DISCLAIMER: Today’s Advertiser helped make the production and delivery of this episode possible, for free, to you! But the themes expressed by the guests do not necessarily reflect the views of Inside Reproductive Health, nor of the Advertiser. The Advertiser does not have editorial control over the content of this episode, and the guest’s appearance is not an endorsement of the Advertiser.
What do embryologists do when they own stake in the company?
If you’re Professor Alison Campbell, Chief Scientific Officer and equity owner at Care Fertility, you’d build massive datasets to train a machine learning system to predict live births.
With Alison we dive into:
Their proprietary Caremaps-AI system (Saving 10 weeks of Embryologist time per year)
Why CARE is building a machine learning system rather than using AI software already on the market
How she pitted 10 of her best embryologists against an AI software she was skeptical of (And who won!)
The one AI solution she likes for egg freezing (Why CARE Fertility uses that rather than building their own)
Prof. Alison Campbell
LinkedIn
Transcript
[00:00:00] Prof. Alison Campbell: And it's saving loads of time. So now all of that manual annotation is over. I mean, we just get the machine learning model, press a button and in one to two seconds, it's generated all of that data that's previously taken us half an hour or so for a whole embryos course of development from fertilization to, to embryo cryopreservation.
And then it feeds into the same BLAST6, we call it the six, the six model, the statistical model, and we get a score. And that score relates to the chance of a live birth for that particular embryo. We obviously choose the embryo with the highest score.
[00:00:38] Sponsor: This episode was brought to you by Future Fertility, the leaders in AI powered oocyte quality assessment.
Discover the power of Violet oocyte assessments by Future Fertility. These AI based reports provide personalized egg quality insights to improve treatment planning and counseling for egg freezing patients. Deliver a superior patient experience and improve satisfaction by empowering your patients with an objective, personalized view of their unique chance of success.
Download a sample Violet report plus a roundup of clinical validation research today to learn the difference this tool makes in patient care. Visit futurefertility. com slash irh. That's futurefertility.com/irh.
Announcer: Today's advertiser helped make the production and delivery of this episode possible for free to you. But the themes expressed by the guests do not necessarily reflect the views of Inside Reproductive Health, nor of the advertiser. The advertiser does not have editorial control over the content of this episode, and the guest's appearance is not an endorsement of the advertiser.
[00:02:00] Griffin Jones: What do embryologists do when they own stake in the company? They do cool stuff like amass massive datasets with huge sample sizes and lots of detail to eventually be able to build a machine learning system that predicts live births. You've met Professor Alison Campbell before. She's been on the program before.
She's the chief scientific officer and an equity owner in Care Fertility, the largest fertility network in the UK and Ireland, now with clinics outside those areas. And Alison talks about Care Maps, their system that they started in 2011. When they started with time lapse imaging that is now in its sixth iteration and powered by artificial intelligence to save 10 weeks of embryologist time per annum to improve success rates to be able to predict live birth and why care fertility decided to put in that work to assemble that massive amount of data, make it safe, put it in one place to find a machine learning partner to do that with all on their own instead of choosing one of the AI systems on the market.
Talks about the value of their data set versus that of cares, the expense of those solutions versus what they were able to do. She talks about an AI solution that she does like called Violet, which is made by a company called Future Fertility and what value she saw Violet brought for their egg freezing patients and why they decided to use that solution instead of make their own.
If you're one of these companies selling into fertility networks, you might pay attention to what made Alison and her team agreeable to even do a pilot with Future Fertility. Why would you take the time of 10 of your best embryologists to see how they stack up against a solution that you're skeptical about in the first place?
Better make it easy for them to do that. I asked Alison if CareFertility will begin to sell CareMaps as a solution to other fertility clinics, IVF labs, and networks throughout the world. I asked her if some of the AI companies listening should just try to build that with them instead of what they might be starting to work on now.
Here's what she has to say about that. And once more, enjoy our interview with Professor Alison Campbell. Professor Campbell, Alison, welcome back to the Inside Reproductive Health podcast.
[00:04:03] Prof. Alison Campbell: Thank you very much, Griffin. It's lovely to be here again.
[00:04:06] Griffin Jones: It was probably a year and a half ago that I had you on Somewhere Thereabout.
And your episode was popular because you talked about embryologists owning equity and how Important. That is, I have a feeling some of the thread of that topic might reappear in our conversation today, but I wanted to talk to you about a tool that your practice has been using to great embryos and perhaps for other applications using artificial intelligence and perhaps other technologies it's called care maps.
You're the group that you. Work with and for and own part of his care fertility what I didn't realize though Is that it's been around for a little while? so I and I went with a default assumption that maybe it's a couple years old and he started as Artificial intelligence got more into the Zeke Geist in that 2018 2019 Timespan, but I looked and you had one YouTube video from almost 11 years ago now, and so that means you've been using it for at least that long.
So please lay the foundation of what CareMaps is and when it started.
