An Innovative Machine Learning-Based Algorithm for Diagnosing Pediatric Ovarian Torsion

Area of Science:

• Medical imaging and diagnostics
• Machine learning in healthcare
• Pediatric surgery

Background:

• Ovarian torsion is a gynecologic emergency in children.
• Accurate and timely diagnosis is crucial to preserve ovarian function.
• Current diagnostic methods can be challenging, necessitating improved approaches.

Purpose of the Study:

• To develop a machine learning (ML) algorithm for diagnosing pediatric ovarian torsion.
• To integrate physical examination, sonographic findings, and laboratory markers into a diagnostic model.
• To achieve high accuracy in differentiating ovarian torsion from other conditions.

Main Methods:

• Retrospective analysis of 70 ovarian torsion cases and 73 controls (2013-2023).
• Inclusion of sonographic findings, laboratory values (WBC, NLR, SII, SIRI, CRP), and clinical status.
• Development and evaluation of supervised ML algorithms (decision trees, random forests, LightGBM) using 5-fold cross-validation.

Main Results:

• Nausea/vomiting and symptom duration were significant predictors (p < 0.05).
• Elevated WBC, NLR, SII, SIRI, and CRP were highly significant (p < 0.001, p < 0.05).
• Ultrasound findings like ovarian size ratio, medialization, follicular ring sign, and free fluid were significant (p < 0.001).
• The decision tree model achieved 98% accuracy, 98% F1-score, and 100% specificity.

Conclusions:

• The study developed the first ML algorithm integrating clinical, laboratory, and ultrasonographic data for pediatric ovarian torsion.
• The algorithm demonstrated high diagnostic accuracy (>98%).
• This ML approach offers a promising tool for improving the diagnosis of pediatric ovarian torsion.