Machine learning algorithms for predicting stone residue and recurrence after lateral decubitus percutaneous...

Area of Science:

• Urology
• Medical Informatics
• Machine Learning

Background:

• Percutaneous nephrolithotomy (PCNL) is a common procedure for kidney and upper ureteral stones.
• Predicting stone residue and recurrence post-PCNL is crucial for patient management.
• Current predictive methods may lack precision, necessitating advanced analytical approaches.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting residual and recurrent stones after PCNL.
• To compare the performance of logistic regression, random forest, and XGBoost algorithms in this predictive task.
• To assess the clinical utility of AI-driven predictions in guiding urological treatment decisions.

Main Methods:

• Retrospective analysis of 271 patients undergoing PCNL for renal and upper ureteral stones.
• Data split into training (n=217) and testing (n=54) sets.
• Development of predictive models using logistic regression, random forest, and XGBoost algorithms.
• Model performance evaluation using accuracy, precision, F1 score, and Area Under the Curve (AUC).

Main Results:

• The XGBoost model demonstrated superior performance for predicting stone residue (AUC: 0.87, Accuracy: 86.8%) and recurrence (AUC: 0.68, Accuracy: 72.4%).
• XGBoost achieved the highest F1 scores for both stone residue (0.866) and recurrence (0.72).
• All tested machine learning models showed potential in predicting stone residue and recurrence post-PCNL.

Conclusions:

• Machine learning models, particularly XGBoost, can effectively predict residual and recurrent stones after PCNL.
• These AI-based predictive models offer valuable support for urologists in making timely and informed treatment decisions.
• Further validation and integration of these models into clinical practice could improve patient outcomes in stone disease management.