[00:05:19] Prof. Alison Campbell: Right. Yeah, no, it's a, it's a beautiful story, I think. So it all began around 2011 when the Embryoscope first came to market. And it actually blew my mind, this device that enabled embryologists to to watch the embryo developing in, in real time, really.
So we set upon a great mission to introduce it into our clinics and to collect data from it, to use it to the max. So, and it wasn't easy. It was a very hard sell because of course there wasn't a lot of data around then. So. I had to try and sell the vision to our chief financial officer. We need to buy this kit, and this is why.
And it, it didn't go down well. It, it, they all thought it was just a toy. It was just a nice to have, but it just made complete sense that if we could get more information, from the developing embryo and the time points that it was reaching each of these subsequent cell stages, that there must be some answers within that information that could help us improve outcomes.
So we, we managed to get one free of charge for a fixed period whilst we did a really rushed evaluation just to make sure that it did do what we expected to do in terms of and functionality and imaging. And then we invested in the first one. And we wanted to get some data fast and we wanted to see if we could predict anything fast.
We're quite competitive and we generally want to be first movers in the field at Care Fertility. So we, we decided to annotate really strictly. So that means once we're, whilst we're looking at these time-lapse videos, the embryologist, using the, the viewer using the software that comes with it. Was recording in great detail everything that they saw so every time The cells divided, we introduced user defined variables very early on as well.
So things that didn't come with the software, we thought, well, that might be interesting. Let's also record as a comment, how the polar bodies, the second polar bodies extruded and little details that didn't come as standard. So we had a strict protocol training program and all of the PGT embryos. that were going through our clinic.
It was started in one clinic at this time. We recorded the ploidy when we got the result back. So then we had about a hundred embryos, and this was our very first publication, and we could predict, or we could classify, the risk of each of those embryos. being aneuploid based on the morphokinetic variables.
So quite simply the embryos that were somewhat delayed had a much higher risk of being aneuploid and we published that and it was the frontispiece on reproductive Biomedicine Online, it was really a well received bit of research and that was the first model that we, we developed. That was CareMap. So MAPS standing for Morphokinetic Algorithms to Predict Success.
[00:08:35] Griffin Jones: These algorithms, did they, were they produced by you also? The hardware is the time lapse imaging. And the software you said that you were doing some things like entering for user entered variables. Is that you all building your own software? Did the hardware come with a software? How, how did those algorithms develop?
[00:08:58] Prof. Alison Campbell: Well, the hardware came with software and a viewer and you could enter your own models. within that software. So the first one was quite simplistic. Now we're on our sixth iteration of ChemApps, which is a logistic regression algorithm. So we've had to work with the device manufacturer, Vitralife, to enable us to implement our own algorithm.
So they've been supportive with that. And then more recently we've introduced the AI element. So that sits outside of the equipment, outside of the device. So we've had six versions of ChemApps getting increasingly complex, getting more and more exciting in terms of what they're predicting. So we started initially predicting the aneuploidy.
And then we went on to clinical pregnancy and now our current models are predicting live birth.
[00:09:53] Griffin Jones: In that video that you had from 11 years ago, it says that it talks about AI was, were you using machine learning at that time or was AI more of a general blanket term compared to what it means today?
[00:10:09] Prof. Alison Campbell: Well, I, I didn't really know that AI was on the, on our agenda 11 years ago.
I've not seen that video for a while, but that was probably just, just looking to the future and imagining. So we've only been introduced, introduced AI to our care maps in the last couple of years. So that's yeah. And what the element that's. Being with, with it's involved with machine learning is the annotation.
So that's now completely automated. So I can talk much more about that and how it's won multiple awards. It's a quite a great phenomenon. We're very proud of, of the recent innovations that we've done with AI.
[00:10:45] Griffin Jones: I want to ask about that. So in the beginning it was, it was all manual. So you're, you know, you're entering these criteria, but at the end of the day, one individual is grading.
Each embryo just looking at it and then how do you compare so is the it was there anything like side by side like criteria that was just entered in there so that when you're viewing the embryo you're seeing it against your criteria or um, Um, You're just seeing the embryo and then you have to take it somewhere else to evaluate or take that image and information somewhere else to evaluate it.
[00:11:22] Prof. Alison Campbell: Well, initially we do the manual annotation, which is really laborious and we have been doing that for a decade and absolutely no regrets because that's the high quality data, quality assured, manual annotation that we've used to train the machine learning models. So what we do is we sit at the. machine which looked at the device and we'd annotate every stage and then the software that came with the device would calculate the scores based on the model that we'd entered into the device.
So it's, it's, It's only really the annotation element that was really laborious and took a lot of embryology time. The actual application of the simplistic statistical model was relatively easy.
[00:12:09] Griffin Jones: And so forgive my ignorance, explain to me what annotation refers to it. Is it the, is it's the, the grading of the embryo?
It's the characteristics of the embryo. It's other notes. It's those, that criteria that you set for the, uh, Self-centered variables. What does annotation refer to?
