Background <p>The value of a deep learning (DL) model in distinguishing intracranial solitary fibrous tumors (ISFTs) from angiomatous meningiomas (AMs) and predicting overall survival (OS) of patients with ISFTs have not been systematically assessed. The aim of this study was to develop and validate an MRI-based DL model for distinguishing ISFTs from AMs and predicting&#xa0;OS for patients with ISFTs.</p> Methods <p>(Transformer + Clinic) and clinical models were developed and validated on retrospectively collected preoperative MRI scans of patients with ISFTs and AMs diagnosed between January 2008 and January 2023 at primary cohort (PC) and external validation cohort (EVC). We randomly selected 139 ISFT patients to form a follow-up cohort. The model with the highest mean area under curve (AUC) of receiver operating characteristic (ROC) on both cohorts was identified as optimal model (OM). The follow-up cohort were stratified into high- and low-risk groups based on a fixed cutoff calculated by the OM.</p> Results <p>The OM (Stepglm[both] + GBM) in (Transformer + Clinic) models outperformed the OM (Lasso + GBM) in clinical models in distinguishing ISFTs from AMs on EVC, with an AUC of 0.936 (95% CI 0.891–0.971) and 0.883 (95% CI 0.812–0.943), respectively. Although there was a significant difference in OS between high- and low-risk groups in ISFT patients when using the cutoff value (2.46) calculated from (Stepglm[both] + GBM) (<i>P</i> = 0.04), the cutoff value (5.01) calculated from (Lasso + GBM) had little value in this aspect (<i>P</i> = 0.15).</p> Conclusions <p>The (Stepglm[both] + GBM) model could distinguish ISFTs from AMs. This OM–computed cutoff could be used as a risk stratification tool for patients with ISFTs.</p> Clinical relevance statement <p>The DL model provides a preoperative distinction between ISFTs and AMs on MRI, which may serve as a decision-support tool for preoperative risk stratification and individualized management planning.</p>