Artificial intelligence-based prediction of stone-free status in patients undergoing mini-percutaneous nephrolithotomy without retrograde insertion of a ureteral catheter.

PURPOSE: This study aimed to develop and validate an artificial intelligence (AI) predictive model for postoperative stone-free status (SFS) specifically in patients undergoing mini-percutaneous nephrolithotomy without retrograde insertion of a ureteral catheter (mPCNL-nRUC). METHODS: This single-center prospective observational study included 181 patients with upper urinary tract stones who underwent mPCNL-nRUC between March 2019 and March 2025. Preoperative clinical, laboratory, and imaging data were collected. Five machine learning (ML) algorithms were employed to construct SFS prediction models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and compared with Guy's stone score and S.T.O.N.E. nephrolithometry system. The SHapley Additive exPlanations (SHAP) method was used for interpretability analysis of the optimal model. RESULTS: All surgeries successfully established percutaneous renal access without intraoperative conversion to retrograde ureteral catheter insertion. The immediate postoperative stone-free rate (SFR) was 72.93% (132/181). The overall complication rate was 24.31%, with the majority being Clavien‑Dindo grade I or II. Among the five models, the support vector machine (SVM) model demonstrated the best predictive performance with AUC of 0.875, significantly outperforming Guy's stone score (AUC = 0.659, p < 0.001) and S.T.O.N.E. nephrolithometry system (AUC = 0.770, p = 0.005). The best-performing SVM model has been deployed on an openly accessible online platform to facilitate clinical application. SHAP analysis revealed that stone burden, preoperative serum creatinine, preoperative white blood cell count, stone complexity, and hydronephrosis were the most critical predictors of SFS. CONCLUSION: mPCNL-nRUC is a safe, effective, and simplified procedure in selected patients. The interpretable AI model provides a powerful tool for clinical individualized decision-making.