[00:12:26] Prof. Alison Campbell: So every five or 10 minutes or so, the time-lapse device is taking an image of the embryo through multiple focal planes. And so this goes on continuously.
Right after ixe, you put the embryos or the virtually inseminated cytes into their time-lapse device, and it's collecting these images. So annotation is when the embryologist sits at the screen of viewer. And reviews all of these images. Like a, it's a time lapse movie. I'm using the software that comes with the same device to click to say now it's two cells.
Now it's three cells. Now I've seen the beginnings of compaction and so on. So it's That annotation is the translation by the embryologist of this image information into, into data really. So it's very important that the embryologists are highly skilled and trained at this. And this is one thing that Carefidelity did really well, I think.
We insisted that we trained people very thoroughly, that we quality assured. their annotations. We didn't say, okay, the most junior member of staff can do all of the annotations. And there are arguments for and against, but I think the fact that we, we did it and we stuck with it for a decade, ensuring that everybody was trained and everyone was performing.
Properly as given us this goldmine of data now that's, that's pretty, really valuable.
[00:13:56] Griffin Jones: I was going to ask about who is doing the annotation. So a junior embryologist could do some annotations, but then what would have a senior embryologist would be doing other annotations or the lab director would have to have the final grade.
Tell me about the delineation of those responsibilities.
[00:14:14] Prof. Alison Campbell: Well, there was a competency, there was a training program and competency assessment to make sure that whoever it was, it didn't really matter what level they were, it's are they capable, are they competent at looking at these videos. I could train you to do it, I'd say.
It's not, you don't have to be a scientist to do it, you have to be well trained and you have to be meticulous and you have to believe and understand. Why are you doing it? And I think that's so important because if you, you understand the end game, you will do it properly. And then we'll do spot checks to make sure that we agree with those annotations.
If there's something really ambiguous, which happens with the human embryo, sometimes they, they'll go backwards. You'll see four cells and then two frames later, they've reverted to three cells. Anything peculiar, we would call a colleague over and we can say, look, can you sense this? This sends check this for me, and then we'd have a quality assurance scheme that we established ourselves whereby all the embryologists, all the annotators across the network would annotate the same set of embryos, and then we'd look at the intercorrelation coefficients to make sure that they are correct.
close enough, they're not always going to be identical. Sometimes they're a frame early or late. So then we'd look, well, if you are a frame early or late for that particular variable, so for the start of blastulation, let's say, for example, then does that impact the score, the model score, and therefore does that impact the selection, which embryo you would choose based on the model score.
So it was, it was a very thorough and very complex process, but it's, and it's taken a decade to get to where we are.
[00:16:00] Griffin Jones: Is it now a requirement for every embryologist in your organization, this assessment for competence and annotation?
[00:16:07] Prof. Alison Campbell: It's always been a requirement at Carefertility, yeah, every embryologist who's annotating needs to be competent to do that, the same way that
[00:16:15] Griffin Jones: Does every embryologist annotate or is the workflow segmented in such a way that some embryologists are annotating while others, you know, might be freezing, thawing, etc.?
So does every embryologist annotate?
[00:16:29] Prof. Alison Campbell: Yeah, everyone who's been trained and is competent can annotate. So if they're on the rota for a particular day to do those, the annotations, then they would do the annotations if they're supposed to do vitrification or end collection. So it's, it's done on a rota type basis.
[00:16:44] Griffin Jones: So now you're on the sixth iteration, machine learning has been introduced to, to now be able to do that annotation. Did that, is that new to the sixth iteration? Did that happen? What iteration did that happen?
[00:16:56] Prof. Alison Campbell: Yeah, it's new to the sixth. So the sixth iteration is the live birth prediction model. It's the most sophisticated model that we've got we've ever had.
It's built on over 6, 000 transferred blastocysts where we know the live birth outcome. And so we realized we are taking so much time to manually annotate all these videos across the network. So. 15 laboratories in the UK, some in Spain and the US that aren't yet fully set up to do this, but at least at the time, what can we do to really save time to improve reproducibility and objectivity because manual annotation is not perfect.
So let's, let's look at this. Let's get the data together, which was no mean feat. Let's find a third party who have experience in machine learning and See what they can do. So we scoped the project. We did the business case and we found a UK company or their international called BJSS and they had no experience in the fertility sector, but they did have experience with.
Machine learning. And I was quite impressed because they'd done some work with, with airports, where they built models to scan suitcases, to identify smuggled animal skulls and things. So I thought, well, it's image analysis. It's very important. And yeah, they were very impressive. So we worked with them very closely for, took about 18 months, I'd say, from them to the release of the minimum viable products.
And it's. It's saving loads of time, so now all of that manual annotation is over. I mean, we just get the machine learning model, press a button, and in one to two seconds it's generated all of that data that's previously taken us half an hour or so. for a whole embryo's course of development from fertilization to to embryo cryopreservation.
And then it feeds into the same BLAST6, we call it the six model, the statistical model, and we get a score. And that score relates to the chance of a live birth for that particular embryo. We obviously choose the embryo with the highest score.
