O-121 Temporal development of scores from a time-lapse based artificial intelligence provides no additional benefit compared to the latest score

Abstract Study question Does the temporal development of scores from an artificial intelligence (AI) based embryo selection model provide additional predictive power compared to latest score? Summary answer Earlier predictions no given prediction. Knowing AI is thus not beneficial for selection. What known already Traditionally, grading takes into consideration final morphology score together with history events. Recent publications have shown that time-lapse models trained on clinical outcomes can automatically rank embryos by likelihood implantation equal or surpass traditional methods, without necessity manual evaluation. AI-based methods data may account impact full implantation. In this study, we examine if addition previous adds performance model. design, size, duration A retrospective multicenter study transferred (n = 2422) (KID) between 2012-2020 21 international clinics. Embryos were cultured in EmbryoScope incubators at least 5 days. Both single and multi-embryo transfers fresh warmed cycles included. classified as positive (KIDp) negative (KIDn) presence fetal heartbeat. Implantation was evaluated cleavage- blastocyst stages a 3D CNN model; iDAScore v2.0. Participants/materials, setting, kernel-based conditional independence test used evaluate x hours post insemination (hpi) extra information y hpi. The estimates prediction after hpi does any regards predicting KIDp when available. These likelihoods (p-values) Bonferroni corrected adjust multiple comparisons. Main results role chance computed all 40, 44, 64, 68, 112, 116 corresponding early median time transfer day 2, 3, 5. For points, showed there added knowing earlier point (p > 0.21). Similarly, it tested later will improve This always case < 0.01) except close (68 vs 64 hpi; p 1.0) (44 40 1.0). reflect little during 4-hour period stage. above shows knowledge using available score. means videos likely considers events context entire history. Limitations, reasons caution only looks videos. There potential bias higher quality having been chosen transfer. It possible do generalize simpler non-3D CNNs. Wider implications findings introduce subjectivity might lower rates significant improvement. Trial registration number applicable