Association between embryo development and early pregnancy loss revealed by artificial-intelligence-annotated kinetic events

Abstract

Research question: Can artificial intelligence (AI)-powered annotation of numerous biological events help to uncover an association between embryonic kinetics and early pregnancy loss?

Design: A multicentric retrospective analysis was conducted on 37,717 embryos (7028 egg retrievals completed between 2017 and 2022, 13 centres in France and Spain, three time-lapse systems). The videos of developing embryos were analysed using AI to detect developmental events, including cleavage from two cells to eight cells (t2, t5, t8), start of blastulation (tSB) and blastocyst (tB). Associations between embryo kinetics and transfer outcomes (lack of early pregnancy, early pregnancy loss and clinical pregnancy) were investigated using univariate and multivariate logistic regressions.

Results: Optimal kinetics for clinical pregnancy were found to differ from those of embryos which led to early pregnancy loss (P < 0.001). Although embryos that developed more quickly overall (tB-t2 = 77.18 ± 8.33 h) were the most likely to lead to early pregnancy, it appeared that, beyond early pregnancy, embryos that showed deceleration during cleavage (tsB-t5 = 46.05 ± 13.60 h for clinical pregnancy versus 41.6 ± 17.08 h for early pregnancy loss) followed by acceleration during blastulation (tB-tSB = 9.81 ± 5.04 h for clinical pregnancy versus 12.73 ± 5.69 h for early pregnancy loss) were the most likely to lead to clinical pregnancy. Conversely, embryos that presented the fastest cleavage and longest blastulation kinetics were the most likely to lead to early pregnancy loss.

Conclusions: Analysing numerous kinetic combinations annotated with AI reveals patterns which can help to distinguish embryos that are competent enough for early pregnancy as opposed to clinical pregnancy. Detecting such subtle kinetic differences could add transparency to algorithms by pinpointing which phases of embryo development may be predictive of pregnancy loss.

Keywords: Artificial intelligence; Early pregnancy loss; Embryo development; Time lapse.