Background: Patients with metastatic nonseminomatous germ cell tumours (NSGCTs) are treated with cisplatin-based chemotherapy, followed by postchemotherapy retroperitoneal lymph node dissection (pcRPLND) for patients with residual nodal masses >1 cm. However, 40–50% of patients who undergo pcRPLND are found to have fibrosis/necrotic tissue alone, which may result in overtreatment. Previous prediction models were based on a single method. We hypothesized that a radiomics-based model combined with serum micro-RNAs (miRNAs) including miR371 and miR375 could exhibit better prediction performance in histopathological presence of visible germ cell tumor, teratoma, and fibrosis/necrotic residual retroperitoneal masses. Methods: A retrospective review of 105 patients with 145 residual retroperitoneal masses undergoing pcRPLND for metastatic NSGCTs was performed to develop a prediction model. The dataset was split into separate training cohorts for model construction and validation cohorts for internal validation. On contrast-enhanced axial computed tomography images, 145 corresponding regions-of-interest (ROIs) of residual masses were segmented and 1130 radiomic features per ROI were extracted. After stepwise feature reduction based on reproducibility, variable importance, and correlation analyses, a logistic regression model was established on the selected most important features. Serum miRNA were used to optimize the classifier. The area under receiver operating characteristics curve (AUC), calibration curve, and decision-curve analyses (DCAs) were used to evaluated the prediction model. Results: The trained machine learning model achieved an AUC of 0.772 (95% CI: 0.670-0.875) in the training cohort and 0.764 (95% CI: 0.589-0.938) in the validation cohort. When serum miR371 and miR375 were added to the radiomics signature, the AUC of training cohort was improved to 0.900 (95% CI: 0.826-0.973), and the AUC of validation cohort was improved to 0.939 (95% CI: 0.847-1.000). Nomogram including radiomics score (Radscore) and miRNA was also performed, and the calibration curve analysis and the DCAs confirmed the agreement between prediction and observation and the clinical utility of the radiomics-miRNA prediction model respectively. Conclusions: Radiomics-miRNAs integrated model represents a novel tools for improved prediction of the histopathology of residual retroperitoneal masses in metastatic NSGCTs, aiming at reducing overtreatment. External validation and prospective datasets are needed to further validate this model.