P-136 An embryo evaluation algorithm based on deep learning correlates strongly with cell number and degree of fragmentation on day 2 and 3

Abstract

Abstract Study question Does the embryo evaluation model iDAScore v2.0 correlate with manually assessed cell numbers and fragmentation grade in early-stage embryos? Summary answer iDAScore v2.0 significantly correlated with morphological assessment of preimplantation embryos on days 2 and 3. Correlation between cell numbers and iDAScore decreased with increasing fragmentation. What is known already Time-lapse technology, including continuous documentation of embryo development, has enabled the implementation of more automated scoring systems, based on images and video sequences. iDAScore (ranging from 1-9.9) is a fully automated annotation-free scoring system based on deep learning and has been shown to discriminate between implanting and non-implanting embryos to a similar extent as KIDScore, which required manual annotations. iDAScore v2.0 also demonstrated a good agreement between predicted probabilities and observed implantation rates. However, not much is known regarding the correlation between specific morphological day 2 and day 3 variables and iDAScore v2.0. Study design, size, duration Retrospective observational study from two ART centers, together performing > 2000 OPU per year. The study includes all normally fertilized embryos (n = 5378) from 1792 treatments between 2016 and 2021 where fresh single embryo transfer was performed on day 2 or day 3 post fertilization. The association of iDAScore v2.0 with manual assessment of cell numbers and grade of fragmentation on day 2 (treatments = 762, embryos = 2170) and day 3 (treatments = 1030, embryos = 3208) was analysed using video sequences. Participants/materials, setting, methods Inclusion criteria were treatment cycles with own (autologous) oocytes and embryo transfer on day 2 or 3. Exclusion criteria were cycles with double embryo transfers or PGT cycles. Clinical and laboratory processes were performed according to standard procedures. Time-lapse culture and embryo annotations were performed using the Embryoscope+® with GTL® culture medium (Vitrolife, Gothenburg, Sweden). Wilcoxon rank sum test was used to compare iDAScores between groups. Main results and the role of chance Day 2: For the number of cells, iDAScore differed between embryos having < = 3 cells compared to > = 4 cells (medians: 2.3 vs. 4.0, p < 0.001). The median iDAScores for fragmentation grade 0%; <20%; <50%; > 50% were 4.7; 3.9; 2.6 and 2.1, respectively. Furthermore, 0% was significantly different from each of the other groups (p < 0.001). When cell numbers and fragmentation grade were combined, it was seen that having < = 3 cells versus > = 4 was significantly correlated to iDAScore when the fragmentation grade was scored as 0%, but this association decreased as the fragmentation grade increased. Day 3: A similar pattern was seen for day 3 morphology analyses. It was found that having < = 5 cells vs. > = 6 cells correlated to a significant difference in iDAScore (medians: 2.5 vs. 4.4, p < 0.001). The median iDAScores for fragmentation grade 0%; <20%; <50%; > 50% were 5.6; 3.8; 2.6 and 2.3, respectively. Furthermore, 0% was significantly different from each of the other groups (p < 0.001). When combining cell numbers and fragmentation grade in the same analysis, the association of cell numbers to iDAScore decreased as fragmentation increased. Limitations, reasons for caution The present study is of retrospective design, with analysis of data from two clinics, and the results may not apply in a wider context. In addition, only two morphological variables, cell numbers and fragmentation, were included. Wider implications of the findings Our results show that there is a strong correlation between iDAScore v2.0 and the number of cells and the fragmentation grade of the embryo, both for each individual variable and when combined. This indicates that iDAScore v2.0 identifies some of the well-established morphological evaluations for cleavage stage transfers. Trial registration number Not applicable