Abstract
BACKGROUND: Early identification of molecular subtypes and WHO grades in adult-type diffuse gliomas (ADGs) provides critical evidence for prognostic evaluation and personalized therapeutic decision-making. This study aims to develop and validate radiopathomics models for the prediction of molecular subtypes and WHO grades in ADGs, addressing the limitations of unimodal approaches. METHODS: In this retrospective multicenter study, 499 consecutive ADG patients from three centers (training set: n = 306, testing set: n = 132, external validation set: n = 61) were included. Radiomics features were extracted from preoperative MRI sequences (T2-FLAIR and CE-T1WI), while pathomics features were derived from whole-slide images (WSIs). Feature selection methods and Multilayer Perceptron (MLP) classifier were performed to construct radiomics, pathomics, and radiopathomics models for molecular subtype classification and ADG grading. The performance of the model was evaluated using receiver operating characteristic (ROC) curves, area under the curve (AUC), accuracy, sensitivity, specificity, and F1 score. Decision curve analysis (DCA) was performed to assess clinical efficacy. The Shapley Additive Explanation (SHAP) analysis was employed to explore the interpretability of models. RESULTS: For discriminating molecular subtypes, the radiopathomics model demonstrated superior performance compared to standalone radiomics or pathomics models, achieving AUCs (macro/micro) of 0.847/0.864 in the testing set, and AUCs (macro/micro) of 0.858/0.867 in the external validation set. For differentiating WHO grades, the radiopathomics model achieved superior performance compared to models based solely on radiomics or pathomics features. The AUCs for the radiopathomics model were 0.849 (95% CI 0.775-0.915) in the testing set and 0.855 (95% CI 0.748-0.945) in the external validation set. DCA confirmed superior net clinical benefit across wider risk thresholds compared to unimodal alternatives. SHAP analysis provided interpretable insights into the predictive significance and contributions of individual features. CONCLUSION: The proposed radiopathomics models demonstrate robust diagnostic performance by synergizing cross-scale features, offering a clinically actionable tool for ADG stratification.