Development of a 3D holographic video microscopy associated with convolutional neural networks (CNN) to identify novel markers of preimplantation embryonic health – LIVE3D_CNN

The aim of this project is to develop a ground breaking holographic label-free 3D time-lapse microscope to follow the development of living embryos directly in an incubator. Our device will measure original physiological parameters to identify new markers of mammalian preimplantation embryonic health and hence to pinpoint the embryos with the highest chances of implantation and development.
Imaging of human preimplantation embryos is performed routinely in tens of thousands of in vitro fertilization (IVF) laboratories worldwide for millions of infertile couples. During the IVF process, many embryos arrest in the initial divisions or after implantation and one of the main challenge of IVF is to select and transfer the embryos that have the highest developmental and implantation potential, in order to maximize the chances of a rapid pregnancy. Currently this choice is largely made by visual inspection taking into account mainly the morphological aspect of the embryos. Embryoscopes have recently been made available, permitting to follow the kinetics of embryonic development and bringing additional criteria to identify the “best” embryos. Their use has however not significantly improved pregnancy rates and no clear-cut criteria has yet permitted to identify the best embryos.
We believe that the use of holographic microscopy with 3D reconstruction driven by powerful physical neural networks will permit to identify novel markers of preimplantation embryonic health that cannot be seen by the technologies currently available. The first phase of Live3D-CNN will be centered on the development of a 3D phase microscope equipped with convolutional neural networks. Our 3D microscope will be compact, easy to use, and will work directly in an incubator. Our neural network algorithms rely on a pioneering 3D reconstruction method, i.e. a physical neural network directly encoding the physical laws of light propagation in terms of refractive index distribution and light propagation. The second part of the project will develop and assess the quantification of dry mass measurements in 3D of different embryonic structures by means of neural networks. Our microscope will be the first system to quantify the dry mass of polar bodies and cell nuclei, and the first to permit cell counting in living embryos up to the blastocyst stage. Measurements of dry mass will also be correlated with cell morphology, nuclear shape, modality of cell division, and 2D intracellular particle movement. Our prototype will be tested on living early-dividing mouse embryos in both healthy and diseased conditions to constitute a strong database for embryonic classification. Genetic tests will be further correlated with the measured novel markers for a more accurate classification of aneuploid embryos. Results from these analyses will feed a predictive neural network to foresee the embryonic fate and guide decision over pre-implantation procedures.
The project will be carried out by a multidisciplinary consortium regrouping engineers and experts from the CEA-Leti in the field of optical instrumentation and artificial intelligence (AI) and infertility and early mammalian specialists from the GETI-IAB team permitting to unite all the required complementary skills required for the success of the project. Both teams have been collaborating on this novel and challenging subject for three years permitting to establish the baseline for this project. Exciting preliminary data give us great confidence in the success of this original project expected to have important repercussions on both disciplines independently.