P-202 Using artificial intelligence platform coupled to an existing time system; external validation of an automatic embryo score to assist in selection

Abstract

Abstract Study question How does a novel artificial intelligence-based embryo evaluation system work in Geri® time-lapse systems? Summary answer Automatic score platform provided externally for embryos cultured in Geri® was associated with conventional morphology, euploidy, implantation and live birth. What is known already Artificial intelligence has been making headway in assisted reproduction in recent years. Many companies have developed different models for automated embryo evaluation and selection, such as CHLOETM software (Fairtility, Israel). According to the developers, it is an orchestration of cutting-edge morphological and morphokinetic AI algorithms, trained over 100,000 embryo videos and tens of millions of images. Different laboratories validated its use in specific time-lapse systems (Embryoscope, Vitrolife). However, this is the first time that an objective and independent review of CHLOETM has been performed on Geri® (Genea Biomedx, Australia) time-lapse system. Study design, size, duration This retrospective analysis included 3568 embryos from 417 patients that underwent IVF treatments in a single center. Embryos were cultured in Geri® (Genea Biomedx, Australia) time-lapse systems by single step media (Geri Medium, Genea Biomedx, Australia) and routinely evaluated by senior embryologists by using Connect and Assess software according to the ASEBIR criteria (from A-high quality to D-low quality and excluded embryos). Then, embryos were automatically scored by CHLOETM from 0 to 1. Participants/materials, setting, methods Automatic embryo score was compared with conventional morphology (n = 3568 embryos), ploidy (n = 467 embryos), and clinical outcomes for single blastocyst transfers (n = 461). Main results and the role of chance The comparison between the embryo score provided by CHLOETM and the category assigned by embryologists showed a direct association*. The means were 0.97 ± 0.10 for A (n = 123); 0.89 ± 0.21 for B (n = 842); 0.74 ± 0.30 for C (n = 607), 0.24 ± 0.31 for D (n = 997) and 0.15 ± 0.25 for excluded embryos (n = 403). The automatic embryo score for oocytes that failed to fertilize was 0.06 ± 0.19 (n = 596). Regarding the chromosomal status, embryos with normal content had significantly higher score than abnormal ones. Following results are presented per quartiles of similar sample size: the euploidy rates were 35.9% for score < 0.81 (n = 117), 40.8% for score 0.81-0.96 (n = 120), 48% for score 0.96-0.99 (n = 125) and 58.1% for score >0.99 (n = 105)*. Implanted embryos achieved significantly higher marks than non-implanted embryos: 0.93 ± 0.15 (n = 251) vs. 0.85 ± 0.25 (n = 210)*. Also, automatic embryo score was higher for embryos that led to a live birth than those that did not: 0.94 ± 0.15 (n = 188) vs. 0.86 ± 0.24 (n = 243)*. Focusing on top quality embryos (A+B), the score means were 0.94 ± 0.16 for implanted good quality embryos 0.88 ± 0.22 for non-implanted ones* (*p<0.05). Limitations, reasons for caution This project is limited by its retrospective and single-center nature. Multicenter validation would be necessary to ensure it is a safe and effective method of embryo assessment. Wider implications of the findings In addition to verifying that automatic scoring agrees with embryologists, this study demonstrated its ability to be applied in Geri Time-lapse incubator to distinguish between potential embryos with similar morphological characteristics helping embryologist to make decisions. Trial registration number NOT APPLICABLE