Predicting Postoperative Outcomes in Pediatric Ureteroscopy Using Machine Learning and Explainable AI-EAU Endourology...

Area of Science:

• Pediatric Urology
• Artificial Intelligence in Medicine
• Surgical Outcomes Research

Background:

• Flexible ureteroscopy (fURSL) is a common procedure for pediatric urolithiasis.
• Predicting postoperative complications is crucial for individualized risk assessment and surgical planning.
• Current methods for risk stratification may not fully capture the complexity of patient-specific factors.

Purpose of the Study:

• To evaluate machine learning (ML) algorithms combined with explainable artificial intelligence (XAI) for predicting outcomes after pediatric fURSL.
• To identify key preoperative predictors of adverse postoperative events.
• To enhance individualized surgical risk assessment and treatment planning.

Main Methods:

• Retrospective analysis of 391 pediatric patients undergoing fURSL for urolithiasis.
• Training 15 ML models and a multitask artificial neural network (ANN) to predict fever, hematuria, sepsis, residual fragments (RF), and reintervention.
• Utilizing SHapley Additive exPlanations (XAI) and decision trees for model interpretability.

Main Results:

• Ensemble ML models achieved high accuracy, with Gradient Boosting predicting fever (92.4%), Extra Trees predicting hematuria (91.1%), and XGBoost predicting sepsis (96.0%).
• Key predictors included preoperative infections, stone burden, operative duration, and anatomical anomalies.
• XAI techniques provided transparent and clinically interpretable insights, aligning with medical reasoning.

Conclusions:

• ML models demonstrate significant accuracy in predicting adverse postoperative outcomes in pediatric ureteroscopy.
• Integration with XAI improves model interpretability, supporting clinical decision-making and personalized treatment strategies.
• Further prospective validation is necessary to develop robust and generalizable predictive tools for clinical application.