analysis and model validation. Results: In 148 meningioma patients, habitat analysis showed higher AUC values than conventional methods for predicting "soft" (0.88/0.84) and "hard" (0.87/0.78) tumors. Impact: This study firstly implemented habitat analysis in the field of tumor consistency prediction of meningioma. Our results showed the meningioma heterogeneity features not only significantly improved predictive performance, but also provided new insights on tumor proliferation mechanisms of meningiomas.
Tan et al. (Tue,) studied this question.