New AI method accurately predicts male infertility from hormone levels, bypassing need for semen analysis

By Dr. Liji Thomas, MDReviewed by Benedette Cuffari, M.Sc.Aug 4 2024

A recent study published in Scientific Reports investigates the accuracy of a novel artificial intelligence (AI) screening method for male infertility based on serum hormone levels.  

Study: A new model for determining risk of male infertility from serum hormone levels, without semen analysis. Image Credit: Lightspring / Shutterstock.com

Semen analysis

Infertility affects about 9% of the global population, which amounts to about 72.4 million men and women. Male infertility, which is responsible for 50% of all infertility cases, is diagnosed through semen analysis with hormonal assays.

Semen analysis provides important information on sperm production and maturation in the testes, the patency of the seminal passages, and the secretory profile of glandular cells in the testes. The World Health Organization (WHO) Laboratory Manual for the Examination and Processing of Human Semen defines standards for the analysis of semen parameters.

There are several limitations associated with conventional semen analytical approaches. Sample collection, for example, remains a challenge, as many men worldwide are unwilling to get tested due to social stigma. Furthermore, the manual inspection aspect of sperm analysis is labor intensive and requires skilled individuals. Thus, alternative screening methods for male infertility are urgently needed.

Hormones and sperm production

Normal spermatogenesis depends on testicular and endocrine function, beginning with the hypothalamo-pituitary-testicular axis. Several hormones are involved in sperm production, including luteinizing hormone (LH), follicle-stimulating hormone (FSH), prolactin (PRL), testosterone, and estradiol (E2).

FSH and LH are released from the anterior pituitary gland in response to the secretion of gonadotropin-releasing hormone (GnRH) from the hypothalamus. FSH triggers spermatogenesis through its effect on Sertoli cells in the testes.

Sertoli and Leydig cells secrete the hormones inhibin B and testosterone, respectively. Testosterone is converted to E2 through aromatase. Both inhibin B and E2 inhibit further enhancement of the loop by suppressing hypothalamic and pituitary activity.

With disorders of spermatogenesis, high FSH levels may be observed without corresponding changes in LH and testosterone levels. Both FSH and testosterone are essential for spermatogenesis.

Prior research indicates a strong link between serum hormone levels and semen analysis profiles. The current study utilized machine learning (ML), an AI-congruent data processing technology, to characterize male infertility based on serum hormone levels alone.

What did the study show?

The study included 3,662 patients who previously underwent semen analysis and serum hormone measurements as part of their evaluation for male infertility. The mean age of the study cohort was 36 years.

About 44% of the study patients had oligozoospermia or asthenozoospermia, which reflects a lower abundance of sperm and poorly motile sperm, respectively. Azoospermia is defined as the total absence of sperm and can be further classified as non-obstructive azoospermia (NOA) or obstructive azoospermia (OA). NOA and OA affected 12.2% and 5.7% of the study cohort, respectively.

Cryptozoospermia, which reflects extremely low sperm counts, was present in 46 individuals. Only six patients had an ejaculation disorder.

Patient age and several hormones, including LH, FSH, PRL, E2, and the ratio of testosterone to E2 (T/E2), were used to train the AI models and predict the likelihood of male infertility. The lower limit for normal sperm levels was defined as a total motility sperm count of 9.408 × 10⁶ (1.4 ml × 16 × 10⁶/ml × 42%).

The AI model based on Prediction One showed an area under the curve (AUC) value of 74.4%. In general, larger AUC values reflect superior ML performance. Using the AutoML Tables, the AI model was associated with an AUC receiver operating characteristic (AUC ROC) of 74%, whereas the AUC precision-recall (AUC PR) was 77%.

The ROC curve is obtained by a sensitivity vs. specificity plot, with a higher value indicating better test performance. In contrast, the PR curve balances accuracy with comprehensiveness, thereby optimizing the model's ability to identify abnormal results with greater reliability.

For both AI models, FSH ranked as the best hormone in terms of its predictive value, followed by T/E2 and LH.

The current analysis's predictions of male infertility indicated a 100% match for NOA and MHH, with actual results in both 2021 and 2022. For OA, the match was 70%.

Conclusions

Previous studies have reported the possibility of using AI to predict endocrine status. Some applications have included predicting postoperative outcomes after pituitary surgery for non-functioning pituitary adenomas or elevated parathyroid hormone (PTH) status in American adults.

The AI-based models developed in the current study achieved highly accurate and consistent predictions of male infertility. Moreover, the feature-based ranking tool indicated that FSH was the most useful hormone for predicting male infertility. The importance of T/E2 is also predictable, as aromatase inhibitors have been historically used to treat patients with a high T/E2 ratio.

We would like to position our AI model as a convenient means of screening for male infertility prior to semen analysis.”

Infertile males are more likely to have a greater number of medical conditions or comorbidities than fertile men. Therefore, this type of screening approach can be used to predict both male infertility and overall health. Although the AI model is not expected to immediately replace semen analysis, it could serve as an alternative to home diagnostic kits.