[00:19:24] Sponsor: Egg freezing programs are experiencing growth worldwide. Yet patients continue to complete their treatment with little to no understanding of their unique chance of success.
To date, there has been no standard of care for assessing egg quality. At best, general age based estimates are provided, which assume that all patients of the same age have the same quality. There is so much untapped potential to better manage your patient's expectations for success while ensuring treatment plans are tailored for each patient's unique oocyte quality.
Future Fertility has addressed this gap in care, enabling your clinic to drive a superior patient experience while optimizing cryopreservation outcomes. Future Fertility are the leaders in clinically validated, AI powered oocyte assessment. Violet, developed specifically for oocyte cryopreservation patients, integrates seamlessly with any lab to deliver objective and personalized oocyte quality insights.
Violet leverages the industry's largest oocyte dataset to provide predictions of each egg's blastocyst development that visual assessments simply can't match. The report also provides a personalized prediction of your patient's chance of live birth, which factors in their unique egg quality, helping you ensure your patients have sufficient oocyte stored to give them the best chances to meet their fertility goals.
Let Violet become your partner in fertility planning, transforming the way your clinic approaches treatment and counseling for egg freezing patients. Visit their website today at futurefertility.com/irh to receive your free sample report plus a roundup of their industry leading clinical validation research.
That's futurefertility.com/irh.
[00:21:31] Griffin Jones: This is all embryo grading. There's no oocyte grading happening with CareMaps?
[00:21:38] Prof. Alison Campbell: No, not with CareMaps. We are interested, um, we've been talking to Uta Fertility, who I know you, you've been speaking to recently. We talked to them about. Trying to do some research together just to see whether Violet, their, or Magenta, their oSight AI tool could add some benefit to care maps.
I'm, I'm a little bit skeptical that it could, but if we could get some marginal gains to improve predictive power, even subtly, then it's worth exploring.
[00:22:08] Griffin Jones: So would that be for, like, you would be using their oocyte grading in to improve your embryo grading or you would be using it separately to create something for, for oocyte, for oocyte grading?
[00:22:21] Prof. Alison Campbell: Well, they've already got their own tool to assess the oocytes, which we use for fertility preservation purposes. So if we work, if we thought about putting it together with ChemApps, it would be to see whether The assessment of the oocytes from their system could improve the predictive power for embryo selection in our live birth prediction models.
[00:22:48] Griffin Jones: So I want to talk about this, about building an AI solution versus going with one and, uh, future fertility might sponsor this episode. Somebody else might, Burger King theoretically could, yeah, they don't have any control over what you say. So you can, you can say whatever the heck you want, but I, I, I see some people really see the value in certain AI solutions and then other times.
Yeah, I've, you know, I think I heard Santee, Dr. Mune say, you know, try to build one yourself for a lower cost. The costs also range a lot. Some of them seem really expensive. Some of them don't see, seem as expensive, but it sounds like for in the case of, Embryos, you wanted to build your own as opposed to using one of the solutions that are out there with these, these companies trying to get them implemented into clinics.
Why did you decide to go the build your own route?
[00:23:45] Prof. Alison Campbell: Well, a few reasons for that. We, we tried the systems on the market at the time, a couple of years ago. With AI, we're talking about the AI aspect and it wasn't good enough for us. It didn't It it wasn't comprehensive enough. We can use a an alternative providers Auto annotation tool, but we weren't getting comprehensive auto annotate.
We weren't getting all the data that we needed to feed our models. There were lots of emissions, lots of inaccuracies, lots of, um, sense checking required. So probably a couple of reasons for that, but the main one may have been that it's trained, they're trained on really heterogeneous data from all over the place.
And I have most confidence in our own data. Our own data was massive. Relatively massive. Arguably, probably the highest quality data set for annotated embryos and largest in the world because of the approach that we took. So why would we use somebody else's tool when our data is stronger, bigger, better?
That was at least the mindset at the time. That might be different now, but we've taken the leap and we're not looking back.
[00:25:05] Griffin Jones: Was that the defining feature of what made those systems on the market not good enough in your view, the, the quality of the dataset? Were there any other reasons?
[00:25:16] Prof. Alison Campbell: They were expensive.
I think when we first looked at it, and it may, may still be the same, that it doesn't seem to be a really Clear offering in terms of what you get, the pricing model, is it per click per price per click, or is it a flat fee? I think people are just trying to find their way. They don't have lots of publications to justify the quality and the predictive power of their algorithms or their systems.
So it's still relatively new and you have to make a leap of faith if you're going to do it. And so we decided we'll, we'll do that with our own data and take it from there. But I think we've done a good job. We've won. Multiple awards for this solution. We, Amazon web services did a case study on, on it.
We've won a Royal College of Pathologists achievement award, UK IT tech awards. We've just been nominated for another one I heard this morning. So I think, you know, when we're scrutinized by the data scientists and machine learning experts and tech people, they can see that this has been done really well.
