O-127 Blastocyst expansion speed assay: a putative artificial intelligence-powered tool for embryo selection. Retrospective study involving 2184 blastocysts and 786 PGT-A cycles

Abstract

Abstract Study question Is blastocyst expansion-speed, examined through Artificial-Intelligence (AI) between time of blastulation (tB) and time of expanding blastocyst (tEB), associated with embryo blastulation and reproductive competence? Summary answer Blastocyst expansion-speed-assay (ESA) was significantly associated with euploidy and live-birth rates, being discordant with clinical embryologists’ ranking in 50% of embryonic cohorts. What is known already Blastocyst expansion is one of the earliest morphogenetic events in mammalian development. It is essential for the establishment of the blastocyst layout and requires trophectoderm integrity. Several studies suggested that expansion is a valuable feature for ranking and prioritizing blastocysts for transfer based on their developmental competence. Time-lapse technology (TLT) implementation in IVF allows a deeper understanding of blastocyst expansion dynamics. In this study, we combined TLT and AI to develop a blastocyst-ESA and investigate its association with embryo blastulation and reproductive competence. Study design, size, duration Retrospective study including 2184 blastocysts cultured in EmbryoScope incubators during 786 PGT-A cycles across 2013-2020. Videos were analyzed through an AI-powered tool (CHLOETM, Fairtility). The expansion-speed was calculated as [(Δ embryo proper area at tEB – embryo proper area at tB) / (Δ tEB - tB)]. ESA was tested for its association with embryo quality, euploidy and live-birth rate (LBR) in 548 vitrified-warmed single euploid transfers. A simulation of ESA putative clinical effectiveness was conducted. Participants/materials, setting, methods ICSI, trophectoderm biopsy of fully-expanded blastocysts without day3 zona-drilling, and qPCR/NGS to assess full-chromosome uniform aneuploidies were performed. tB and tEB, expressed as hours-post-insemination (hpi), embryo proper area (in µm2) at both these stages, and blastocyst quality (EQ-Score from 0 to 1) were all automatically recorded through CHLOE. Possible confounders (e.g., maternal age, male factor, and cause of infertility) were considered. Regression analyses were conducted to adjust the data. Main results and the role of chance The average ESA was 705±458 µm2/hour. Higher ESA was associated with higher EQ-Score. Euploid blastocysts showed higher ESA than aneuploid (761±465 µm2/hour versus 667±449 µm2/hour; p < 0.01). ESA was also associated with LBR per euploid blastocyst transfer (LB: 873±438 µm2/hour versus 736±467 µm2/hour; p < 0.01). Of note, maternal age showed no association with ESA. Multivariate regressions outlined a + 5.1%-increase in the chance of euploidy and a + 6.3%-increase in chance of LB, every +100 µm2/hour-increase in ESA. ESA and embryologists were discordant in grading top-quality embryos in each cohort in 50% of the cases. In 59% of the 352 cohorts where both euploid and aneuploid blastocysts were obtained, ESA would have ranked the former embryos as top-quality. In the 216 cycles with ≥2 euploid blastocysts obtained and ≥1 transferred, ESA would have prioritized (i) a competent embryo in 46% of the cases, (ii) an incompetent embryo in 20%, but (iii) in 33% its value could not be assessed (i.e., the top embryo according to ESA has not been transferred yet). When compared to the embryologists, ESA would have been (i) equally-effective in 49% of the cases, (ii) more effective in 12-24%, and (iii) less effective in 5%-26%. Limitations, reasons for caution Retrospective single-center study. Both continuous and sequential media were used, although they were not associated with the outcomes under investigation. To assess the clinical value of ESA, a prospective study among first transfers is warranted. Wider implications of the findings ESA is an automated, unbiased, and easily-applicable measurement that certainly deserves further appraisal. If validated in prospective studies, AI-based selection algorithms could include this assessment to increase their predictive power. Trial registration number None