P-150 Predicting fetal heartbeat in vitrified blastocysts through artificial intelligence analysis of post-warming images and videos

Abstract

Abstract Study question Is it possible to predict the fetal heartbeat of frozen blastocysts from post-warming time-lapse images and videos captured with EmbryoScope? Summary answer A significant positive correlation was found between the score given by the artificial intelligence algorithm and the fetal heartbeat rate of the vitrified blastocysts analyzed. What is known already Vitrification is the safest and most successful technique for cryopreserving human embryos, although some of the best morphologically graded blastocysts before freezing do not recover properly when thawed, collapsing or failing to re-expand. Artificial intelligence (AI) has been applied on videos and images of fresh embryos at different stages, and also on images of vitrified-warmed embryos. However, a single image does not give information about events such as collapses or re-expansion rate. This study evaluated the ability of an AI software to predict the fetal heartbeat of vitrified blastocysts from images and videos taken in the post-warming period. Study design, size, duration This retrospective single-center study included a video of the period between thawing and transfer of 537 vitrified blastocysts, and the last image of that video of 518 blastocysts (Feb 2022-May 2023). All blastocysts were vitrified and warmed using the Cryotop method (Kitazato, Biopharma, Japan), and placed in EmbryoScope (Vitrolife, Denmark) time-lapse incubators for 2-5 hours to let them re-expand until embryo transfer. Warmed embryos were assessed by experienced embryologists according to the ASEBIR morphological criteria. Participants/materials, setting, methods Images and videos were provided to the algorithm (EMBRYOAID v1.0, MIM Fertility), which returned scores from 0 to 10. Multivariate logistic regression analysis included embryo score and confounders: patient age and body mass index, oocyte age, oocyte origin (autologous vs. donated) and handling (fresh vs. vitrified), day of blastulation (5 vs. 6), preimplantation genetic testing (PGT-A), and post-thaw incubation time. Model performance was calculated using the area under the receiver operating characteristic (ROC) curve (AUC). Main results and the role of chance The mean score was different between embryos with negative (5.9±1.2) and positive (6.3±1.2)* fetal heartbeat outcome. The contribution of each increased unit of algorithm score to the fetal heartbeat was significant in both imaging (OR = 1.36 [95%CI: 1.16-1.60]**) and video (OR = 1.29 [95%CI: 1.12-1.50]*). The score predicted fetal heartbeat with AUCs of 0.61 in images and 0.60 in videos, and it performed better than the ASEBIR conventional morphology, which achieved an AUC of 0.58. The algorithm performance was higher in embryos not subjected to PGT-A (AUC=0.62) in images (n = 418) and videos (n = 435). Fetal heartbeat prediction was optimal in images of non-PGT-A embryos taken between 3.3-4h of post-warming incubation (n = 107) (OR = 1.76 [1.22-2.55]*), reaching an AUC of 0.74. In videos, the optimal prediction in non-PGT-A embryos was at 3.3-4.1h (n = 126) (OR = 1.93 [1.36-2.74]*), with an AUC of 0.69. 9.9% of the embryos collapsed and/or did not re-expand after thawing. These damaged embryos had an implantation rate of 15.2%, lower than the rest (46.8%**). Mean score differed between poor prognostic and normal embryos, both in images (4.8±1.1 vs. 6.1±1.1**) and videos (4.9±1.1 vs. 6.2±1.2**). Embryo score was related to the odds of collapse/non-re-expansion in images (OR = 3.07 [2.20-4.29]**) and videos (OR = 2.94 [2.13-4.05]**), with an AUC=0.81 *P<0.01, **P<0.001. Limitations, reasons for caution The algorithm tested in this study was developed externally. It was designed to evaluate videos of fresh embryo development and images at 120 h post-ICSI, but not post-thaw videos and images at days 5 and 6. The developers did not have access to any data or labels from our clinic. Wider implications of the findings Algorithm scoring on videos and images was related to fetal heartbeat rate and to collapse/non-re-expansion of vitrified blastocysts. Fetal heartbeat prediction performance reached 74% in images of non-PGT-A embryos taken at 3.3-4h post-warming. The application of predictive models in vitrified cycles may facilitate the choice of the best frozen embryo. Trial registration number Not applicable