[00:26:28] Griffin Jones: Tech bros are scrutinizing you? Yeah. In marketing, we would call them internet Ian's, but that doesn't mean that they don't have a certain expertise. What do they feel that you're, so you've In your view, the established solutions didn't have a high enough quality data set. What do the machine learning geeks critique your solution for?
[00:26:53] Prof. Alison Campbell: Well, they can just see the information that we give them and we're not letting them under the bonnet to just scrutinize it. So they can scrutinize the way we did the build or the way BGSS, our partner did that build, the size of the data. The, the outputs, the predictive power, because we've used this model now respectively for over 2000 transfers.
with the AI element. It's predicting really well, really, really successfully, accurately. And, and the time saving we've quantified, we estimated it beforehand. And now we've quantified it. We've say we're saving 10 working weeks embryology time per year. So that's, that's a great output as well. So there's so many benefits that we've seen from this.
So yeah, patient attraction has been also. A good one, staff retention. We've had embryologists saying, I could not work anywhere else. I could never go back to manual allocation. I could not live without this system and you know, non professional recognition for our work. We've presented it at conferences and we're still writing up.
We've got a lot of information still to share and to publish, but it's, yeah, it's, it's on track.
[00:28:08] Griffin Jones: You developed this system, you all at CareFertility developed this system because you weren't satisfied with what was on the market and you felt you got a better data set, a bigger data set, better predictive power.
Now that you have something that you feel is better than what was on the market, are you going to take it to market? Are you, should we expect CareFertility to spin off CareMaps and be selling that to the EVRMAs and the EUGENs and the. the, the inceptions, et cetera, of the world?
[00:28:43] Prof. Alison Campbell: Well, never say never. It's a possibility.
I would say we've got to consider our priorities. And we've, we've got the data. We've got a lot of expertise in time lapse in this type, this area of machine learning, but we don't have a sales force. We've never done regulatory, got regulatory approvals for our, our products. So this is an in house, developed tool.
So we could use it in our clinics, but we couldn't sell it as it stands without certifications, FDA approval, CE marks, and all of those things. So almost certainly if we did that, we would be looking for a partner to help us get there.
[00:29:25] Griffin Jones: I wonder if any of the current partners who are thinking, man, this is tough.
This is really hard to sell in, into these clinics. Why don't we just do that? Why don't we just try to, to, to take what HairMaps is doing and then make that our product. And we've got the Salesforce and we've got the venture raising infrastructure. And I think that might happen.
[00:29:47] Prof. Alison Campbell: That could happen. Yep. I think it could happen.
And yeah, open to conversations. And let's say we, we don't have a Salesforce, but actually. You can probably tell, you know, I'm so passionate about it, it's, I probably, some of our team members would be the best people to sell it if it's, uh, if we were to take it to market because we, we've lived and breathed it for 10 years and we trust it.
[00:30:09] Griffin Jones: I'm sure.
[00:30:10] Prof. Alison Campbell: And so, it.
[00:30:13] Griffin Jones: And so maybe future fertility can help with the, with the embryo grading, but, but you're, you're a bit skeptical of that, but they are helping you with oocyte grading for egg freezing. Why go with them in that situation, as opposed to then trying to develop your own oocyte solution. So the embryo grading systems on the market weren't sufficient.
And, but, and so you built your own. didn't go that route for egg freezing. Why not?
[00:30:46] Prof. Alison Campbell: Well, yeah, we considered it, of course, and, and it may still happen that we, we do our own thing, but we focused on one, one thing at a time and we focused on the embryo selection. We, Caught Future Fertility's Violet through its paces early on, because I was really sceptical.
A static image of an egg to predict outcomes, but saying that, we, we assessed it, got 10 of our expert embryologists to compete against Violet, and it beat us. So yeah, we, it could, it could assess and predict better than we could. Not at a very high rate, because there are so many other variables. Sperms takes a huge part to play in it and lots of other factors, but at least for patients who, who want a bit more information and they say, well, how are my eggs?
Without Violet, we'd give them our best judgment, but it wasn't particularly accurate. And with Violet, they get more information, they get images of their eggs. And so it was a nice to have. So, yeah, I, and generally I'm skeptical of static image assessment because human, human embryo development is a very dynamic process and yeah, there's so many things that can impact it.
So yeah, we just got to focus on what we have the most faith in, I think at any one time and put our efforts into that.
[00:32:12] Griffin Jones: You were skeptical of the static images, and then you put it against your team of 10 embryologists, and it won, but to do that pilot test, you must, they must have communicated some sort of value to you, or if they didn't, you You just perceived the need, you know, it's tight grading to be that great.
I remember last time we spoke, I asked you how many of these different companies pitch you over the course of the year. And I think it is probably a couple dozen that you said. And then I asked how many in a year do you do any, even like a pilot? Program with, and you said, you know, maybe three, I think it's something along those lines.
So you were talking about one out of every 10 or, or, or something like that, that you're actually piloting. What was it about that pilot that you said, this is worth the time of 10 of my embryologists to, to put them against.
