O-099 The inner cell mass (ICM) regions can be used to predict pregnancy with a similar accuracy as the original images of the entire embryo

Abstract Study question Can clinical pregnancy be accurately predicted using only ICM and TE regions in an embryo microscopy image? Summary answer The AI model segmented images with a similar accuracy as original images. What is known already Previous research suggests that are key factors determining grade pregnancy. Grad-CAM on the CNN models various literature also highlight significance of regions. However, there have been no studies to compare assess importance two prediction. This study aims test each region by evaluating performance after segmenting design, size, duration We performed retrospective single static 2,555 Day 5 blastocysts from seven vitro fertilization (IVF) clinics between June 2011 May 2022. were collected standard optical light microscopes matched metadata such outcomes. manually labeled examined embryologists over 10 years experience. Participants/materials, setting, methods defined positive indication presence gestational sac (G-SAC). built 4 (“Original model”), (“ICM (“TE model”) & model”). performances measured through 3-fold cross validation. Main results role chance In validation, mean deviation AUROCs 0.753±0.005 for Original model, 0.733±0.003 0.724±0.016 0.690±0.011 respectively. DeLong showed significant differences AUCs (p = 0.01), 0.0003). Positive correlations found prediction values pearson correlation [Confidence interval] image (0.426 [0.343, 0.495]), (0.345 [0.266, 0.419]) (0.223 [0.140, 0.304]). highest was least accurate expected. it noteworthy outperformed model. Although well plays implantation, forcing utilize both predict could muddled analysis. proved can successfully Limitations, reasons caution has limitations due its nature, IVF centers. Further larger dataset warranted offset focus, magnification color Wider implications findings similarly high lowest performances. Utilizing well-focused may enhance models. Trial registration number not applicable