Thanks to the St John’s College Student Opportunities Fund, Lilly Lees found herself in Edinburgh presenting research on AI generated embryo images.
“I can’t believe it’s not a real embryo! Can embryologists tell the difference between synthetic and real embryos?”. This was the title of my abstract that I had the privilege of presenting at the Fertility 2024 conference in Edinburgh, thanks to the support of the St. Johns academic opportunity fund.
Earlier that year, I conducted research under the guidance of Chloe He on synthetic embryo image generation using diffusion networks (AI-generated embryo images). I know, on the surface, it doesn’t sound all that important. Given I hadn’t acknowledged academic biology since my GCSEs, a year earlier I would have thought the same thing. However, once I learnt more about reproductive health and the increased reliance on embryology for conception, my view changed.
As I stood beside my poster at the conference, I found myself fielding questions from curious attendees, all wondering, “Why?” Why bother with AI-generated embryo images? Well, let me break it down for you. With AI becoming increasingly intertwined with embryology, the dataset used to train these systems is crucial—enter synthetic embryo images. By generating synthetic embryo images makes an infinite amount of fake images (so no more data scarcity), it allows anyone to create AI models that need embryo datasets (not just big fertility or academic institutions), it allows us to create more images of sorts of embryos that are rarely documented (yay, more dataset diversity!) and lastly, it circumvents some of the privacy issues of storing real biological data, as opposed to synthetic.
But why does any of this matter? Excellent question. We developed a diffusion model capable of generating synthetic embryo images and then put them to the test. When presented alongside real embryo images, a third of embryologists couldn’t distinguish between the two. That’s quite a lot when we consider our computational and time limitations for the project, and it certainly shows the feasibility!
Presenting at the conference was nerve-wracking, to say the least, but it was also incredibly rewarding. Engaging with fellow attendees—from senior embryologists to fellow undergraduates, in a whole range of fields—opened my eyes to the vast landscape of fertility tech and its implications. I connected with biology students intrigued by machine learning, explored conversations on personal identity and fertility with psychologists, and attended talks on areas of reproductive medicine I had never heard of (notably TESE, don’t look at Google images).
In the end, the conference was an invaluable experience. It wasn’t just about presenting my project; it was about forging connections, honing my public speaking skills, and gaining insight into the future of reproductive health. Plus, I got to explore the beautiful streets of Edinburgh—a bonus I couldn’t pass up. Thank you to the Johns Opportunity Fund for the opportunity and experience!
John's Chronicle 