P-111 External validation of a model for prioritizing day 3 embryos for transfer based on deep learning and time-lapse images

Abstract Study question Could an objective embryo assessment by iDAScore version 2.0 perform as well as a conventional morphological assessment according to the Istanbul consensus report? Summary answer iDAScore version 2.0 performed as well or even better than conventional embryo assessment in fresh day 3 embryo transfer cycles. What is known already The pregnancy prediction performance of iDAScore version 1.0 has been confirmed in a large cohort study with vitrified-warmed blastocyst transfer, the validation results showed that it performs as well or better than conventional embryo evaluation. iDAScore version 2.0 was developed for transfers on both day 2, day 3 and day 5 transfers, however, the performance of iDAScore version 2.0 has not been verified with independent datasets. Study design, size, duration A single-center large cohort retrospective study from an independent validation test was carried out, including 4328 cleavage embryo transfers cycles performed between January 2020 and June 2022. Participants/materials, setting, methods The selection of embryos for transfer was based on conventional morphological assessment model according to Istanbul consensus report, and the transferred embryos were retrospectively evaluated by the iDAScore version 2.0 model. The pregnancy prediction performances of the two day 3 embryo assessment models were compared using the area under curve (AUC) for predicting the FHB for all cycles and subgroup analysis(SET and DET; female age <35 and ≥ 35 year; number of blastomeres <8c and ≥8c). Main results and the role of chance The FHB rate in the iDAScore (2.0-3.9) group was lower than in all iDAScore group, iDAScore (4.0-5.9), and (6.0-8.0) groups within the poor, fair and good group. Additionally, for fair and good group, the FHB rate in iDAScore (6.0-8.0) group was higher than that in other iDAScore groups. The AUCs of iDAScore were significantly greater than the conventional morphology assessment, for all cycles (0.62 vs. 0.59, P < 0.05), SET cycles (0.63 vs. 0.59, P < 0.05) and DET cycles (0.61 vs. 0.59, P < 0.05). For the sub-groups of age, the AUC of iDAScore was significantly greater than the conventional morphology assessment in < 35 group (0.62 vs. 0.59, P < 0.05). However, there was no significant difference between the two assessment model in ≥ 35 group (0.60 vs. 0.58, P > 0.05). For the sub-groups of the number of blastomeres, a significantly greater AUC was observed for iDAScore than for conventional morphology assessment, for both the <8c group (0.67 vs. 0.56, P < 0.05) and the ≥ 8c group (0.58 vs. 0.55, P < 0.05). Limitations, reasons for caution the limitation of the findings is that all analyzed embryos were selected previously by morphology. Although the conventional morphological assessment is a widely used morphological grading system, it is subjective and susceptible to inter-observer variation. Wider implications of the findings iDAScore version 2.0 may be a promising tool for selecting day 3 embryos with the highest likelihood of implantation. A fully automated evaluation system also significantly reduced the time used for embryo selection and eliminates subjectivity. Trial registration number none