AI sperm analysis shows potential in reducing suboptimal fertilization outcomes in assisted reproduction

A research team from the Department of Obstetrics and Gynaecology at the University of Hong Kong (HKU) has developed the world’s first artificial intelligence (AI) model that can accurately predict the fertilization potential of spermatozoa – a breakthrough that could reshape fertilization failure diagnosis and assisted reproductive treatments worldwide.

Using deep-learning techniques, a pretrained AI model (VGG13) was fine-tuned on >1,000 images of zona pellucida (ZP)–bound and –unbound spermatozoa and independently verified with an additional 220 images. The VGG13 model was fine-tuned to distinguish images of spermatozoa capable of binding to the ZP (the crucial first step in fertilization) based on their morphological features, with high sensitivity (97.6 percent), specificity (96.0 percent), accuracy (96.7 percent), and precision (95.2 percent). The model exhibited low learning variance (average accuracy, sensitivity, and specificity: 97.4, 96.0 and 98.5 percent, respectively) across subgroups, with primary emphasis on the sperm head and mid-pieces in all images as indicated by pixel importance. [Hum Reprod Open 2025;2025(3):hoaf024]

The discriminatory performance of the fine-tuned VGG13 was clinically validated on >33,000 sperm images collected from 117 men with male-factor or unexplained infertility in 2022–2024. Overall, the model exhibited excellent generalization ability, as reflected by the strong correlation between the predicted percentages of spermatozoa with ZP binding per sample and their fertilization rates. A clinical threshold of 4.9 percent (specificity, 89.3 percent; sensitivity, 90.0 percent) was established to differentiate sperm samples with normal and defective ZP binding ability. Pairwise comparisons among 30 patients found that the predicted values generated by the model outperformed conventional semen analysis by the team’s in-house embryologists in identifying patients who were likely to experience failure with conventional in vitro fertilization (IVF).

Semen analysis is a standard clinical assessment for male fertility potential before assisted reproduction procedures such as IVF. Conventional semen analysis is performed by microscopically examining individual spermatozoa and manually assessing them based on WHO’s sperm morphology grading.

“This traditional method is not only labour-intensive and time-consuming, but also highly dependent on the subjective visual judgement of laboratory technicians. This leads to significant variations between individuals and across laboratories, making it difficult to standardize sperm quality criteria, undermining the accuracy of male fertility evaluations,” said Professor William Shu-Biu Yeung of the Department of Obstetrics & Gynaecology, HKU, a leader of the team.

“Prediction of the true fertilization potential of sperms is limited with these traditional semen parameters. Complete fertilization failure or low fertilization rates of <30 percent still occur during IVF for 5–25 percent of men,” noted Yeung. “Assisted reproduction technique failure not only prolongs the time to conception, but also increases psychological stress and financial burden.”

“Our AI model precisely analyzes subtle traits in sperms, enabling more accurate prediction of fertilization potential,” said co-leader of the team, Professor Philip Chi-Ngong Chiu from the Department of Obstetrics & Gynaecology, HKU. “It serves as a novel diagnostic tool for detecting fertility issues that may be overlooked in conventional semen analysis, allowing clinicians to tailor more effective treatment plans and improve pregnancy outcomes.”

“The advent of AI allows us to assess sperm fertilization capacity in a standardized, reproducible manner to improve overall infertility management, reduce fertilization failure rates, and shorten the time to pregnancy,” concluded Yeung.