AI Embryo Selection in IVF: How Does It Work?

Picture the moment in the IVF lab. Your eggs have been collected, fertilised, and developed into embryos over a few days. Now, several embryos are growing in the incubator, and to an untrained eye, they may all look similar. But which one has the best potential for successful implantation? AI embryo selection in IVF uses advanced technology to analyse embryo development, assess quality, and help fertility specialists identify the most promising embryo for transfer.

That question has long rested in an embryologist’s trained eye. Lately, it also lands on an algorithm. “AI embryo selection in IVF” is one of the most talked-about tools in modern IVF, and also one of the most misunderstood; plenty of patients assume it can tell whether an embryo is healthy, or that it’s quietly overruling the humans. Neither is quite right. So let’s open up the hood and look at what actually happens, step by step, when AI helps choose an embryo.

The Problem AI Embryo Selection in IVF Is Trying to Solve

Start with why this is hard. With a single embryo, there’s no choice to make. The difficulty shows up when you’ve got several decent ones and have to decide which goes back first because transferring the strongest one sooner can shave months, sometimes more, off the road to a pregnancy.

For decades, that call has been made by eye. An embryologist grades each embryo at set checkpoints: its shape, its cell structure, how neatly it’s developing. Some labs go further and track the timing of each cell division (the field’s word for this is “morphokinetics”). Both methods work. Both also wobble. The same embryo can earn slightly different grades from two embryologists, or from the same one on a frantic day versus a quiet one. Pick the second-best embryo first, and you may be in for another round of waiting. That inconsistency, in a decision this consequential, is the gap AI stepped into.

How AI embryo selection actually works, step by step

How AI embryo selection actually works, step by step

Strip away the mystique, and it’s a fairly logical pipeline. Here’s each stage.

Step 1 – Capturing Images for AI Embryo Selection in IVF

It all begins with embryo images, and there are two broad approaches.

Some systems work from a single still image, one clear photo of the blastocyst, fed to the AI. Life Whisperer is a well-known example. Others use time-lapse: the embryo grows inside a special incubator (the familiar EmbryoScope) that photographs it continuously, every few minutes, over five or six days. That produces a time-lapse movie of the embryo’s early life, and the AI gets to watch the whole thing. iDAScore, from Vitrolife, is the best-known tool in this camp.

Still photo or full movie- that’s the first fork in the road, and it matters, because the movie carries far more information.

Step 2: AI Embryo Selection in IVF Using a Deep-Learning Model

Now the clever part. The AI is a neural network, specifically the kind built to interpret images, a convolutional neural network. The more sophisticated video-based models, like iDAScore, use a design that reads two things at once: what the embryo looks like (its morphology) and how it behaves over time (the rhythm and timing of its divisions). It does this automatically, frame by frame, with nobody having to hand-label anything. No embryologist sitting there annotating. The machine extracts the patterns itself.

Step -3, training on known outcomes

Here’s where the “intelligence” actually comes from. Before any of this touches your embryos, the model is trained on enormous libraries of past embryos whose endings are already known: this one implanted, that one didn’t; this one led to a live birth, that one to a chromosomal abnormality, gathered from many clinics across many countries. Over thousands and thousands of examples, the model learns which visual and developmental signatures tend to go with a good outcome. So when it finally meets a new embryo, it’s comparing what it sees against everything it learned.

Step 4: the score

The output of all that is, in the end, a number. iDAScore, for instance, hands each embryo a score on a scale of 1.0 to 9.9. The higher the number, the higher the model’s estimate of that embryo’s chance of implanting and progressing to a pregnancy. One embryo might come back at 8.4, another at 5.1, a clear, consistent ranking where the human eye might have hesitated.

Step 5: ranking, not deciding

And this is the step the brochures tend to blur. The score doesn’t choose. It ranks. The embryologist takes that ranking, weighs it against everything else they know about your embryos and your case, and makes the decision: which embryo to transfer first, or, if you’re having genetic testing, which ones to prioritise for biopsy. The AI is a sharp, consistent second opinion. The person is still in the chair.

What the score can, and can’t, tell you

This is the bit worth reading twice, because it’s where the biggest misunderstanding lives.

What the score can do: rank your embryos by their likely potential to implant, flag the strongest candidate, bring consistency to a judgment that used to swing between observers, and save time. The scores also tend to correlate with an embryo being chromosomally normal (euploid); higher-scoring embryos are, on average, more likely to be euploid.

