Improving the assessment of embryo developmental potential via morphokinetic forecasting of future events using language modeling

In IVF treatments, accurate assessment of the developmental potential of embryos to implant is essential for reaching reasonable pregnancy rates while shortening time-to-pregnancy. Hence, clinical guidelines recommend extended incubation to blastocyst transfers, which provide better evaluation of embryo developmental potential. However, cleavage stage transfer is often favored owing to various clinical considerations. To improve embryo assessment of cleavage stage embryos without extended incubation, we present a computational strategy for forecasting future morphokinetic events. Motivated by the advances in language modeling, we adapt generative pre-training to forecast future morphokinetic events based on the sequence of present events. We demonstrate < 12% forecasting error in forecasting up to three consecutive events. A new policy is proposed that combines morphokinetic forecasting and assessment of the risk of embryo developmental arrest. Using this policy, we demonstrate an improvement in the prediction of known implantation outcome of day-3 embryos from AUC 0.667 to 0.707. We expect morphokinetic forecasting to address the inherent hurdles in the selection of cleavage-stage embryos for transfer. In addition, we hope that demonstrating for the first time the utilization of language modeling on non-textual data in healthcare will stimulate future applications in reproductive medicine and other disciplines.