WOFAPS 2025 8th World Congress of Pediatric Surgery

Mohammed J Aboud ¹, Manal M Kadhim ², Mustafa J Radif ³

¹ Ministry of Health
² College of Medicine University of Al Qadisiya
³ College of Computer Science and Information Technology University of Al Qadisiya

Background: Ureteropelvic junction obstruction (UPJO) is a leading cause of pediatric hydronephrosis, and while dismembered pyeloplasty is the standard surgical treatment, accurately predicting postoperative success remains a clinical challenge. This study explores the use of machine learning (ML) models to predict surgical outcomes based on preoperative imaging and clinical parameters.

Methods: We retrospectively analyzed 63 pediatric patients (mean age 3.1 ± 2.4 years; 69.8% male) who underwent pyeloplasty between 2005 and 2023. Preoperative assessments included ultrasound parameters, anteroposterior (AP) pelvic diameter, parenchymal thickness, Society for Fetal Urology (SFU) grade, and diuretic renography metrics, specifically half-time (T½) and differential renal function (DRF). Surgical outcomes were classified as successful or failed (persistent obstruction or need for reintervention). Univariate analysis was performed to identify statistically significant predictors, and five supervised ML algorithms were trained using 5-fold cross-validation on an 80:20 train-test split. Our health authorities approved this study.

Results: Of the 63 patients, 52 (82.5%) had successful outcomes. Failures (n = 11; 17.5%) were significantly associated with prolonged T½ (48.1 ± 9.7 min vs. 28.9 ± 10.1 min, p < 0.001), lower DRF (36.2% ± 7.5 vs. 45.3% ± 6.4, p < 0.001), and greater AP diameter (33.5 ± 7.2 mm vs. 24.8 ± 6.5 mm, p = 0.002). Among models tested, XGBoost achieved the best performance (Accuracy: 89.1%; AUC-ROC: 0.92; Sensitivity: 90.9%; Specificity: 90.4%). SHAP analysis confirmed T½, DRF, and AP diameter as the top predictors.

Conclusion: This study demonstrates the potential of ML-driven prediction tools in pediatric urology. The high-performance XGBoost model, leveraging routine imaging metrics, offers a clinically relevant decision-support tool for preoperative risk stratification. Integration into practice may enhance individualized care and postoperative planning in children with UPJO.