P-134 An artificial intelligence method based on time-lapse images and deep learning may predict if a day2/3 embryo will form a utilizable blastocyst

Abstract

Abstract Study question Can iDAScore v2.0 predict likelihood of development into a utilizable blastocyst from time-lapse images captured during the first two or three days of embryo development? Summary answer iDAScore v2.0 predicts utilizable blastocyst formation from time-lapse images of early embryo development. Specific score thresholds can be established to ensure high specificity of prediction. What is known already Treatment efficacy would benefit from a reliable method to predict the potential of early embryos to develop into utilizable blastocysts and select fresh IVF treatment cycles suitable for blastocyst culture. Reports indicate that time-lapse images captured during the first few days of development and analyzed by deep learning methods may be an accurate approach. A fully automated AI model, iDAScore, is reported to be more accurate than routine morphology and morphokinetics for prediction of implantation. However, the applicability of iDAScore calculated on day 2 and 3 as a tool to predict blastocyst development has yet to be tested. Study design, size, duration This is a retrospective multicenter cohort study including time-lapse videos of 9111 embryos cultured to blastocyst stage, from 1283 fresh IVF treatments between 2016-2021 at two ART clinics together performing > 2000 OPU per year. Images captured during the first days of culture were used to calculate iDAScores (at 48 and 66 hours post insemination) and determine the ability to discriminate embryos likely to develop into utilizable blastocysts according to Gardner score of 3BB or more. Participants/materials, setting, methods Patients were required to be undergoing IVF treatment using their own (autologous) oocytes whereby all zygotes were cultured up to the blastocyst stage in GTL® culture medium using the Embryoscope+® time-lapse system (Vitrolife, Gothenburg, Sweden). At least two experienced embryologists graded blastocysts on day 5/6, selecting utilizable blastocysts for transfer/vitrification. To evaluate the performance of iDAScore to predict development into utilizable blastocysts, sensitivity, specificity, and area under ROC curve (AUC) values were calculated. Main results and the role of chance The blastocyst utilization rate was 43.8% (3990 utilizable blastocysts of 9111 day 2/3 embryos developing from 2PN zygotes). Prediction of utilizable blastocysts obtained AUCs of 0.768 (0.759 - 0.778) and 0.782 (0.773 - 0.791) for day 2 and day 3, respectively. These results suggest IDAScore prediction performed slightly better when calculated on day 3 than on day 2. The ability to discriminate between utilizable and non-utilizable blastocysts at different thresholds of iDAScore on day 2 and on day 3, suggests that embryos should have a minimum iDAscore of 5 on day 2 to obtain a high specificity of 89%, and a threshold of 6 on day 3 to obtain a high specificity of 88%. Lower sensitivities were observed for these potential thresholds, 35% on day 2 and 41% on day 3. Limitations, reasons for caution This is a retrospective analysis where most cycles have at least five zygotes that supported subsequent blastocyst development, limiting relevance for patients with fewer zygotes. IDAScore is developed to predict implantation potential, not utilizable blastocyst, thus could be an underestimation of the potential performance of a model for utilization prediction. Wider implications of the findings The iDAScore v2.0 may provide valuable information for embryologists and be useful as a prognostic tool to select patients with high chance to obtain blastocysts based on day2/3 time-lapse images and thereby may achieve similar live birth rate per transfer with no need for extended culture. Trial registration number Not applicable