Artificial Intelligence (AI) is transforming in vitro fertilisation (IVF), offering new ways to make embryo selection and sperm selection more efficient. This week, we speak with Dr Gavin Sacks, a leading Australian fertility specialist, about the integration of AI into IVF clinics.
We discuss an AI system that analyses thousands of embryo images, using historical data to predict which embryo will have the highest chance of leading to pregnancy. Whilst the AI tool has proven just as effective as human experts and ten times more efficient, it also raises ethical questions about the future of AI in fertility treatment. Join us as Dr Sacks explores the clinical benefits, challenges, and future implications of AI in reproductive medicine.
To start, could you share what clinical problem IVF Australia was trying to solve using the AI tool?
Dr Sacks: IVF is a fast-moving science with constant pressure to improve efficiency, success rates, and cost-effectiveness. One of the biggest challenges in IVF is embryo selection. When we do IVF, we collect and fertilise eggs, then we have embryos. And the embryos have a grading system, which is where we see how they look biologically. You can have a range of embryos and choosing which one is likely to work is very subjective, you know, the scientists on the bench grading embryos and deciding.
We’ve developed lots of systems that observe the embryo growing across a few days. But the game changer was when a medical student at UNSW proposed using AI to analyse embryos. The embryos grow in a culture system for five days and it’s possible to take a picture every 10 minutes, so it captures over 700 images of each embryo over five days. The AI is trained on historical data to identify patterns in which embryos worked and which didn’t. It’s impossible for a human to analyse so much data. Previously humans have looked at certain time points, saying “We’ll look at day one and at day three”, to identify the stages that embryos should go through, but it’s quite crude.
This technology has since been distributed worldwide so AI is quite common in IVF now. Every IVF lab would be doing it, nearly doing it, or thinking of doing it.
Can you describe more about the process? Does the AI select the embryos first or does it check the work of the embryologists?
Dr Sacks: No, the embryologists pick the embryos, and the AI is a backup. But it could depend on how good the lab is. In labs with more inexperienced scientists, the AI could take a more central role. The reason people want to use AI is first as a protective backup. Second is to save some lab time, in theory. Speaking as a clinician, I’d still like a person deciding. I’m not personally sold on AI taking over, but I think that it will free up our embryologists to do other things.
There is a related element about consent. Do the patients know whether AI is being used to choose out the embryo? And are there any opt out mechanisms if someone didn’t want AI involved?
Dr Sacks: Yes, the patients all know that AI is being used. And to be honest, that’s another driver that pushed us to do it. With the patients, there is a lot of competition and pressure to have high-end technology. I think they’d prefer it almost. Regarding opt-out mechanisms, there is no formal mechanism, but patients can request that AI not be used, though this is rare.
What other AI applications exist in IVF?
Dr Sacks: We’ve pioneered another kind of AI to look for sperm in low sperm count individuals. Some people have severe male infertility, and even when we do surgery to try and find sperm, have really low numbers. Embryologists can spend four to six hours in a day going through a sample looking for a sperm. That’s hard to believe right? It’s a really skilled person taking an enormous amount of time. In the last couple of years, IVFAustralia have developed this new AI tool that finds sperm quickly. It assists the embryologist to look get through a sample dramatically more quickly and with more accuracy. We’ve published and presented on this in the last couple of years.
There are other people looking at AI in the IVF industry. Rostering in the lab is another big thing. IVF is extremely complex, and we have lots of different kinds of people that have to come together for any particular treatment. It’s hard to know ahead of time which staff and how many staff are needed on a particular day. So, people are using AI to predict how treatmentcycles are coming to make that more efficient.
Regarding the operational aspects of IVF, do embryologists need special training to use the software and was there an onboarding process?
Dr Sacks: For the sperm search there is a little training, but it’s easy and straightforward. For the embryo selection, the process is virtually automated. The embryos are in an incubator, we have a computer that analyses the images, then the embryologists receive a score for each embryo, which they use to inform their decisions. The score is the AI tool’s output.
Were there any clinical governance practices implemented alongside the AI system to ensure its safety?
Dr Sacks: In terms of the embryo work, we did a big randomised controlled trial, published in Nature Medicine showing that AI is ten times faster and just as effective as human embryologists at achieving pregnancy outcomes. It’s not currently better, but that is still very useful for us. On the human side, people are very busy and it’s good to have an assistive tool for the scientists to decide. Sometimes it is a difficult call. The scientist doesn’t know “Should we pick this one or this one?”. And the AI will suggest a particular one.
Looking at the bigger picture, what do you think have been the greatest benefits or challenges from using AI in IVF practice?
Dr Sacks: In IVF, people are always looking for better success rates, but AI hasn’t yet shown better success rates overall, and I’m not sure that it will ever do that. But it will definitely save embryologists time, such as with the massive reduction in time for the sperm searching method. So, I think the real goal of AI in IVF is to save time and save the work required by embryologists, to make it cheaper, which hopefully flows down to the patients to have less cost.
But we’re heading towards a place in reproductive medicine where we’re learning more about embryos and people have started looking at selecting embryos based on characteristics, such as sex and risk of genetic disease or disease in later life. You know, how do you put that all in a choice? Whilst this could help parents have healthier children, it also opens the door to “designer babies”. I worry that we’re heading towards a future where machines decide the next generation of humans, because who knows what the coding is for those machines? That is a real reason for us to always be cautious, especially in reproductive medicine.
You’ve mentioned a lot of fears about AI and IVF in the future. Do you think there are enough safeguards in place to address these concerns?
Dr Sacks: There are government regulations on reproductive medicine. But each country has its own regulations. Some are stricter than others, so then people will go a country with more permissive laws to do it. Currently, there is a law in Australia against the testing of embryos like a designer baby. But it can be done in America and people in Australia are beginning to request it. Where does the law stand on this?
I don’t think it’s just a matter of policing and stopping AI, it’s a matter of us understanding as a society the kinds of directions we want to move in. I believe in fostering public understanding and debate so that people understand what’s really happening. There has to be responsibility on the companies that do it too. I agree that we need a regulatory framework, but for me that would be more like a guideline than a law, because where do you even draw the line?
You have to think of the realities – IVF itself was in that position. That’s the nature of human society, for people to be learning and pushing boundaries. So, it’s a tricky situation.