Researchers at Toho University in Japan have introduced an AI-based model that can evaluate male infertility risk solely through blood tests, eliminating the necessity for semen analysis. It is anticipated that this could become a standard infertility screener in the near future, replacing semen-based assessments.

The World Health Organization (WHO) in 2017 reported that men contribute to roughly 50% of global infertility cases. Although semen analysis is crucial for identifying male infertility, it's often not easily accessible outside facilities focused on infertility treatments, making it difficult for men to pursue the test due to the associated barriers.

Under the leadership of Associate Professor Hideyuki Kobayashi from Toho University’s Department of Urology, a team created an AI model capable of predicting male infertility risk through hormone level measurements obtained from blood tests, sidelining the need for semen analysis. Using an AI-generating system that requires no programming expertise, the project was featured in the British journal Scientific Reports. Data samples from 3,662 patients were used to construct the model, which demonstrated a 74% overall accuracy rate. Particularly noteworthy was the model's 100% accuracy in identifying cases of non-obstructive azoospermia, a severe infertility condition.

The study compiled records from 3,662 male participants who underwent semen and hormone testing between 2011 and 2020, measuring semen volume, sperm concentration, and motility, alongside hormone levels for LH, FSH, PRL, testosterone, and E2. The T/E2 ratio was also considered. The total motile sperm count was calculated by multiplying semen volume, sperm concentration, and sperm motility rates. According to the latest WHO semen reference guide from 2021, counts that fell below a threshold of 9.408 x 10⁶ were classified as abnormal (assigned “1”), while higher counts were deemed normal (“0”). With these parameters, the AI model achieved an accuracy of roughly 74%.

Further validation was performed with data collected from 2021 and 2022. In 2021, accuracy was 58% when tested on 188 patients while 2022’s 166 patient data showed 68% accuracy. Importantly, the AI consistently predicted non-obstructive azoospermia accurately across both years, achieving a 100% success rate in these instances.

Associate Professor Kobayashi noted that “This AI-driven prediction model acts as an initial screener before semen analysis. While it does not replace traditional semen tests, its simplicity allows it to be used at facilities that do not specialize in infertility. Especially with its consistent success in detecting non-obstructive azoospermia, which is a serious condition, it can prompt patients toward further testing at specialized institutions, ensuring timely interventions.”

Collaborating with CreaTact, Inc. from Ibaraki Prefecture, Japan, the team is advancing the model toward commercialization through software refinement and data analysis. Kobayashi added, “Our hope is that with this new model, regular clinical labs and health checkup centers will be able to implement male infertility screenings seamlessly, lowering the entry barriers many men face when considering infertility testing.”

The research findings were officially published in Scientific Reports on July 31, 2024.

Journal

Scientific Reports

Method of Research

Observational study

Subject of Research

People

Article Title

A New Model for Determining Risk of Male Infertility from Serum Hormone Levels, without Semen Analysis

Article Publication Date

31-Jul-2024

COI Statement

No conflicts of interest have been declared with regards to this publication.