Abstract: Objective: To develop a machine learning model integrating MRI visually accessible Rembrandt images (VASARI) features and apparent diffusion coefficient (ADC) parameters for the preoperative classification of the three subtypes of adult-type diffuse glioma. Methods: This study retrospectively enrolled 206 patients with pathologically confirmed adult-type diffuse glioma and divided them by scanner type into a training set (n=114) and an independent validation set (n=92). Tumor multi-region segmentation was conducted using nnU-Net, followed by extraction of VASARI features and ADC-related parameters. Recursive feature elimination with cross-validation (RFECV), XGBoost feature importance ranking, and incremental feature selection were applied to identify the optimal feature subset, and an XGBoost-based multiclass fusion model was subsequently constructed. Results: Twelve VASARI features and four ADC parameters (ADCmin, ADCmean, rADCmin, and rADCmean) differed significantly among the three subtypes. Eight features were finally selected for model construction. The fusion model achieved accuracies of 96.49% in the training set and 80.43% in the validation set. In the validation cohort, the AUCs for predicting astrocytoma and glioblastoma were 0.96 and 0.97, respectively; both higher than those of the standalone VASARI and ADC models, while the AUC for oligodendroglioma was 0.59. SHAP analysis revealed that diffusion restriction, ADC parameters, and the margin of the non-enhancing region were the primary contributors to the model. Conclusion: The XGBoost-based fusion model integrating MRI-VASARI features and ADC demonstrated a high diagnostic performance in distinguishing glioblastoma and astrocytoma, but its ability to differentiate oligodendroglioma remained limited. Future studies should incorporate additional imaging features, such as calcification, to further improve the model performance.