Noninvasive time-lapse 3D subcellular analysis of embryo development for machine learning-enabled prediction of blastocyst formation

Abstract In developmental biology and in vitro fertilization (IVF), image-based assessment of embryos is pivotal. Traditional methods in clinical IVF have been constrained to 2D morpho-kinetic profiling and manual selection, hindered by the absence of noninvasive techniques for quantitative 3D imaging over extended durations. Here, we overcome these limitations by employing low-coherence holotomography to monitor mouse preimplantation embryo development from the 2-cell stage to the expanded blastocyst. This approach enables the generation of 3D refractive index tomograms of unlabeled embryos, facilitating the observation of subcellular developmental dynamics. We investigated the 3D spatiotemporal profiles of embryo development, identifying key morpho-kinetic parameters that distinguish between embryos with differing developmental outcomes—specifically, Grade A embryos that successfully progressed to expanded blastocysts within 72 hours, and Grade C embryos that did not. Using machine learning, we demonstrate the 3D morpho-kinetic parameters can offer a noninvasive, quantitative framework for predicting embryos with high developmental potential.