P-565 Moving towards non – invasive euploid blastocyst prediction. Retrospective euploid blastocyst evaluation using different morphokinetics and Artificial Intelligence (AI) models

Abstract

Abstract Study question Could Preimplantation Genetic Testing for Aneuploidies be replaced by an Aritificial Intelligence model which evaluates blastocysts and selects the most suitable ones for embryo transfer? Summary answer The use of AI for euploidy prediction has not been validated in a large scale and thus it should not be used to replace PGT-A. What is known already PGT-A is an invasive method that may affects blastocysts implantation potential. Both the biopsy protocol and the experience of the embryologist who performs the biopsy can influence the outcome. Efforts are made to develop non-invasive methods for PGT. Many AI systems have been developed in assisted reproduction.Time- Lapse systems have a specialized AI software, that assist in the evaluation of the embryos and the selection of the most suitable ones based on their morphokinetic characteristics. Other AI systems, such as ERICA, evaluate embryos and provide a measure of their euploidy, serving as an alternative method for non-invasive preimplantation genetic testing. Study design, size, duration The retrospective study was conducted between January 2018 and December 2020. A total of 585 blastocysts, derived from 144 women who had undergone an IVF – PGT-A cycle, were included. The embryos were cultured in a TL incubator and evaluated using KID Score D5 (Vitrolife), iDa Score (Vitrolife) and ERICA (ERICA). Participants/materials, setting, methods The study was conducted in a private IVF clinic. Embryos from women with an average age of 38 years who had a history of recurrent miscarriages and/or >3 failed IVF cycles were included. Biopsy was performed in all blastocysts that were cultured in TL incubators up to day 5-6. Statistical and logistic regression analysis were conducted to assess the comparison between the score of the three AI programs and the result of the biopsy. Main results and the role of chance ERICA seems uncapable to separate euploid from aneuploid embryos. T-test showed that there is no statistically significant difference between the two categories (mean_difference=0,02, p = 0,3>0,001), while the effect size (clinical value) of the difference was very low (0,085). This means that this program cannot replace PGT-A. According to the iDA Score, although there is a difference in scores between euploid and aneuploid embryos, their difference, according to t-test, is not statistically significant [t(581)=2,56, p = 0,01>0,001], while the effect size is low (0,217). According to KID Score D5, there is a statistically significant difference in scores between euploid and aneuploid embryos [t(514)=4,4, p < 0,001], which shows that this program could replace PGT-A in some cases. However, the effect size of the difference (0,37), as well as the determination coefficients are quite low. Finally, the KID Score D5 succeeds in discovering aneuploid embryos in a higher percentage than euploid ones. Limitations, reasons for caution Women under the age of 35 were not included in the study. Permission from the National Greek Authority of Medical Assisted Reproduction was needed from every patient to undergo PGT-A. The size of the sample might limit the conclusions. A larger study would help verify the results of the study. Wider implications of the findings AI models can be used as good indicators for embryo selection/prioritization and predict the euploid status of the embryos. Non-invasive euploid prediction models could be a very useful tool for embryologists especially in countries where law restrictions are strict against the cohort of patients that are interested in PGT-A cycles. Trial registration number Not Applicable