82 Advancing embryo evaluation: generative artificial intelligence to assess embryos in routine embryo transfer practice

Generative artificial intelligence (GAI) utilizing computer vision is a powerful tool that is marked as a dynamic and proficient means to rapidly process data and generate predictive output with high precision and accuracy. With these capabilities, integrating GAI into embryo transfer (ET) practice could allow advanced automated embryo analysis to accurately predict embryo viability. Such analysis could surpass the ability of human evaluators and thus break the barriers currently restricting the potential of ET to generate genetic advancement and limiting economic returns. To explore these capabilities, this study had two aims: (1) to use GAI to assess embryos and reproductive outcomes in a field trial, and (2) to survey embryologists’ assessment of bovine embryos with traditional methods and compare these assessments with GAI results. A subset of 30-s videos (n = 6900) of both in vivo-derived and in vitro-produced bovine embryos were recorded from 10 ET practitioners during routine ET with a smartphone mounted to a microscope. These embryos were subsequently transferred into eligible recipients. Clinical data including embryo stage, grade, pregnancy, and parturition outcomes were recorded and used to train GAI. Forty-two bovine embryologists of varying experience were surveyed to assess their ability to evaluate 10 embryo images based on stage, grade, and pregnancy likelihood. The 30-s video data of the same 10 embryos were processed by the trained GAI, and results from the GAI assessment and the embryologists’ evaluations were compared. An additional 194 bovine embryos were evaluated by a very experienced bovine embryologist, and GAI stage and viability predictions were compared. Results were analyzed with ANOVA or Student’s t-test using P < 0.05 for significance. Embryologists’ experience ranged from 1 to 40 years: 13 had 0–4 years of experience, eight had 5–9 years, 10 had 10–19 years, and 11 had >20 years. The survey revealed significant disparities in embryo stage assessments among embryologists of different experience levels (P < 0.05), with only 59.8% agreement across all participants. Agreement notably increased to 74.6% among experienced (10–19 years) and very experienced (>20 years) embryologists. In contrast, GAI demonstrated 70% agreement with all participants and 85% agreement specifically with very experienced embryologists, showing no statistical difference compared with the expert embryologists (P = 0.3). Notably, GAI matched or exceeded embryologists’ proficiency in identifying unfertilized oocytes, a task typically mastered by seasoned embryologists. In the broader study, GAI achieved (139/194) 71.6% agreement with the very experienced embryologists in identifying embryo stage and (193/194) 99.5% agreement in transferability (marked as Embryo Quality grade 1–2 = transfer, 3 = marginal, 4 = non-transferrable). In the event the GAI and embryologist did not agree on embryo stage, 48 of 55 predictions (87.3%) were only one stage apart, typically in disagreement over stage 6 and stage 7 embryos. This outcome again showed that GAI evaluates embryos comparatively to very experienced embryologists (P = 0.5). This study highlights the current state-of-art utilizing GAI to evaluate embryos, offering the reality of an automated and standardized embryo analysis and a transformative pathway to improve status quo.