How AI helps select sperm for ICSI — and what the evidence actually says

In ICSI, an embryologist picks one sperm from a sample that may contain millions and injects it directly into the egg. That choice is made by eye, in seconds, and it is one of the least standardised decisions in the IVF laboratory. Software that measures how each sperm moves is now being built to support it. What it has not yet done is show, in a randomised trial, that it leads to more babies — and neither have most of the older "advanced" sperm selection methods it is often compared with.

This article sets out what the published evidence says about sperm selection for ICSI, where AI fits into that picture, and what laboratories and patients can reasonably expect from it today.

Key takeaways

• The 2026 Cochrane review of advanced sperm selection included five randomised trials (3,752 women). Only hyaluronic acid selection had enough data to judge, and it may make no difference, or a slight increase, to live birth (low-certainty evidence).¹
• The largest single trial, HABSelect (2,772 couples), found no significant difference in term live birth: 27.4% with hyaluronan-selected ICSI versus 25.2% with standard ICSI.²
• For ultra-high-magnification selection (IMSI), a Cochrane review of 13 trials judged all the evidence to be very low quality and concluded it could neither support nor refute clinical use.³
• ESHRE's 2023 add-ons recommendations advise that IMSI, PICSI, hyaluronan binding assays and MACS are not recommended for routine clinical use.⁴
• AI sperm selection is at an earlier stage still. The published clinical data are associations from retrospective analyses and small proof-of-concept studies, not randomised outcome trials.⁵⁶⁷

How is sperm chosen for ICSI today?

Intracytoplasmic sperm injection (ICSI) bypasses the natural competition between sperm. Instead of thousands of sperm meeting the egg, an embryologist selects a single one, immobilises it, and injects it through the egg's outer layer with a fine glass pipette.

The selection itself is a visual judgement. The embryologist watches the prepared sample under the ICSI microscope and looks for a sperm that is moving progressively and appears normally shaped. There is no score and no measurement readout — just a trained eye and a few seconds of observation per candidate.

A 2023 systematic review in Fertility and Sterility described the task plainly: embryologists face selecting a single sperm from millions based on various parameters, a process that can be time-consuming and subjective, and which may even damage the sperm.⁵ That is the problem both advanced selection methods and AI are trying to address.

What are the "advanced" sperm selection methods?

Over the past two decades several techniques have been developed to try to pick sperm that are more mature, structurally intact, or less likely to carry DNA damage. The main ones are:

• Hyaluronic acid (HA) selection, often called PICSI. Mature sperm bind to hyaluronan, a major component of the cells surrounding the egg. Sperm that bind are selected for injection.⁴
• IMSI (intracytoplasmic morphologically selected sperm injection). Sperm are examined at very high magnification — over 6,000× — so that fine features such as vacuoles in the sperm head can be seen before selection.³
• Microfluidic sorting. Sperm swim through tiny channels designed to separate out the better swimmers without centrifugation.⁴
• Magnetic-activated cell sorting (MACS) and surface-charge (zeta potential) methods, which select according to markers of cell death or surface charge.¹⁴

Each is biologically plausible. The question that matters to patients is whether any of them increases the chance of a live birth.

Do advanced sperm selection methods improve live birth rates?

For most of them, the honest answer is that we do not know, and for the best-studied one the effect is small at most.

The most recent Cochrane review, updated in 2026, looked for randomised controlled trials comparing advanced sperm selection with standard IVF or ICSI. After removing four previously included studies because of concerns about the trustworthiness of their data, it included five trials involving 3,752 women.¹ The certainty of the evidence ranged from very low to moderate.

• Hyaluronic acid selection: may result in no difference or a slight increase in live birth (RR 1.09, 95% CI 0.97 to 1.24; one trial, 2,772 participants; low certainty). It probably slightly reduces miscarriage (RR 0.59, 95% CI 0.44 to 0.80; two trials, 3,327 participants; moderate certainty).¹
• Zeta potential selection, microfluidic sorting, and head-to-head comparisons: each was assessed in a single trial, and the reviewers were uncertain of the effect on live birth because the evidence was of very low certainty.¹

The reviewers concluded that further high-quality studies — including data awaited from 19 ongoing trials — are needed before any of these techniques can be recommended for routine practice.¹

The single largest trial behind those figures is worth looking at directly. HABSelect randomised 2,772 couples across 16 UK fertility units to hyaluronan-selected ICSI or standard ICSI. The term live birth rate was 27.4% versus 25.2% (odds ratio 1.12, 95% CI 0.95 to 1.34; p = 0.18) — not a statistically significant difference. The authors concluded that wider use of PICSI was not recommended at present.²

What about high-magnification selection (IMSI)?

IMSI is the closest conceptual relative of image-based AI, because it rests on the same idea: look at sperm in more detail, and choose better.

A Cochrane review of IMSI versus ICSI included 13 randomised trials with 2,775 couples. The reviewers were uncertain whether IMSI improves live birth (RR 1.11, 95% CI 0.89 to 1.39; five trials, 929 couples). A possible increase in clinical pregnancy (RR 1.23, 95% CI 1.11 to 1.37) was based on very low-quality evidence, downgraded for risk of bias, inconsistency, and a strong indication of publication bias.³ Their conclusion was that the current evidence does not support or refute clinical use of IMSI.

