A team of obstetricians and gynecology at the University of Hong Kong's LKS School of Medicine (HKUMED) has developed the world's first artificial intelligence (AI) model that can accurately identify human sperm with the potential for fertilization. This breakthrough could restructure diagnosis and support reproductive treatments around the world.
The AI model evaluates sperm morphology based on its ability to bind to Zona Pellucida (ZP), the outer coat of the egg. By automating processes that traditionally relied on manual and subjective analyses, the model demonstrated a clinical validation accuracy rate of >96%. This innovative approach outperforms traditional methods in terms of speed and reliability, reducing artificial errors and significantly improving the accuracy of male fertility assessments. The findings were published in the International Journal Human Reproduction Open [link to the publication] It won the Silver Award at the 50th Geneva International Invention Fair in 2025.
Infertility is a serious global health concern, affecting about one in six couples of reproductive age around the world, with male factors accounting for 20-70% of cases. According to the World Health Organization (WHO), infertility is predicted to become the third most common disease in the world, following cancer and cardiovascular disease. Adjusted reproductive therapy (art) continues to be the most effective treatment for infertility, but its success rate is limited by the accuracy of existing diagnostic tools.
Limitations of traditional semen analysis
Semen analysis is a standard clinical assessment of the potential fertility of men prior to ART. This analysis, traditionally performed manually under a microscope, evaluates sperm morphology according to WHO guidelines. However, Professor William Yong Shubiwu, a graduate of the Faculty of Obstetrics and Gynecology at the Faculty of Clinical Medicine, explained: This leads to significant variation between individuals and laboratories, making it difficult to standardize sperm quality standards and impair the accuracy of male fertility assessments.
Typical male ejaculation contains 10-200 million motile sperm per milliliter, but only about 7% of these sperm have the potential to fertilize. During natural conception, the selection mechanism within the female genital tract eliminates the underlying sperm, allowing only sperm with fertilization ability to begin fertilization. However, ART labs currently lack an equally efficient method of sperm selection, and instead lead to fertilization methods for ART, such as in vitro fertilization (IVF) and sperm injection (ICSI) in enterocytopathy, depending on parameters from semen analysis such as sperm concentration, motility, and morphology.
Professor Yeung explained: Even with the results of normal semen analysis, 5% to 25% of men still experience low fertilization rates (less than 30%) or complete fertilization failure during IVF. Art failures not only increase the time it takes for a couple to become pregnant, but also increase psychological stress and financial burden.
The world's first AI model to redefine high-quality sperm from an egg perspective
Binding of sperm to ZP is an important first step in fertilization. This layer selectively binds to sperm with normal morphology, intact chromosomes and fertilization ability. This is a natural screening mechanism that ensures that only high-quality sperm will fertilize the egg. “Based on this physiological process, our team has developed a highly automated AI model that analyzes morphological features to accurately determine the percentage of human sperm that can be bound to ZPs and provide a highly reliable assessment of male acceptability,” said Professor Philip Chiu Chiu Gong, an associate professor in the same department and co-leader of the research.
The AI model developed by the HKU team is based on this selective binding mechanism, assessing sperm quality from an egg perspective, and the clinical threshold is established at 4.9%. Men with less than 4.9% of sperm that exhibit binding ability are considered to be at higher risk of fertilization problems. “The AI model provides early warning of fertilization problems and helps identify patients with fertilization disorders with IVF,” added Professor Chiu. “This serves as a new diagnostic tool for detecting fertility issues that traditional semen analysis may overlook, allowing clinicians to tailor more effective treatment plans and improve pregnancy outcomes.”
Deep learning technology delivers promising results
Using advanced deep learning techniques, Hkumed researchers trained AI models with over 1,000 sperm images, achieving accuracy of over 96%. From 2022 to 2024, the team further validated the model by examining over 40,000 sperm images, including 117 men diagnosed with infertility or unexplained infertility. The results confirmed a strong correlation between the percentage of sperm that can bind to ZP and the success rate of the ART procedure.
Professor Chiu emphasized the clinical value of AI when assessing male fertility. “Traditional assessment methods rely heavily on subjective visual judgments with inherent limitations. In contrast, AI models accurately analyze the subtle properties of sperm, allowing them to more accurately predict the likelihood of fertilization.
For couples struggling with infertility, repeated attempts of art are often necessary, and inevitably lead to significant stress, disappointment and financial burdens. The Hkumed team is committed to seeking a breakthrough in medical care. This innovative technology reflects scientific advances, supports couples in need, and helps them realize their reproductive dreams more quickly.
Professor Yeung said, “The advent of AI allows us to assess sperm fertilization ability in a standardized and reproducible way, improve clinical decision-making, and enable personalized treatment plans. This innovation could improve overall management of infertility, reduce fertilization failure rates, and reduce time to pregnancy. We are currently conducting large clinical trials as we hope to further validate the application of AI models and benefit more patients.
About the research team
The study was led by Professor Philip Chiu Chiu Gong of the Obstetrics and Gynecology Department of Clinical Medicine and Gynecology Department and Professor William Yong Shu Biu of the Obstetrics and Gynecology Department. Professor Yu Lequan, Professor of Statistics and Actuarial Sciences, Computational and Data Sciences School, HKU. The main research was conducted with the support of research and clinical teams by Dr. Erica Leung Tsz-King, a postdoctoral researcher at Obstetrics and Gynecology.
Acknowledgments
This study was funded by the Health and Medical Research Fund of the HKSAR Government's Health Bureau and the Center for Advanced Biomedical Instruments. Hong Kong's Hensen University Hospital was a clinical research partner.
