Predicting orchiectomy in testicular torsion using hybrid machine learning and explainable AI: a web-based clinical
Ali Çift1,2, Hüseyin Kutlu3, Ferhat Çoban4
1Faculty of Medicine, Department of Urology, Adıyaman University, Adıyaman, Turkey. dr.alicift@gmail.com.
World Journal of Urology
|June 8, 2026
Summary
A new machine learning model accurately predicts the need for orchiectomy in testicular torsion (TT). Platelet Distribution Width (PDW) is a novel biomarker, and a web app aids clinical decisions.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Medicine
- Urology
Background:
- Testicular torsion (TT) is a urological emergency requiring prompt diagnosis.
- Distinguishing between testicular salvage (detorsion) and removal (orchiectomy) is critical for patient outcomes.
- Accurate prediction of surgical outcomes aids preoperative planning and patient counseling.
Purpose of the Study:
- To develop a high-accuracy prediction model for differentiating orchiectomy from detorsion in TT using hybrid machine learning (ML) and explainable artificial intelligence (XAI).
- To create an interactive web application and nomogram for practical clinical decision support.
- To identify novel predictive biomarkers for TT surgical outcomes.
Main Methods:
- Retrospective analysis of 117 TT patients (detorsion: 83, orchiectomy: 34).
- Feature selection using a hybrid Particle Swarm Optimization-Grey Wolf Optimization (PSO-GWO) algorithm.
- Model development with CatBoost, addressing class imbalance with SVMSMOTE, and interpretability via SHAP, LIME, PDP, and ICE analyses.
- Development of a seven-feature nomogram and a Python Dash web application.
Main Results:
- The CatBoost model achieved high accuracy (AUC: 0.923, Accuracy: 89.5%).
- Symptom duration was the strongest predictor, followed by MLR and PDW.
- Platelet Distribution Width (PDW) emerged as a novel biomarker with strong discriminatory power (cut-off < 17.9 fL).
- The nomogram showed good discrimination (AUC: 0.818) and calibration; a simplified risk score was developed.
- The web application provided real-time risk calculation.
Conclusions:
- A hybrid ML and XAI model accurately predicts orchiectomy in TT.
- PDW is identified as a novel, practical, and cost-effective biomarker for TT.
- The developed web application and risk scoring system can support clinical decision-making in emergency settings.
- External validation and prospective studies are required before widespread clinical adoption.