AI Model Developed Using Machine Learning for Predicting Sperm Retrieval in Micro-TESE for Nonobstructive Azoospermia Patients

Abstract

Azoospermia is a severe problem that prevents couples from having their own children through natural pregnancy. In nonobstructive azoospermia (NOA), microdissection testicular sperm extraction (micro-TESE) is required to collect sperm and, at 40%–60%, the sperm retrieval success rate is not very high. Previous studies identified no single clinical finding or investigation that could accurately predict the outcome of sperm retrieval. It would be very valuable to have a factor for predicting the possibility of sperm retrieval in patients with NOA before performing micro-TESE. We retrospectively obtained data from the medical records of 430 patients who underwent micro-TESE from 2011 to 2020. Parameters extracted were age, height, body weight, body mass index, luteal hormone, follicle-stimulating hormone, PRL, total testosterone, E2, T/E2, sperm retrieval, G-band, AZF, medical history, Rt testis, and Lt testis. Prediction One, which does not require coding, was used to create the AI prediction model for sperm retrieval. Prediction One makes the best prediction model using an artificial neural network with internal cross-validation. Prediction One also evaluates the “importance of variables” using a method based on permutation feature importance. The AUC for the AI model was 0.7246, which is acceptable. In addition, among the variables, T/E2 ratios contributed most to predicting whether sperm retrieval was possible or not. However, the difference in T/E2 between successful and unsuccessful sperm retrieval was not statistically significant. In addition, our analysis of data from 20 patients who underwent micro-TESE in 2021 found that in 85%, the actual result matched the result predicted using our novel AI model. We created an AI model for predicting sperm retrieval in patients with NOA before undergoing micro-TESE. In addition, we found that T/E2 ratios contributed most to predicting possibility of sperm retrieval in NOA using machine learning.