A machine learning algorithm to predict the success of a second microsurgical testicular sperm extraction

Abstract

Introduction: Testicular sperm extraction (TESE) is a common procedure for retrieving sperm in men with azoospermia. However, the success rates of a second TESE following an initial unsuccessful attempt remain low. This study aims to develop and evaluate a machine learning algorithm to predict the success of a second microsurgical TESE (microTESE).

Methods: Medical records of 47 patients who underwent a second microTESE were analyzed. The dataset included variables such as procedure side, histopathology, preoperative Follicle-stimulating hormone (FSH) and testosterone levels, testicular volume, and comorbidities. Supervised machine learning algorithms, including support vector machine (SVM), were employed to predict the success of the second microTESE. The dataset was split into training (80%) and testing (20%) sets.

Results: The SVM model achieved an accuracy of 80% after hyperparameter tuning. Bilateral procedures and longer intervals between surgeries were associated with higher success rates, while a history of cancer correlated with negative outcomes. FSH and testosterone levels were also identified as predictive factors. The SVM model's feature importance analysis highlighted histopathology, varicocele, hormone levels, and the interval between procedures as highly correlated with the success of a second microTESE.

Discussion: The machine learning model accurately predicted the presence or absence of spermatozoa in patients with non-obstructive azoospermia undergoing a second microTESE. The findings are consistent with previous studies and provide valuable insights into the predictive factors for the success of a second microTESE. However, the study's limitations include selection bias and reliance on retrospective data.

Conclusion: The SVM model shows promise in predicting the success of a second microTESE by incorporating factors such as age, hormonal levels, testicular volume, and genetic evaluation. Further validation and refinement are needed to ensure the model's accuracy and applicability across different populations.

Figure 1.

Relationship between the interval between previous and current surgery and microTESE outcome. Patients with shorter intervals tended to have negative results, while longer intervals correlated with positive outcomes, suggesting recovery time improves efficacy.

Figure 2.

Correlation between procedure side (unilateral vs. bilateral) and microTESE outcome. Bilateral procedures were associated with higher success rates.

Figure 3.

Association between prior cancer history and microTESE outcome. All patients with a cancer history had negative results, indicating a potential risk factor for failure.

Figure 4.

Stratification of patients by Follicle-stimulating hormone (FSH) levels. Positive outcomes were more frequent in patients with high FSH (>15.4 mIU/mL), while low FSH correlated with negative results.

Figure 5.

Testosterone level stratification and microTESE outcomes. Patients with low testosterone (<10 nmol/L) had higher positive result rates, whereas normal/high levels were linked to negative outcomes.

Figure 6.

Reiteration of interval-surgery relationship (similar to Fig. 1). Shorter intervals corresponded to negative results, while longer intervals improved success, likely due to recovery time.

Figure 7.

Confusion matrix of the support vector machine (SVM) model after hyperparameter tuning, demonstrating 80% accuracy in predicting microTESE outcomes.

Figure 8.

Feature importance ranking from the SVM model. Histopathology, varicocele, testosterone, FSH levels, and inter-surgery interval were strongly correlated with second microTESE success.