Multimodal MRI-based nomogram integrating clinical-radiological, radiomic, and habitat features to discriminate solitary fibrous tumors from atypical meningiomas.

This study aimed to develop and validate a fused model integrating clinical-radiological, radiomic, and habitat features from magnetic resonance imaging (MRI) for differentiating intracranial solitary fibrous tumors (ISFTs) from atypical meningiomas (ATMs). This retrospective study included 155 patients (39 ISFTs, 116 ATMs), divided into training (n = 108) and testing (n = 47) cohorts. Clinical-radiological factors were evaluated, and a clinical-radiological model (CRM) was constructed using multivariable logistic regression. Radiomic features from whole tumors and habitat features from subregions were extracted to build radiomics model (RM) and habitat model (HM) via four classifiers. A clinical-radiological-habitat-radiomics model (CHRM) incorporating all features was developed. Performance was assessed using area under the receiver operating characteristic curve (AUC) and integrated discrimination improvement (IDI). The CRM identified cyst/necrosis and absent dural tail sign as independent predictors of ISFTs. In the testing cohort, the HM achieved an AUC of 0.829, higher than the CRM (0.696) and RM (0.764). The CHRM demonstrated the best performance, with AUCs of 0.928 (training) and 0.888 (testing), showing significant improvements in IDI (P < 0.05). The HM revealed distinct tumor spatial heterogeneity between ISFTs and ATMs, and the CHRM provided superior preoperative discrimination. These findings provide clinical value for treatment planning, though future validation is warranted to confirm generalizability.