Early assessment of the developmental potential of preimplantation embryos to implant and generate a live birth is important for obtaining insights into the critical developmental cues and for advancing in vitro fertilization (IVF) based assisted reproductive technologies (ART). To address this unmet need, we combine advanced machine learning methods and in situ mechanical characterization of oocytes and embryos. We established an expansive database that consists of 70,000 preimplantation human embryos, which includes 3D video recordings, morphokinetic annotations, maternal information, clinical protocols, and transfer, implantation and live birth rates for each embryo. Based on this database, we trained convolutional neural networks (CNN’s) and deep learning classifiers. Our classification algorithms perform automated morphokinetic annotation and predict embryo blastulation and implantation rates at early stages of development with exceptional accuracy. In parallel, we designed a culture-plate based device for performing non-invasive measurements of the viscoelastic properties of the embryos. We define mechanical parameters that differentiate between low-competence and high-competence embryos. Taken together, we present automated, non-invasive, standardized and accurate tools for early evaluation of embryonic competence and highlight physical and dynamic parameters that correlate with embryonic functions.
