Machine Learning Predicts Stone-Free Outcomes After RIRS for Lower Pole Stones

Overview

This study developed and validated a machine learning model to predict stone-free rates (SFR) following retrograde intrarenal surgery (RIRS) for lower pole renal stones (LPS). Using clinical and anatomical features from 327 patients, the model demonstrated robust predictive performance and identified key factors influencing surgical success.

Background

Lower pole stones represent 25–35% of renal stones and pose challenges for RIRS due to anatomical constraints like acute infundibulopelvic angle (IPA) and narrow infundibulum, which limit access and fragment removal. Despite advances in endoscopic technology and lithotripsy, SFR for LPS remains lower than for stones in other locations. Preoperative prediction of SFR can guide treatment planning, and machine learning offers a novel approach to integrate multiple patient and stone characteristics for improved prediction accuracy.

Data Highlights

Key Findings

• The machine learning model was trained on 135 patients and internally validated on 58 patients from KMUH, then externally validated on 134 patients from SNUH.
• Input features included demographic data, stone characteristics (size, number, HU), and anatomical parameters (infundibular width, pelvic stone angle, infundibular length).
• Stone-free status was defined as no visible stones or residual fragments ≤5 mm on CT one month postoperatively.
• Model performance metrics included accuracy, AUC, recall, precision, F1-score, Cohen’s kappa, and Matthews correlation coefficient, with accuracy and F1-score guiding model selection.
• Feature importance analysis using gain importance and SHAP values identified key predictors positively and negatively associated with stone-free outcomes.

Clinical Implications

The machine learning model enables personalized prediction of stone-free outcomes after RIRS for lower pole stones, facilitating better preoperative counseling and surgical planning. Understanding the impact of anatomical and stone-related factors can help clinicians optimize patient selection and tailor surgical techniques to improve success rates.

Conclusion

Integrating clinical, anatomical, and stone characteristics through machine learning provides an effective tool to predict stone-free outcomes following RIRS for lower pole stones. This approach supports enhanced decision-making and may improve patient outcomes in this challenging clinical scenario.

References

1. Dresner SL et al. -- Role of Infundibulopelvic Angle in Stone-Free Rates
2. Huang Y et al. -- Scoring System for Predicting SFR of Lower Pole Stones