[00:33:09] Prof. Alison Campbell: Well, it was a relatively easy thing to do. It was a quick pilot. It was on a, an app.
And so we could do it quite quickly, gather information fast. There were nice people. There were also passionate. Dan Neo, particularly when he first knocked on the door, very passionate about his product, made sense for them to have looked into that element. And I wondered why so few people had ever really tried to come up with a tool to.
assess the quality of an oocyte. But yeah, they were there right at the start and simple and effective. It's not going to change the world. It's not highly predictive, but it's better than we can do. So I think that's a positive for patients.
[00:33:53] Griffin Jones: Do you think that's the future of AI companies in the fertility space, like more segmentation?
Or do you think so that they can find a place where they really can be valuable one and then to make easier pilots? Or do you think that somebody has to win this battle to become the AI solution for, for all, you know, embryo and, you know, site machine learning? I should say all embryo and gamete machine learning.
[00:34:25] Prof. Alison Campbell: Yeah, I think it'll, um, it'll be quite a slow journey. There's a lot of competition, a lot of people trying to get a piece of the pie at the moment. Um, but eventually, I think there'll be just a couple of High quality solutions, which incorporates gametes and embryo assessment. I don't think it's really going to be, we're going to see hundreds of different options.
I think eventually the cream will rise to the top. There'll be collaborations, partnerships, merging of solutions. Cause what we want in the clinic is, is simplicity. We don't want to be moving between systems and causing confusion. We need integration of, of good systems and simple, simple tools.
[00:35:10] Griffin Jones: Are other networks doing this to your knowledge, developing their own embryo grading, machine learning?
[00:35:18] Prof. Alison Campbell: Not to my knowledge. I don't think at the scale that CareFertility have been moving and developing this, this CareMaps AI, I don't think I, I haven't seen that or heard of that. Now I've spoken to some big group scientific leaders who've said it's so difficult to get the together. So I think, I know it was a huge undertaking for us to get all of that data off all of those servers into one safe place.
And so we were fortunate to have the expertise or have the partnerships to enable us to do that. So that was the first step. And I think that some of the big groups might, will be struggling with that. And also if they didn't embark on the journey like Fertility did, annotating rigorously and religiously, comprehensively.
Then they won't have that data set, but you can accumulate it very quickly. Some of the networks now are enormous, and if they just decided to change tack and do exactly what we've done, it wouldn't take 10 years to do it.
[00:36:25] Griffin Jones: It's difficult to get the data together. You talked about it was possible for you because you had partners, but they could go out and get the data.
Adequate partners to help them with that. What made it possible for you all to bring that data together?
[00:36:42] Prof. Alison Campbell: Well, teamwork and shared vision, I'd say it's, uh, but it's
[00:36:47] Griffin Jones: gotta be something in the shared vision because if they wanted to, they could, they could align the teamwork to it. So there was something about your shared vision that prioritized it in a way that maybe others haven't.
What do you, what was that? Why was this a priority?
[00:37:01] Prof. Alison Campbell: Well, because we'd already got Care Maps, so we already, we were getting great outcomes for our patients, we were generating revenue, we, we loved the technology, we were getting publications from it, so it was part of our DNA, so it was the next step, really, to bring machine learning into that, to save time, we, we were never going to say goodbye to, to Care Maps, we wanted to keep developing it, we'd done that six times over over the last six years, 10 years.
So it seemed to be the next step. And it's quite possible now, you asked the question, that These guys tapping on our shoulders saying, well, look at our solution. Look at our solution that they catalyzed our actions because we we'd been talking about it. But once we realized that other people are starting to bring tools to the market that can automatically annotate and predict outcomes, we should, we should be.
Lead in the way. We should be doing that. Let's, let's get on with it because time moves quickly and We have the potential to to make all those benefits that I described and particularly saving time for our embryologists It was a huge driver.
[00:38:11] Griffin Jones: Again, my ignorance are other labs not annotating to this degree of detail?
[00:38:16] Prof. Alison Campbell: I don't believe so. No, they're not. I think we were annotating every single embryo for a very long time and And Some of the clinics will have not annotated at all, or else they'll just annotate the blastocysts, or just the euploids. So, if you do that, you, you're restricting your dataset because you're only annotating the good quality ones.
You've not annotated the ones that have arrested the patients. On the third day at five cells, for example, or degenerated at the more realist stage, we annotated everything. So the data that trained us, we've used the manual annotation data that we've trained the machine learning models is, is really comprehensive.
[00:39:04] Griffin Jones: Did that set out, was the vision for that originally to eventually compile a massive data set? Was that the, the main or only driving reason, or were there other reasons for that level of detail in your annotation?
[00:39:22] Prof. Alison Campbell: Yeah, no, that was the main reason is because we want to the data and in the data will be the answers.
So unless we collect the data really thoroughly and comprehensively, we're not going to get all the answers that we want to find.
[00:39:36] Griffin Jones: And so now you're at a point where you're predicting live birth rates. Tell me more about that.
