Predicting sperm retrieval success in salvage testicular sperm extraction: a machine learning perspective.

Abstract

This multicenter retrospective observational study aimed to identify predictive factors for successful secondary testicular sperm extraction (TESE) using advanced machine learning (ML) models. Data from 503 infertile men who underwent secondary TESE between 2021 and 2024 after a previous unsuccessful attempt were analyzed. Preoperative characteristics and laboratory findings were assessed using ten ML algorithms, including Xtreme Gradient Boosting, Random Forest, Gradient Boosting, Decision Tree, AdaBoost, Logistic Regression, Multi-Layer Perceptron, Support Vector Machine, K-Nearest Neighbors, and Naive Bayes. Model performance was evaluated using accuracy, sensitivity, specificity, predictive values, F1 score, Youden Index, and area under the ROC curve. Sperm retrieval succeeded in 211 patients (41.9%) and failed in 292 (58.1%). Mean infertility duration was 7.25 years. Xtreme Gradient Boosting consistently outperformed other algorithms across all performance metrics. Key predictors of successful secondary TESE included preoperative body mass index, TESE location, luteinizing hormone level, and semen volume. Testicular volumes and infertility duration also contributed significantly to prediction accuracy. Incorporating multiple clinical and laboratory parameters into ML-based predictive models can improve surgical planning and preoperative counseling. For men undergoing secondary TESE, these models may identify those with a very low chance of success, reducing unnecessary repeat surgical interventions.