AI in the IVF lab: What time-lapse imaging, PGT, and automation can do

How to assess medical AI

An AI system learns from data and produces a prediction, classification or ranking for a defined purpose. In embryo assessment, this might be a score that combines developmental features from images. The score is not a neutral law of nature: its value depends on the training data, measurement method, laboratory and intended patient group.

The World Health Organization describes a full evaluation cycle for AI-based medical devices, covering training, validation, assessment and monitoring after implementation. What matters is not only how a model performed in one study, but whether it was tested independently and works under the conditions in which it is actually used. The WHO framework for AI-based medical devices explains this approach.

• Purpose: Which specific decision should the system support?
• Data: For which populations and laboratory conditions was it developed and tested?
• Comparison: Was its performance checked independently of the training data?
• Human oversight: Who reviews the score, and when may the team override it?
• Operation: How are updates, errors and declining model performance detected?

High technical accuracy does not by itself answer the clinical question. Always ask which outcome was measured: agreement with embryologists, implantation, clinical pregnancy or live birth.

Time-lapse imaging and AI in embryo assessment

Time-lapse incubators take images at short intervals while embryos develop. This lets the laboratory team follow development without removing the embryos from the incubator for every check. Some systems add algorithms that score or rank embryos from the image sequence.

Possible benefits for documentation and workflow must be considered separately from treatment success. In its current evidence assessment, the UK Human Fertilisation and Embryology Authority concludes that time-lapse imaging with manual or automated assessment does not improve the chance of having a baby for most patients. It also recognises that the technology can provide continuous observation and practical laboratory benefits. The evidence and evaluated studies are available in the HFEA guidance on time-lapse imaging.

This does not mean that a time-lapse system is useless. It means a clinic should not promise better treatment outcomes when robust evidence is lacking. Useful questions include:

• Is the system used for incubation, documentation, selection or several purposes?
• Which decision does the score change in practice?
• How was the model checked in the clinic's own laboratory?
• Are there additional costs, and what demonstrated benefit supports them?

If you want to understand the treatment steps first, read the guides to IVF, ICSI and IUI.

Automation in the IVF laboratory: Less spectacular, often more important

A modern laboratory consists of more than incubators and microscopes. Its technical infrastructure also includes identity checks, sensors, alarm routes, maintenance records, access controls, data backups and documented approvals. These systems are designed to keep critical steps under control and help prevent mix-ups or unnoticed failures.

The revised 2026 ESHRE recommendations for IVF laboratories cover quality management, patient identification, traceability of gametes and embryos, cryopreservation and emergency procedures. Much of genuine technical innovation lies here: not in one eye-catching device, but in a system that remains traceable during periods of pressure, staff changes or technical faults.

• Traceability: Are samples, work steps, responsible staff and times documented unambiguously?
• Alerts: Who is notified if the temperature deviates or a device fails?
• Resilience: Are backup devices, data backups and tested emergency procedures available?
• Change control: After software updates or new consumables are introduced, does the laboratory check that the process remains stable?
• Auditability: Can deviations be reconstructed later and used to improve the system?

Electronic checks do not replace qualified professionals. Good technology makes responsibility visible and supports the team; it does not move decisions into a black box.

PGT-M, PGT-A and non-invasive tests: Similar abbreviations, different aims

The exact purpose matters when embryos are tested genetically. PGT-M looks for a specific single-gene condition known in a family. PGT-A, by contrast, checks whether the number of chromosomes appears abnormal in the cells tested. These procedures answer different questions and are not interchangeable.

PGT-A is a selection method, not a guarantee of pregnancy or a healthy child. The current HFEA assessment of PGT-A explicitly distinguishes between outcomes. For most patients, it does not automatically increase the chance of having a baby, while the risk of miscarriage may fall in certain analyses. Age, starting point, the number of embryos and the chosen goal all affect the balance.

Non-invasive PGT examines genetic traces in the culture fluid instead of removing cells through a conventional embryo biopsy. The method sounds attractive, but it is not yet an equivalent routine replacement. The ESHRE recommendations on treatment add-ons do not currently recommend non-invasive PGT for routine clinical use.

