Can AI help an embryologist choose the sperm used for ICSI? A study of 61 ICSI cycles and 894 injected eggs found the same reported fertilization rate—79.3%—with AI-assisted and expert manual sperm selection. Early embryo development and blastocyst formation also showed no statistically significant differences. The results suggest a possible role for decision support, but they do not show that AI is better or that it increases pregnancy or live birth rates.
Among the many sperm available during IVF treatment, who chooses the single sperm that will enter an egg? When intracytoplasmic sperm injection, or ICSI, is used, that responsibility belongs to an embryologist. Looking through a microscope, the embryologist considers movement and shape, selects a suitable sperm, and injects it directly into the egg.
What happens when artificial intelligence helps with that decision? Can it identify a better candidate, or does an experienced embryologist’s eye remain difficult to match?
A study reported in Human Reproduction in 2026 compared conventional selection by an embryologist with more than 10 years of experience against selection supported by real-time AI analysis of sperm movement. It did not find a clear advantage for either approach on the laboratory outcomes measured.
Why choosing one sperm matters in ICSI
In conventional IVF, sperm are placed around the eggs and fertilization is allowed to occur. ICSI works differently: an embryologist chooses one sperm and uses a very fine needle to inject it into an egg.
Instead of allowing many sperm to compete to reach the egg, a person makes the decision: ‘This is the sperm we will use.’
The choice therefore matters. Through the microscope, the embryologist assesses how actively a sperm moves, whether it progresses forward, and whether its head and tail have a suitable shape. This is why training and practical experience play such an important role.
Our related article explains what an embryologist assesses when selecting sperm.
How did the AI-assisted system work?
The system used in this study, the Sperm Identification Device (SiD), analyzes moving sperm in microscope images in real time. It tracks how individual sperm move and ranks potential candidates for use.
‘AI selected the sperm’ needs to be understood precisely. The system did not pick up a sperm and inject it into an egg without a person. It analyzed movement and suggested candidates; the embryologist used that information, checked morphology—the sperm’s shape—and made the final selection.
At this stage, AI-assisted sperm selection is therefore a more accurate description than an autonomous AI procedure.
What happened in the 61 treatment cycles?
The researchers studied 61 ICSI cycles at one fertility center between May 2025 and January 2026. Sperm were injected into a total of 894 eggs.
Of these, 459 eggs received sperm selected manually by the experienced embryologist, while 435 eggs received sperm selected with AI assistance. In most cycles, eggs from the same patient were divided between the two approaches.
The primary abstract specifies that all eggs in three cycles received AI-selected sperm. In the remaining cycles, eggs were randomly allocated between the approaches. The authors nevertheless describe the overall design as a prospective observational study.
The reported fertilization rate was 79.3% with manual selection and 79.3% with AI assistance—identical to the first decimal place. There was no statistically significant difference between the groups.
Did embryos develop into blastocysts at different rates?
Fertilization alone does not tell the whole story in IVF. The fertilized egg must continue dividing and developing, and whether it reaches the blastocyst stage, potentially suitable for transfer, also matters.
The reported day-2 embryo development rate was 94.5% with manual selection and 91.9% with AI assistance. Blastocyst formation rates were 53.6% and 51.0%, respectively.
The manual-selection figures were numerically a little higher, but the differences were not statistically significant. Across fertilization, early embryo development and blastocyst formation, the study did not detect a clear difference between AI-assisted selection and selection by an embryologist with more than a decade of experience.
Absence of a statistically significant difference should not be confused with definitive proof that the methods are equivalent in all settings.
This was not a study in which AI outperformed a person
It would be incorrect to conclude that AI now chooses better sperm than an embryologist. In this study, AI assistance did not produce a higher fertilization rate or blastocyst formation rate than expert manual selection.
That does not make the result uninteresting. The comparison was with a highly experienced embryologist, not a beginner. Similar observed fertilization and development outcomes suggest that AI support warrants further investigation.
Embryologists are people, and assessments can vary with their experience and skill. An AI system can repeatedly analyze movement using a consistent framework. Its possible value may therefore lie in reducing variation and supporting standardized decisions, while keeping the embryologist responsible for the final choice.
Another question is what happens when less experienced embryologists use the system. If AI assistance helps them make selections associated with similar laboratory outcomes, its practical value could be greater. That remains a possibility to test, rather than a result established here.
The researchers also called for further evaluation with embryologists who have less experience.
Does this mean higher IVF pregnancy or live birth rates?
No such benefit was established. The outcomes reported here were principally fertilization, early embryo development and blastocyst formation. This was not evidence that choosing sperm with AI assistance led to more pregnancies or a greater chance of taking home a baby.
The study included 61 cycles and 894 eggs from one center. All procedures were carried out by the same senior embryologist. It was reported as a conference abstract in a Human Reproduction supplement, rather than as a large, definitive randomized clinical trial.
The evidence is therefore insufficient to claim that AI selects better sperm and increases IVF success rates.
It is also useful to separate questions about how sperm are selected from questions about when ICSI itself is appropriate. See our article on the evidence for routine ICSI when there is no male factor infertility.
A possible partnership in the embryology laboratory
The study is interesting because AI is beginning to contribute to a task that has long depended on human observation and experience: choosing one sperm from many.
In this comparison, AI assistance did not outperform the experienced embryologist. It did, however, produce similar observed fertilization and early embryo development results.
The next question may be less about whether AI can replace embryologists and more about whether combining an experienced eye with computerized analysis can make sperm selection more consistent and objective.
Human experience and AI analysis could complement one another. Whether that partnership meaningfully improves laboratory practice—and ultimately patient outcomes—requires further study.
Research source and limitations
This article is based on the conference abstract by S. Zakarneh and colleagues at the Razan Center for Infertility, Artificial intelligence-assisted versus manual sperm selection in ICSI cycles: a prospective comparative study, published on July 8, 2026, in Human Reproduction, volume 41, Supplement 1, article deag083.578. DOI: 10.1093/humrep/deag083.578.
This was a single-center prospective observational study. It did not demonstrate that AI-assisted sperm selection increases pregnancy or live birth rates.
Image disclosure: The image used in this article was created with generative artificial intelligence (ChatGPT, OpenAI). Any people depicted are fictional and are not real individuals.
Copyright © 2026 Fertility Information Talk Talk — Fertility News | THE FERTILITY NOTES. All rights reserved. Unauthorized republication, reproduction, redistribution, scraping, AI training, or use of any content—including articles, photographs and images—for data purposes is strictly prohibited. Violations may result in legal liability.