Machine learning–based personalized prediction of sperm retrieval in patients with non‐obstructive azoospermia prior to microdissection testicular sperm extraction: A multi‐center cohort study

作者：Xi Yu, Bailing Zhang, Yun Zhang, Lianming Zhao, Defeng Liu, Jiaming Mao, Wenhao Tang, Haitao Zhang, Haocheng Lin, Xiaoyan Wang, Pengcheng Ren, Yanlin Tang, Yuzhuo Yang, Kai Hong, Jingtao Guo, Zhe Zhang, Hui Jiang · 发表于：Andrology · 年份：2025 · DOI：10.1111/andr.70114 · 被引用次数：3 · 研究领域：Sperm and Testicular Function、Ovarian function and disorders、Reproductive Health and Technologies

BACKGROUND: Non-obstructive azoospermia represents the most severe form of male infertility. The heterogeneous nature of focal spermatogenesis within the testes of non-obstructive azoospermia patients poses significant challenges for accurately predicting sperm retrieval rates. OBJECTIVES: To develop a machine learning-based predictive model for estimating sperm retrieval rates in patients with non-obstructive azoospermia. MATERIALS AND METHODS: This multi-center study included more than 2800 men with non-obstructive azoospermia who underwent microdissection testicular sperm extraction. Preoperative clinical variables were used to train, test, and validate multiple machine learning models. The predictive performance of eight models was assessed with several metrics, including area under the receiver operating characteristic curve, overall accuracy, etc. RESULTS: Of the eight models evaluated, Extreme Gradient Boosting, Random Forest, and Light Gradient Boosting Machine consistently outperformed the others. Extreme Gradient Boosting, which achieved the highest mean area under the receiver operating characteristic curve (0.9183), was selected to power SpermFinder-an online calculator for sperm retrieval rates prediction. The model maintained strong discriminatory ability in both validation sets, with an area under the receiver operating characteristic curve of 0.8469 in the internal cohort and 0.8301 in the external cohort. DISCUSSION AND CONCLUSION: By leveraging routine clini...