[00:39:42] Prof. Alison Campbell: Well, when I'm describing it to patients, I'll say, well, we've Transferred embryos in good faith.
We transferred blastocysts in good faith and we've put the ones that have resulted in a baby in one bucket and the data from the ones that haven't, they've been transferred in another. And we've analyzed to see the differences in their morphogenetic values, in their developmental timings and morphological scores to see what the differences are.
And then we've built these predictive models. So we, when we do apply our models, to predict live birth, we get a score and the score one will mean that embryo, it's made of blastocysts because these models are applied to all of the blastocysts, the live birth chance is about five percent, so really low chances.
of live birth, even though we, we can see a blastocyst, which is sometimes seemingly beautiful. And then it just goes up to a score of 10 and the chance of live birth is over 50 percent with that embryo. So of course we choose the highest score and we've used the So we've retrospectively validated these models and now we've prospectively validated the models and they work exactly as predicted.
So we achieve the birth rates or the clinical pregnancy rates just as we predicted because it's, it's so accurate. And when you look at morphology alone, which is the alternative, really a standard practice, you have trophectoderm quality and inner cell mass quality, and you have a stage of expansion. And those variables are nowhere near as predictive of life birth.
They, it's just over flipping a coin. It's not, it's not good enough. And it upset, it upset me a lot over the years working with standard practice that one embryologist would choose one blastocyst from a cohort and another embryologist will choose another. Now with CareMaps, we will choose the same one and we'll choose the best one.
[00:41:45] Griffin Jones: With this level of detail and all of the data that you've assembled, would that even be technically possible without time lapse imaging?
[00:41:56] Prof. Alison Campbell: No, it wouldn't be possible.
[00:41:58] Griffin Jones: And so it was 2011 where you first started using time lapse imaging. I would say in the U. S. Probably fewer than 20 percent of clinics are using time lapse imaging in their labs right now.
Maybe it's around there. It sounded like you had to make that case in the beginning, but I want to ask what, what percentage of embryos are PGTA tested in the UK about?
[00:42:31] Prof. Alison Campbell: It's much lower than the U S I believe it's probably closer to 20%. Whereas in the U S it's It's, it's more than 50%.
[00:42:38] Griffin Jones: Yeah. I think it might be 60, something like that.
And so are, are these two things either or in many people's view that we either do time lapse or we do PGTA? Is there a reason why it's not both?
[00:42:53] Prof. Alison Campbell: Well, yeah, I'd say in most people's view, they think right. And embryo selection, is it PGT or is it time lapse? If we're talking about modern or more sophisticated.
embryo selection, but actually there are synergies. between them. And we've shown from our own data that you're, you've got a patient with multiple euploid embryos, then you can apply care maps to distinguish between those euploid embryos. And of course we want the best embryo transferred. So if we've got that technology, then, then we should be using it.
[00:43:27] Griffin Jones: So you had to make the case though for the, for the time lapse imaging back some 12, 13 years ago. And This might tie back into the first conversation we have about embryologists owning equity in the clinic and the network, because I think you said something to the effect of that. They thought it was just kind of a nice toy and you had to convince them of a greater business value.
And I think When I just kind of ask around, I've started every embryologist that comes on the show. I asked them, do you think time lapse imaging is a nice to have or a must have? And it seems like everyone is saying they think it's a must have. And yet we have so many networks that don't have time lapse imaging.
So you had to convince them of that. Of that value. And you also had to have seen the value yourself because you own equity in the company. What was that business case that you had to make to your colleagues and, and to your self and that you feel maybe isn't being made strongly enough?
[00:44:35] Prof. Alison Campbell: Well, you know, it's a long time ago, but the business case related to it being a differentiator and having the potential to improve outcomes.
So it wasn't um, a rock solid because we didn't know if it would definitely improve outcomes. We didn't know how much, how much it was going to cost us as a whole network if we were going to end up rolling it out across all of the clinics and we hadn't really been certain about what we would charge patients for it and if that was appropriate or not.
So there were lots of discussions. We did invest in it because the hardware is very expensive. We did start to charge patients. Once we'd got confidence in it, we had, we didn't charge for six months while we collected the initial data and we built our preliminary models. So once we had demonstrated that it was going to help us with outcomes and it could predict, predict Ploidy at that stage.
Or risk classify and the patient feedback was really positive. And one of the questions that we asked the patients was how, how, what, what, uh, did you like about the time lapse? What did what, how could you relate to it? What did you feel about it? And the patient feedback was mostly we, it really aided our understanding as to what went on in the laboratory.
And we also asked them, do you think the price. Is appropriate. And the vast majority said yes. So that was really, that was really promising and that helped us invest in further devices and keep rolling it forward and then invest in the statistical analysis and then more recently in the, in the machine learning.
[00:46:20] Griffin Jones: So it was more of a longterm play though, if you're thinking about differentiation that way, because I think if, if you're not looking on a. Five, 10, 12 plus year horizon. Maybe it's more expensive. If you're, if you're looking on a three year horizon, then it's pretty big expense to have for all of those IVF labs, isn't it?
