Machine learning-based prediction of pregnancy outcomes in couples with non-obstructive azoospermia using micro-TESE for ICSI: a retrospective cohort study

Abstract

Objective: To develop a clinically applicable tool for predicting clinical pregnancy, providing individualized patient counseling, and helping couples with non-obstructive azoospermia (NOA) decide whether to use fresh or cryopreserved spermatozoa for oocyte insemination before microdissection testicular sperm extraction (mTESE). Methods: A total of 240 couples with NOA who underwent mTESE-ICSI were divided into two groups based on the type of spermatozoa used for intracytoplasmic sperm injection (ICSI): the fresh and cryopreserved groups. After evaluating several machine learning algorithms, logistic regression was selected. Using LASSO regression and 10-fold cross-validation, the factors associated with clinical pregnancy were analyzed. Results: The area under the curves (AUCs) for the fresh and cryopreserved groups in the Logistic Regression-based prediction model were 0.977 and 0.759, respectively. Compared with various modeling algorithms, Logistic Regression outperformed machine learning in both groups, with an AUC of 0.945 for the fresh group and 0.788 for the cryopreserved group. Conclusion: The model accurately predicted clinical pregnancies in NOA couples.