MRI delta-radiomics for prediction of tumor response in muscle-invasive bladder cancer patients undergoing neoadjuvant chemotherapy

Abstract Background To investigate whether MRI-based delta-radiomics can predict treatment response in patients with muscle-invasive bladder cancer (MIBC) undergoing neoadjuvant chemotherapy (NAC). Materials and methods In this prospective study, 52 patients with localized MIBC underwent multiparametric MRI (mpMRI) at three time points: pre-treatment, mid-treatment, and post-treatment. Radiomics features were extracted from the manually segmented primary tumor, and relative delta-radiomics were calculated. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) within a nested cross-validation framework. Predictive models were developed using Support Vector Machines with radial basis function kernel (SVM-RBF), Random Forest (RF), and Light Gradient Boosting Machine (LGBM). The discrimination and calibration of different models were evaluated. Clinical complete response (CR) and overall response (OR) were the primary and secondary endpoints, respectively. Results Clinical CR and OR were achieved in 36.5% and 59.6% of patients, respectively. Post-treatment diffusion-weighted imaging (DWI) yielded the highest predictive performance for both CR and OR. For the prediction of clinical CR, the SVM-RBF model achieved an area under the receiver operating characteristic curve (AUROC) of 0.817 using mid-treatment multi-sequence MRI and improved to 0.844 with post-treatment DWI. For OR, the LGBM model demonstrated higher discriminative performance, with AUROCs of 0.887 at mid-treatment using multi-sequence imaging and 0.891 at post-treatment using DWI. Calibration analyses and precision–recall metrics supported the robustness of the best-performing models. Conclusion Delta-radiomics derived from longitudinal mpMRI enables accurate, non-invasive prediction of NAC response in MIBC, with mid-treatment imaging allowing early treatment adaptation during NAC and demonstrating potential clinical utility for personalized decision-making in bladder cancer.