This is an important lesson for AI. Seeing more detail is not the same as improving outcomes. A method has to show that the extra information changes decisions in a way that leads to more healthy pregnancies — and that is a much harder bar to clear.

What does ESHRE recommend?

ESHRE's 2023 good practice recommendations on add-ons in reproductive medicine reviewed these methods as part of a wider assessment of 42 add-on practices. The overall evidence base was thin: none of the 42 recommendations could be based on high-quality evidence, and only four on moderate-quality evidence.⁴

For sperm selection specifically, the working group stated that IMSI, physiological ICSI (PICSI), the sperm hyaluronan binding assay and MACS are currently not recommended for routine clinical use. Microfluidics "can be considered", with further research needed.⁴ The document does not make a recommendation on AI-based sperm selection.

How does AI sperm selection work?

AI sperm selection takes a different approach from the methods above. Instead of changing how sperm are prepared or how closely they are magnified, it adds a measurement layer to what the embryologist is already watching.

Typically, a computer-vision system analyses the live video from the ICSI microscope, tracks individual sperm across frames, and calculates quantitative features for each one: how fast it swims in a straight line, how linear its path is, how its head moves, and aspects of its shape. Those measurements are combined into a score, and the highest-scoring candidates are highlighted for the embryologist.

The appeal is the same one we described in our article on how consistent embryo grading is: a measurement is more reproducible than a judgement. Two embryologists watching the same droplet may favour different sperm; a tracking algorithm given the same video will compute the same velocities every time.

What does the evidence on AI sperm selection show so far?

It is early, and the studies are small. Three examples illustrate where the field currently stands.

Association studies. In a retrospective analysis of recorded ICSI procedures, the SiD software analysed 383 individual sperm from 78 ICSI cycles, calculating straight-line velocity, path linearity and a head-movement measure for each. Path linearity and head movement differed significantly between sperm that did and did not achieve fertilisation, and higher overall software scores were associated with successful fertilisation (P = 0.004) and blastocyst formation (P = 0.013).⁶ This shows that the measurements carry information. It does not show that choosing by the score improves outcomes, and the study was carried out by the software's developers.

Automated ICSI. A 2026 proof-of-concept study in Human Reproduction used a sequence of automated systems — covering sperm preparation, sperm selection, laser immobilisation and piezo-ICSI, with several AI models deployed across them — in 11 cases. Automated procedures achieved fertilisation in 64.3% (45/70) of injected eggs, compared with 81% (47/58) for manual procedures on sibling eggs, and five healthy live births followed.⁷ The authors note that the small sample prevented statistical comparison, that some steps still needed operator support, and that the study was sponsored by the company developing the system. It demonstrates feasibility, not superiority.

Reviews. The 2023 systematic review in Fertility and Sterility concluded that AI algorithms have the potential to address the challenges of sperm selection and could provide valuable assistance to embryologists, improving as larger datasets become available.⁵ That is a reasonable summary: promising, plausible, and not yet proven on the outcome that matters.

What is missing is the study design that settled the question for PICSI and IMSI: an adequately powered randomised trial with live birth as the primary outcome. Until one is published, no one can say that AI sperm selection increases live birth rates.

So is AI sperm selection worth using?

It depends on what you expect it to do.

If the expectation is more pregnancies, the evidence does not yet support that claim, and it is worth remembering that more established methods with far larger trials have not clearly delivered it either.¹²³

If the expectation is a more consistent, documented and objective selection process, the case is stronger. Quantifying motility and morphology gives every case the same criteria, creates a record of why a sperm was chosen, and may reduce the time and fatigue involved in screening a difficult sample. Those are laboratory quality benefits in their own right — and they are also the precondition for ever running the outcome trials the field needs.

What this means for your laboratory

• Separate quality claims from outcome claims. Objective measurement, consistency and workflow are legitimate reasons to adopt a tool. Improved live birth rates are a different claim that requires randomised evidence.
• Ask vendors what their score was validated against. Association with fertilisation in a retrospective dataset is a starting point, not a clinical endpoint. Ask whether it was tested on data from clinics that were not involved in building it.
• Keep the embryologist in control. A ranked shortlist should support the injection decision, not replace it — particularly with severely compromised samples, where the software may have little to choose between.
• Audit your own practice. If you adopt any selection aid, record fertilisation, blastocyst and pregnancy outcomes before and after, so your laboratory generates its own evidence rather than relying only on marketing claims.
• Be careful with add-on pricing. ESHRE advises that several sperm selection add-ons are not recommended for routine use.⁴ Patients should be told clearly what the evidence does and does not show.

What this means if you are a patient

If you are having ICSI, the embryologist will choose the sperm for each egg. In standard ICSI this is done by eye, and it is a well-established procedure that has helped a great many people have children.

