Related Experiment Video
Updated: Jul 26, 2025
06:28
Vessel-Sparing Microsurgical Longitudinal Intussusception Vasoepididymostomy to Treat Epididymal Obstructive Azoospermia
Published on: May 27, 2022
3.8K
A Machine Learning Approach for the Prediction of Testicular Sperm Extraction in Nonobstructive Azoospermia:
Guillaume Bachelot1,2,3, Ferdinand Dhombres3, Nathalie Sermondade1,2
1Saint Antoine Research Center, L'Institut national de la santé et de la recherche médicale UMR 938, Sorbonne Université, Paris, France.
Journal of Medical Internet Research
|June 21, 2023
Summary
Machine learning models can predict the success of testicular sperm extraction (TESE) in men with nonobstructive azoospermia (NOA). The random forest model demonstrated high accuracy, identifying inhibin B and varicocele history as key predictors.
Area of Science:
- Reproductive Medicine and Urology
- Biostatistics and Machine Learning in Healthcare
Background:
- Testicular sperm extraction (TESE) is crucial for male infertility but has limited predictive success rates.
- Existing models lack sufficient power to accurately predict TESE outcomes based on clinical and laboratory parameters.
Purpose of the Study:
- To compare various predictive models for TESE outcomes in nonobstructive azoospermia (NOA) patients.
- To identify the optimal mathematical approach, study size, and relevant biomarkers for predicting TESE success.
Main Methods:
- Analysis of 201 patients (175 training, 26 testing) undergoing TESE, collecting 16 preoperative variables.
- Training and optimization of 8 machine learning models using retrospective data and random search for hyperparameter tuning.
- Evaluation of models on a prospective cohort using AUC-ROC, sensitivity, specificity, and accuracy metrics.
Main Results:
- Ensemble models, particularly random forest, showed superior performance (AUC=0.90, sensitivity=100%, specificity=69.2%).
- A study size of 120 patients was found sufficient for effective model training.
- Inhibin B levels and a history of varicocele were identified as the most significant predictive factors.
Conclusions:
- Machine learning algorithms can effectively predict TESE success in NOA patients.
- Further prospective, multicentric validation is recommended before clinical application.
- Future research should incorporate advanced biomarkers like noncoding RNAs for improved prediction.