Whether a genetic test is medically useful and locally permitted depends on its purpose, the individual situation and local rules. Before deciding, ask what exactly is tested, how the team handles uncertain or mosaic findings, and what each possible result would mean. The guide to preimplantation genetic testing explains the terms in more detail.

An add-on is not automatically an innovation

Add-ons are often promoted with a plausible mechanism: a test measures more, a medium is intended to support implantation, or an algorithm is meant to improve selection. A plausible mechanism is a starting point, not proof of a benefit that matters to patients.

ESHRE has assessed add-ons in reproductive medicine across diagnostics, laboratory procedures and clinical applications. The recommendations show that the evidence varies widely between procedures. The HFEA overview of treatment add-ons also assesses specific outcomes and patient groups rather than “innovation” as a marketing label. Its ratings are not worldwide approval rules, but they provide a transparent summary of the evidence reviewed.

Place an offer in a clear category:

• Routine: The procedure is part of an established workflow with a defined purpose.
• Add-on: It is offered in addition to standard care, but the extra benefit may be limited or supported only for particular groups.
• Research: Its safety, accuracy or clinical benefit is still being studied.

Terms such as “AI-powered”, “personalised” or “next generation” do not tell you which category a service belongs to.

Robotics and fully automated laboratory steps: Promising, but not self-running

Automated ICSI, robot-assisted preparation of culture dishes, and automated methods for freezing gametes or embryos are being researched and refined. In 2026, an HFEA scientific committee included AI, robotics and automation in fertility treatment as a dedicated area of its evidence work. The committee's meeting papers show the range of applications under review.

A technology appearing in studies or professional debate does not establish a clinical benefit. The same questions apply to every automated step: Is the purpose clear? Was the process tested against a meaningful comparator? What happens in an edge case or failure? And does the step merely become faster, or does it improve an outcome that matters to patients?

Digital care: A good interface or a good process?

Patient portals, medication plans, appointment management and secure messages can reduce organisational friction. The benefit does not come from the app alone. It depends on whether responsibilities, response times and escalation routes work behind the interface.

• What information does the app show, and which decisions should you not make from it alone?
• Who answers medical questions, and within what time?
• How can you reach the team about urgent symptoms or outside regular hours?
• Which data is collected, how is it used and who receives it?
• Can you download, correct or request deletion of your data?

A digital interface helps when it makes a good care process visible and accessible. It cannot repair missing accountability.

Wearables and LH tests: A timing window, not a diagnosis

Wearables and cycle apps can collect temperature, sleep and other trend data. These patterns may help with timing, but they cannot reliably explain why pregnancy has not occurred. A urine LH test also does not measure ovulation itself. It detects the rise in luteinising hormone that typically appears about one to one and a half days beforehand. The FDA information on ovulation tests clearly separates this prediction from a broader assessment of fertility.

For practical context, read the guides to ovulation, LH tests and the comparison of ovulation tracking devices.

Checklist: Eight questions for a clinic or provider

1. Which specific problem is the technology intended to solve?
2. Is the goal a better workflow, more accurate selection or a better treatment outcome?
3. Which outcome was actually measured in studies?
4. Which patient groups and laboratory conditions do the data apply to?
5. How was the system checked locally, and how is its performance monitored?
6. Who remains responsible when a person and the software reach different conclusions?
7. What risks, possible errors and alternatives are there?
8. What additional costs arise, and what demonstrated benefit supports them?

A good answer does not need to sound technical. It should be specific enough for you to understand the purpose, limits and consequences.

Conclusion

The strongest innovation in an IVF laboratory is not necessarily the most eye-catching algorithm. Reliable identity checks, stable equipment, traceable data and clearly assigned responsibilities often provide greater technical value. AI, time-lapse imaging and genetic analysis can complement these systems, but only an appropriate endpoint can show whether they also improve treatment outcomes.

This article examines technology from a software and product development perspective. It does not replace individual medical or genetic advice. If an add-on could affect your treatment plan, embryo selection or another far-reaching decision, ask a qualified treatment team to explain its benefits, limits and alternatives.