Forecasting Stone-Free Status Following Percutaneous Nephrolithotomy Utilizing Explainable Machine Learning.

Area of Science:

• Nephrology
• Urology
• Artificial Intelligence
• Medical Informatics

Background:

• Percutaneous nephrolithotomy (PNL) is a common procedure for kidney stones.
• Predicting stone-free status post-PNL is crucial for patient management.
• Explainable machine learning offers potential for improved prediction accuracy.

Purpose of the Study:

• To develop and validate explainable machine learning models for forecasting stone-free status after PNL.
• To identify key predictors influencing stone-free outcomes.

Main Methods:

• Retrospective analysis of 2144 adult PNL patients (2010-2024).
• Training of Extreme Gradient Boosting (XGBoost), Random Forest, Light Gradient Boosting Machine (LightGBM), and Adaptive Boosting (AdaBoost) models.
• Utilized clinical, radiographic, stone, and surgical data; Synthetic Minority Oversampling Technique for imbalance; SHapley Additive exPlanations (SHAP) for interpretability.

Main Results:

• Overall stone-free rate was 84.8%.
• XGBoost achieved the highest predictive performance (accuracy 0.916, ROC-AUC 0.975).
• SHAP analysis highlighted anatomical anomalies, access sheath size, and stone burden as significant predictors.

Conclusions:

• Explainable AI, particularly XGBoost, accurately predicts PNL stone-free outcomes.
• SHAP enhances model transparency, enabling use as decision-support tools for personalized surgical planning.