AI support, not AI authority: a field guide for embryologists
What an AI score in the IVF lab is, what it is not, and five questions to ask before you trust any ranking at the bench.
Artificial intelligence is now part of many IVF laboratories. Used well, it brings consistency to tasks that depend on long hours at the microscope. Used without understanding, it can quietly shift decisions away from the people who are accountable for them. This guide is about the first case.
What an AI score is
An AI score in embryology is the output of a model trained to recognise patterns in images or videos. In ERICA, for example, the score orders the embryos of one procedure by their estimated euploid potential, so the team can review them in a sensible order. It is a structured, repeatable reading of what is visible in the image.
That has real value. Two embryologists looking at the same blastocyst on a busy afternoon will not always grade it the same way. A model reads every image with the same criteria, every time.
What an AI score is not
A score is not a diagnosis, and it is not a prediction of pregnancy. It does not know the patient's history, the endometrium, the transfer technique or anything that happens after the embryo leaves the lab. That is why, at IVF 2.0, we describe our software as AI support, never AI authority: the model proposes an order; the embryologist and the clinical team decide.
Five questions to ask before trusting any ranking
- What exactly was the model trained to estimate? A ranking by euploid potential, by morphology grade or by implantation are different things. Read the intended purpose in the Instructions for Use.
- What does it see, and what does it not? A model working from a single static image cannot see dynamic events. A model working from time-lapse video cannot see what the camera did not capture. Know the input.
- How are the results shown? Score bands with names (not colour alone), and the possibility to review and undo any decision, keep the human in control.
- Is it regulated for that purpose? In the EU, software whose intended purpose is to inform clinical decisions about individual patients can qualify as a medical device under Regulation (EU) 2017/745 (see the MDCG 2019-11 guidance). Ask for its classification and its intended purpose.
- What happens to my own data? Can you see how scores relate to your clinic's recorded outcomes, with sample sizes? Can you export everything?
Building the habit in your team
AI literacy is not a one-off training. A few practices help:
- Keep the score next to your own grading, not instead of it. Disagreements are worth discussing, not hiding.
- Record outcomes consistently. A model's usefulness in your lab is only visible if follow-up data exists.
- Make the review explicit: a report signed by the embryologist states who decided.
- Revisit the Instructions for Use when the software changes version.
Where to start
If you want to see what this looks like in practice, our product tour walks through real ERICA and SiD screens with demo data, and the research page lists the peer-reviewed work behind them.
ERICA is a Class I medical device under Regulation (EU) 2017/745 for clinical decision support. It does not predict pregnancy and does not replace clinical judgement.