Fig. 6: Development and validation of a deep learning-based pathomics model and an integrated pathomics–radiomics model for RE subtype prediction in bladder cancer.

A H&E-stained WSIs were analyzed using the RetCCL model to extract deep learning features from the WSIs. B, C The pathomics-based SVM model achieved high accuracy, sensitivity, specificity, PPV, and NPV in both training and validation cohorts. D ROC curves showed AUCs of 0.87 and 0.81 for the training and validation sets, respectively. E, F Confusion matrices demonstrated reliable classification of high and low RE score subtypes. G Representative histopathological images revealed distinct morphology: high RE score tumors showed dense tumor cell clusters, while low RE score tumors exhibited more stromal infiltration. H Schematic overview of the integrated pathomics–radiomics model incorporating treatment response–related parameters. I, J The integrated model showed improved predictive performance (AUC = 0.89 in training and 0.85 in validation) with enhanced diagnostic metrics compared to the pathomics model alone.