#147 : Artificial Intelligence is Equally Capable of Predicting Viability of Embryos Created from Vitrified Oocytes, Despite Observed Differences in Morphokinetic Behaviour
Gabriella Pereira, Kelli Sorby- General Medicine
Background and Aims: With oocyte vitrification increasingly employed and the ever-expanding presence of artificial intelligence (AI) in the laboratory, it is essential to understand whether the same weight can be given to AI models’ predictions when vitrified oocytes are utilized.
Method: Embryos were cultured in Embryoscope+ and assigned an iDA score, generated after 112 hours in culture.
Data was collected from 186 vitrified oocyte (VO) cycles using a total of 1577 vitrified oocytes, along with control data from 43,957 fresh oocytes.
Data analysed included all standard morphokinetic parameters, iDA scores, embryo utilisation and clinical pregnancy.
Statistical analyses were performed using Logistic Regression, Fisher’s exact test and unpaired t-tests.
Results: The majority of cycles utilised clinic-recruited donor oocytes (70.7%), followed by autologous (25.9%) and known-donor oocytes (3.4%). No difference was observed in blastocyst utilisation between the three groups (48.0%, 48.1% and 52.9% respectively). In patients undertaking a single cycle utilising both fresh and vitrified oocytes, blastocyst utilisation was significantly higher in fresh oocytes (58.98% vs. 41.12%, p<0.0001), however, no difference in clinical pregnancy rate was observed (40.0% vs. 40.5%).
Fresh and vitrified oocytes displayed significantly different morphokinetic behaviour across all time points assessed. The magnitude of difference was greatest during compaction and blastulation, with VO embryos compacting on average 6.53 hours later than control embryos (p<0.0001) and commencing blastulation 4.58 hours later (p<0.0001).
When analysing iDA scores against the clinical pregnancy rate of fresh vs. vitrified oocytes, logistic regression showed no difference in areas under the curve (AUC) between the two groups (0.6431 vs. 0.6390, 95% Confidence Intervals 0.6111–0.6751 and 0.5474-0.7300 respectively), demonstrating no reduction in the AI’s ability to predict embryo viability despite their morphokinetic differences.
Conclusion: AI technology predicted clinical pregnancy equally well whether embryos originated from fresh or vitrified oocytes, despite confirmation of significantly delayed morphokinetic parameters in embryos formed from vitrified oocytes.