Develop and validate clinical-radiomics models to predict the risk of postoperative bleeding after percutaneous nephrolithotomy for single stone

To develop and validate clinical-radiomics models for predicting the risk of severe postoperative bleeding in patients with solitary renal or upper ureteral stones undergoing percutaneous nephrolithotomy (PCNL), clinical and imaging data of 190 patients who underwent PCNL at a single tertiary care center from January 2022 to March 2024 were retrospectively analyzed. Patients were divided into a bleeding group and a non-bleeding group based on the occurrence of severe postoperative bleeding. Clinical variables with statistically significant differences between the two groups were incorporated into the models. After delineating regions of interest (ROI) on preoperative CT images, radiomics features were extracted, and the least absolute shrinkage and selection operator (LASSO) algorithm was used for feature selection and dimensionality reduction. A total of 12 clinical-radiomics machine learning (ML) models were constructed by combining clinical factors with the selected radiomics features. The predictive performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and decision curve analysis (DCA). The results showed that Linear Support Vector Classifier (Linear SVC), Gradient Boosting (GB), and Logistic Regression (LR) demonstrated superior predictive accuracy and discriminative ability, with GB achieving the best performance. GB, Extreme Gradient Boosting (XGBoost), Linear SVC, Support Vector Machine (SVM), and LR showed balanced sensitivity and specificity. DCA revealed that most of the models in this study have high clinical applicability. In conclusion, the ML models incorporating clinical variables and CT-based radiomics features demonstrate good performance in the early prediction of severe postoperative bleeding in patients with solitary renal or upper ureteral stones undergoing PCNL, and can assist clinicians in making early interventions to enhance the safety of PCNL.