What it can’t do: tell you, for certain, that an embryo is genetically normal. That correlation is just a tendency, not a diagnosis, and its accuracy in predicting euploidy is modest. If you need to actually know an embryo’s chromosomal status, that still requires genetic testing (PGT-A), where a few cells are biopsied and analysed in a lab. A camera and an algorithm can suggest; only the biopsy can confirm. Please don’t read an AI score as a green light that skips genetic testing; they answer different questions.

Does AI pick better embryos than an expert?

Here’s the honest state of play, and it’s more sober than the marketing.

The headline trial on this was a large, rigorous study across fourteen clinics in Australia and Europe, comparing iDAScore against standard manual grading for choosing which single blastocyst to transfer. Its finding was striking for what it didn’t show: the study couldn’t demonstrate that the AI was non-inferior to expert embryologists on clinical pregnancy rate. In other words, in that setting, the algorithm didn’t beat skilled humans, and couldn’t even prove it matched them. Other studies are kinder, showing AI landing at parity with experts and improving consistency; in validations, it would have questioned the embryologist’s ranking in a sizeable chunk of cases. But the two biggest trials pointed in different directions.

So where does that leave us? AI embryo selection is genuinely good at making the choice more consistent and faster, and it performs at least as well as a skilled embryologist in many settings. What it has not reliably shown is that it produces more babies than a good embryologist would. And because results differ by tool, “AI embryo selection” can’t be treated as one proven thing; it’s a family of tools of varying quality. Anyone promising it reliably boosts your success is getting ahead of the evidence.

The catches worth knowing

A few honest flags, because they explain why a human stays in the loop:

It can be a black box; a model might rank an embryo top without being able to spell out, in checkable terms, why. It inherits whatever it learned, so a tool trained mostly on one population can falter on another. Bias is documented: embryo-selection algorithms, AI and humans alike, have shown a lean toward male over female embryos, the kind of quirk that needs a person watching. Video-based tools also need the right kit: a time-lapse incubator, which not every lab has. And regulatory clearance varies from tool to tool. None of that makes AI a bad idea. It makes the case that AI with an expert beside it beats AI on its own, every time.

What it means for you as a patient

Here’s the practical takeaway. AI embryo selection can give you a steadier, more objective read on which embryo to transfer, so your result leans a little less on which embryologist is on shift. It might get you to a pregnancy a touch sooner. It’s a useful second opinion in a genuinely hard call.

What it isn’t: a guarantee, a genetic test, or a replacement for your embryologist’s judgment. If a clinic mentions it, the questions worth asking are simple: do you use a single-image or a time-lapse system, which tool, is it validated, and how does it fit with PGT if I’m testing my embryos? A clinic that knows its stuff will answer each one without hand-waving. And hold on to the bigger truth: the lab and the team matter more than the software. Good tools in skilled hands beat flashy tools every time.

The Imprimis IVF approach

At Imprimis IVF, selection technology earns its place by helping patients, not by looking good on a brochure. Tools like time-lapse imaging and AI scoring add real value by making embryo assessment steadier and choice more objective, but they support experienced embryologists, who still make the final decision. Just as important is honesty about the limits: a score can rank embryos and sharpen a judgment call, but it can’t diagnose an embryo’s genetics or promise a particular outcome, and it’s never treated as a substitute for genetic testing when your case needs it. If you want to understand exactly how your embryos are assessed, by people and by tools, the team will walk you through it plainly.

Weighing up a clinic, or just trying to make sense of the jargon? A conversation with a specialist is the quickest way to see how any of this applies to your embryos and your treatment.

8. FAQs

How does AI embryo selection in IVF work? The AI is shown images of your embryos, either a single photo or a continuous time-lapse recording from a special incubator, and a deep-learning model reads their shape and development. Trained on thousands of past embryos with known outcomes, it gives each one a score ranking its likely potential to implant. The embryologist then uses that ranking to help choose which to transfer.

Does AI or the embryologist choose the embryo? The embryologist. AI produces a score that ranks the embryos, but it doesn’t make the decision; it’s a consistent second opinion, and the human weighs it alongside everything else before choosing.

What is iDAScore, or an embryo AI score? It’s a number a model assigns to each embryo to rank its potential. iDAScore, for example, runs from 1.0 to 9.9, with a higher number meaning a higher estimated chance of implanting and leading to a pregnancy.

Does AI embryo selection use time-lapse or a single photo? Both approaches exist. Some tools work from one still image of the blastocyst; others analyse a time-lapse recording made over five or six days inside a special incubator. The time-lapse versions have more information to work with.

Can AI tell if an embryo is genetically normal? Not for certain. AI scores tend to correlate with an embryo being chromosomally normal, but that’s a tendency, not a diagnosis, and the accuracy is modest. To actually confirm genetic status, you need PGT-A genetic testing, where cells are biopsied and analysed.