O-120 Predicting Embryo Ploidy Status Using Time-lapse Images

Abstract Study question Can deep learning models using time-lapse images of embryo development be used to predict ploidy and provide supplemental information embryologists for clinical decision-making? Summary answer We developed a general MDBS-Ploidy model that uses maternal age quality scores status. What is known already Ploidy status, or the presence lack chromosomal abnormalities, an important factor in successful pregnancy. Embryos with abnormalities are classified as aneuploid, whereas those without euploid. With advent artificial intelligence computer vision collection large IVF-related datasets combining images, videos, outcomes, variety methods have been automatically assess other characteristics from sequences. To our knowledge, video classification validated by heterogeneous data both score lacking. design, size, duration The training dataset consisted 1,998 sequences captured Embryoscope®. Time-lapse image around 360–420 frames during 5 days development. PGT-A results were ground truth labels all prediction tasks, embryos euploid aneuploid. also included such blastocyst (BS) at time oocyte retrieval. Participants/materials, setting, consists two steps. first step day-5 input Bidirectional Long Short-Term Memory architecture added output layers multitasking. In second step, predicted score, addition age, status logistic regression model. evaluated performance area under receiver-operating-characteristic (AUROC) on validation dataset. Main role chance For module, Pearson correlation between 0.70 dataset, suggesting moderate strength. As aneuploidy can discriminate euploidy AUROC 0.76 ± 0.002. performs comparably trained embryologist-annotated score. replicate these comparative external well, namely Embryoscope+® 1,000 IVI Valencia Spain 543 embryos. Moreover, completely automated, requiring only 96 112 hpi This allows adapted clinically interrupting ongoing workflows. provides certain level explainability; use intermediate (from module) determine why recall 0.84 0.004 shows promise successfully selecting believe most viable embryo. Limitations, reasons caution Our primarily microscopy. Clinics access this technology will unable Wider implications findings relevant making decisions whether may chromosomally normal. Automated instrumental who currently annotating manually. Future iterations theoretically adopted into practice because it end-to-end fully automated. Trial registration number not applicable