Predicting Flexible Ureteroscopy Outcomes in Renal Anomalies Using Explainable AI
Overview
This study analyzed 569 patients with congenital renal anomalies undergoing flexible ureteroscopy (fURS) for urolithiasis, applying explainable AI models to predict surgical outcomes. The model identified key predictors influencing stone-free rates and complications, providing interpretable insights to guide clinical decision-making in anatomically complex cases.
Background
Flexible ureteroscopy (fURS) is widely used for managing renal and ureteric stones but poses challenges in patients with congenital renal anomalies such as horseshoe, malrotated, or pelvic ectopic kidneys. These anatomical variations complicate access and may reduce stone clearance success. Although fURS is considered safe with low major complication rates, standardized guidelines for these complex cases are lacking. Explainable AI (XAI) offers a promising approach to improve outcome prediction and support clinical decisions by clarifying model predictions.
Data Highlights
Characteristic
Value
Number of patients
569
Mean age (years)
44.51
Female (%)
73.46
Male (%)
26.54
Hypertension (HTN)
122
Diabetes Mellitus (DM)
125
Cardiovascular disease
47
Horseshoe kidney (HSK)
50.62%
Pelvic ectopic kidney (PEK)
22.67%
Malrotated kidney (MK)
26.71%
Multiple stones (%)
59.58%
Stone size < 20 mm (%)
76.80%
Stone density < 1000 HU (%)
56.94%
Significant intraoperative haematuria
34 patients
Abandoned operations
16 patients
Postoperative sepsis
53 patients
On-table stone-free rate
78.56%
3-month CT-KUB stone-free rate
72.06%
Key Findings
Majority of patients had horseshoe kidney (50.62%), followed by malrotated (26.71%) and pelvic ectopic kidneys (22.67%).
Most patients presented with multiple stones (59.58%), predominantly small (<20 mm) and soft stones (density <1000 HU).
Intraoperative complications included significant haematuria in 34 patients and procedure abandonment in 16 cases.
Postoperative sepsis occurred in 53 patients, requiring prolonged intravenous antibiotics.
Stone-free rates were 78.56% on-table and 72.06% at 3-month follow-up CT-KUB.
Explainable AI models, including decision trees and SHAP analysis, identified key preoperative features influencing outcomes, enhancing interpretability and clinical applicability.
Clinical Implications
The use of explainable AI models can aid clinicians in preoperative risk stratification and procedural planning for patients with renal anomalies undergoing fURS. Understanding the influence of anatomical and stone-related factors on outcomes may optimize patient selection and counseling. Additionally, recognizing predictors of complications can improve perioperative management and reduce adverse events.
Conclusion
Explainable AI approaches provide valuable, interpretable insights into predicting flexible ureteroscopy outcomes in patients with congenital renal anomalies. These data-driven tools have the potential to enhance personalized treatment strategies and improve surgical success in complex anatomical scenarios.
References
EAU Endourology Guidelines -- Flexible Ureteroscopy in Renal Anomalies
SHAP and Explainable AI in Clinical Decision-Making -- 2021
