AI in Fertility and IVF Embryo Selection

In IVF, eggs are fertilised in the laboratory and the embryos grow for about five days to the blastocyst stage. When several embryos are available, the clinic chooses which to transfer first. Traditionally embryologists do this by eye, grading features such as how expanded the blastocyst is and the quality of its inner cell mass and outer layer, the trophectoderm, often with the Gardner grading system. Time-lapse incubators, such as Vitrolife's EmbryoScope, photograph each embryo every few minutes without removing it from the incubator. This produces a video of development, including when each cell division happens. AI models trained on these videos, and on whether transferred embryos went on to implant, generate a ranking score. Examples include Vitrolife's iDAScore, Fairtility's CHLOE, and Presagen's Life Whisperer, which works from single still images. The evidence is more modest than the marketing. A large randomised trial published in Nature Medicine in 2024 compared iDAScore with standard morphology grading and did not show the AI was non-inferior on clinical pregnancy rates, although it greatly reduced the time embryologists spent assessing embryos. Cochrane reviews of time-lapse imaging have found the evidence uncertain on whether it improves live-birth rates. A crucial point is often missed: ranking cannot create a better embryo. It only changes the order in which a patient's existing embryos are used. For patients who will eventually transfer all their embryos, the main possible gain is getting pregnant sooner, not a higher overall chance. Ethical questions include charging for add-ons with weak evidence, opaque scores patients cannot question, models trained on data from particular clinics or populations, and a slide toward selecting embryos for traits. Polygenic embryo screening, which claims to rank embryos by genetic risk scores, has been widely criticised by genetics societies as unproven.