You may be offered an additional sperm selection method, sometimes at extra cost. It is reasonable to ask which one it is and what the evidence shows. For hyaluronan selection (PICSI), the best evidence suggests little or no difference in live birth, with a probable small reduction in miscarriage.¹ For most other methods, including AI-based selection, the evidence is not yet strong enough to say whether they improve your chances.

None of this means these tools are useless. It means they should be offered honestly — as ways of making the laboratory's work more consistent, not as a promise of a better result.

Where Garbha.ai fits

Garbha.ai is a CDSCO-licensed AI decision-support platform for the IVF laboratory. Its sperm quality and selection module uses computer vision to analyse live microscopy footage, tracking velocity, linearity and morphology so that candidate sperm can be scored against the same criteria in every case and surfaced to the embryologist for injection. The aim is objective motility and morphology assessment, faster candidate identification, and less manual screening fatigue.

The sperm module is currently in validation. As with our embryo assessment module, the embryologist remains the decision-maker, and the software does not diagnose, treat, or guarantee an outcome.

Frequently asked questions

How is sperm selected for ICSI?

In standard ICSI, an embryologist watches the prepared sample under a microscope and chooses a single sperm that is swimming well and looks normally shaped. The sperm is then immobilised and injected directly into the egg. The choice is a visual judgement made in seconds.

Can AI choose the best sperm for ICSI?

AI can measure how each sperm moves and what it looks like, and rank candidates consistently. Retrospective studies show these measurements are associated with fertilisation and blastocyst formation. No randomised trial has yet shown that AI-selected sperm lead to higher live birth rates, so the embryologist's judgement remains central.

Does PICSI improve my chances of having a baby?

The largest trial found no significant difference in term live birth: 27.4% with PICSI versus 25.2% with standard ICSI. The 2026 Cochrane review concluded that hyaluronan selection may make no difference or a slight increase to live birth, and probably slightly reduces miscarriage.

Is IMSI better than ICSI?

The evidence is too weak to say. A Cochrane review of 13 randomised trials rated all the evidence as very low quality and concluded that it neither supports nor refutes the clinical use of IMSI. ESHRE does not currently recommend IMSI for routine use.

Does AI sperm selection check for DNA damage?

Current AI sperm selection tools mainly analyse movement and shape from microscope video. Whether these features reliably reflect DNA integrity in the individual sperm chosen for injection has not been established in randomised outcome trials.

Should I pay extra for an advanced sperm selection method?

Ask your clinic what the method is, what the evidence says, and whether it is recommended in your particular situation. ESHRE currently advises that IMSI, PICSI, hyaluronan binding assays and MACS are not recommended for routine clinical use. A good clinic will explain the evidence openly.

References

1. Garg A, Nakhla O, Steeper M, Afroz A, Lepine S, McDowell SJ, et al. Advanced sperm selection techniques for assisted reproduction. Cochrane Database of Systematic Reviews. 2026;6:CD010461. doi:10.1002/14651858.CD010461.pub4
2. Miller D, Pavitt S, Sharma V, Forbes G, Hooper R, Bhattacharya S, et al. Physiological, hyaluronan-selected intracytoplasmic sperm injection for infertility treatment (HABSelect): a parallel, two-group, randomised trial. The Lancet. 2019;393(10170):416–422. doi:10.1016/S0140-6736(18)32989-1
3. Teixeira DM, Miyague AH, Barbosa MA, Navarro PA, Raine-Fenning N, Nastri CO, Martins WP. Regular (ICSI) versus ultra-high magnification (IMSI) sperm selection for assisted reproduction. Cochrane Database of Systematic Reviews. 2020;2:CD010167. doi:10.1002/14651858.CD010167.pub3
4. ESHRE Add-ons working group; Lundin K, Bentzen JG, Bozdag G, Ebner T, Harper J, et al. Good practice recommendations on add-ons in reproductive medicine. Human Reproduction. 2023;38(11):2062–2104. doi:10.1093/humrep/dead184
5. Cherouveim P, Velmahos C, Bormann CL. Artificial intelligence for sperm selection—a systematic review. Fertility and Sterility. 2023;120(1):24–31. doi:10.1016/j.fertnstert.2023.05.157
6. Mendizabal-Ruiz G, Chavez-Badiola A, Aguilar Figueroa I, Martinez Nuño V, Flores-Saiffe Farias A, Valencia-Murilloa R, et al. Computer software (SiD) assisted real-time single sperm selection associated with fertilization and blastocyst formation. Reproductive BioMedicine Online. 2022;45(4):703–711. doi:10.1016/j.rbmo.2022.03.036
7. Chavez-Badiola A, Mendizabal-Ruiz G, Flores-Saiffe Farías A, Costa-Borges N, Murray A, Alikani M, et al. Automated oocyte retrieval, denudation, sperm preparation, and ICSI in the IVF laboratory: a proof-of-concept study and report of the first live births. Human Reproduction. 2026;41(2):214–230. doi:10.1093/humrep/deaf240

---

Medical disclaimer: This article is for educational purposes and is not medical advice. Decisions about fertility treatment should be made with a qualified reproductive medicine specialist who knows your clinical history.