[00:46:43] Prof. Alison Campbell: It's a huge expense, but the return on investment. Didn't take much, it wasn't too long before it came, came back because if you're charging 500 pounds per cycle and the device was 60, 70, 000 pounds at that time and you're getting good uptake and because it's a patient choice, it's an add on. It, it wasn't too difficult to get the money back in order to then buy the next device and so it just kept rolling and so it's been a great success financially, success rate wise.
Staff wise, time savings, efficiencies, and R and D wise.
[00:47:22] Griffin Jones: It seems to me that in order to meaningfully improve success rates, and in order to have differentiation, people have to have the data. They have to have the data for everything. And so that refers to the tools in the lab that allow you to capture embryo and gamete data refers to, uh, software that allows you to capture clinical data and.
Other inputs and outputs. I don't think people will be able to differentiate without it. We talked about the market possibilities of CareMaps. Maybe somebody listening will say, Hey, why don't we throw in the towel on what we're doing and try to build CareMaps out into a, a side company that could sell into.
Other networks. We talked about the possibility of some of those companies merging as competition thins out and someone emerging as the ultimate AI solution for the IVF lab from a technical perspective, what's on the horizon for care maps. This is how I want to conclude our conversation today. What would we expect from iteration seven?
[00:48:33] Prof. Alison Campbell: Right. Iteration seven would ideally. be device agnostic. It will be cloud based. It will not be tied to one particular time lapse device. It would be accessible to this maybe version. This is the future. This is the dream. It would be nice if we could get it certified and enable other people outside our network to use it and to see, feel the benefits of it.
And for those patients of those. competitor clinics to also feel about the benefits of it. So we'll see.
[00:49:11] Griffin Jones: When do you work on the next iteration? Does it like, does that work immediately begin as once you've completed an iteration or is you, you work on an iteration for once an iteration is implemented, you wait a little while to see what the needs are.
How does that work?
[00:49:28] Prof. Alison Campbell: Yeah, we wait a little while. We usually Keep one version going for 18 months, two years before we see like how big is the data set now? Are there any, where, where, where's the area we could tweak and improve? So yeah, it's, we'll be coming up for those port processes soon, but version six is, is working phenomenally well.
[00:49:50] Griffin Jones: And how often do priorities shift in what you expect the next iteration is going to need versus what you end up actually doing? So in other words, going into iteration five, you implement. the fifth iteration and you're thinking this is probably what we're going to need for the sixth iteration. How much does your expectation, your assumption match up with what you end up actually needing for that next iteration?
[00:50:18] Prof. Alison Campbell: That's a really good question. We, I'm not sure so much thought goes into it as you might imagine. Really, it's because the data set is growing. We've got the ultimate, in my opinion, the ultimate outcome measure, which is live birth. So previously we were just reaching for the better outcome measure. So it took time for us to feel confident we've got enough data now to predict live birth because you have to wait, obviously, quite some months to get the live birth data after you've done the annotation, after you've collected the embryo data.
So we'd started with ploidy prediction and we'd moved to clinical pregnancy prediction. And then we got there with the live birth prediction because we had the data, we had the numbers. So I believe we've got it. Got that in terms of outcome measure. It is the best. And people do argue that it might not be this.
There are other variables. Pregnancies can be lost, needs to be a euploid. So we'll see that that mindset could change, but I don't think it will. So going forward, what we're going to be looking to do is to save more time, make the model, uh, better. faster and better and bigger and more accurate, but always I think looking for the, for the live birth is the outcome measure.
So it'll come now, the improvements will be more for user friend, more user friendliness. It's more time savings, and I think now with predictive power.
[00:51:46] Griffin Jones: I look forward to having you back in another year or two to see the progress that you're making with CareMaps, with the other technologies that you're paying attention to.
This has been yet another fun conversation that I think our audience, especially our lab audience, is going to like to hear. Quite a bit, but I think also the executive leadership is going to appreciate your take two with differentiation and new market opportunities. Professor Alison Campbell, chief scientific officer of care fertility.
Thank you very much for coming back on the program. Thank you very much.
[00:52:18] Sponsor: This episode was brought to you by future fertility, the leaders in AI powered Oocyte quality assessment. Discover the power of violet Oocyte assessments by future fertility. These AI based reports provide personalized egg quality insights to improve treatment planning and counseling for egg freezing patients.
Deliver a superior patient experience and improve satisfaction by empowering your patients with an objective, personalized view of their unique chance of success. Download a sample Violet report plus a roundup of clinical validation research today to learn the difference this tool makes in patient care.
Visit futurefertility.com/irh that's futurefertility.com/irh.
Announcer: Today's advertiser helped make the production and delivery of this episode possible for free to you. But the themes expressed by the guests do not necessarily reflect the views of Inside Reproductive Health nor of the advertiser. The advertiser does not have editorial control over the content of this episode and the guest's appearance is not an endorsement of the advertiser. Thank you for listening to Inside Reproductive Health.