To explore or not: machine learning models for intraoperative decision on testicular exploration in infants under 3 months with incarcerated inguinal hernia.

purposeTo develop a machine learning (ML) model for preoperative prediction of testicular necrosis risk in male infants under 3 months with incarcerated inguinal hernia (IIH), addressing the limitations of current assessment methods.

methodsWe retrospectively analyzed 288 male infants under 3 months with IIH who underwent emergency surgery. Key preoperative variables (testicular blood flow, echotexture, incarceration duration, procalcitonin, neutrophil-to-lymphocyte ratio) were used to train ten ML models. Performance was evaluated using ROC AUC, PR AUC, accuracy, precision, recall, and F1-score. SHAP analysis assessed interpretability.

resultsThe Gradient Boosting model performed best, achieving a ROC AUC of 0.940 and a recall of 0.889. SHAP identified absent testicular blood flow, heterogeneous echotexture, prolonged incarceration, and elevated procalcitonin and neutrophil-to-lymphocyte ratio as top predictors.

conclusionThis ML model predicts testicular necrosis risk preoperatively. By integrating color Doppler ultrasound, serological markers, and clinical data, it offers an interpretable tool to guide selective testicular exploration, potentially optimizing outcomes. However, given the single-center retrospective nature of this study, external validation in multi-center prospective cohorts is required before clinical implementation.