Developing the MechanoPLATE:
We have designed a new device-based technology termed the MechanoPLATE. It that facilitate continuous assessment of the viscoelastic properties of oocytes during in vitro maturation and embryos during preimplantation development. Using this device, mechanical assessment of embryo quality can be performed in parallel to time-lapse recording of preimplantation development which allow further evaluation of the developmental quality using machine learning-based classifiers. These methods are non-invasive and are performed inside incubators that maintain optimal culture conditions that are used for example in IVF clinics.
Oocyte and embryo mechanics:
Using the MechanoPLATE, we tested the role of whole-oocyte and embryo mechanics in predicting developmental potential but unfortunately it proved not to correlate with functional parameters. However, we further investigated and found discovered that the deformation dynamics of the ooplasm, which is the internal cellular mass of the oocytes and the embryos, responds to applied forces in a very interesting fashion. Low quality oocytes/embryos demonstrate a continuous deformation starting from the time at which forces were applied. However, high quality oocytes/embryos resist the applied loads for a few seconds before yielding. We further studied the mechanisms that are responsible for this interesting phenomenon. We found that the cortical actin cytoskeleton, and specifically a branching initiator of actin filaments termed the Aep-2/3 complex, was responsible for rendering mechanical resilience to the ooplasm of high quality oocytes/embryos.
Assessing embryo quality using machine learning:
One of our project's main effort was to develop machine-learning based algorithms that will be able to improve IVF treatments by evaluating the developmental potential of embryos during preimplantation stages based on their continuous visual appearance. To this end, we assembled the a large dataset of time-lapse imaging of human embryos that we collected from four medical centers in Israel. Each embryo is clinically labeled and the transferred embryos are annotated by outcome. Using this expansive database, consisting of ~70,000 embryos, we developed machine-learning algorithms for improving IVF treatments as described below.
Assessing the potential of embryo to implant in the uterus:
We harnessed advanced deep learning tools to generate fully automated embryo classification algorithms. One algorithm utilizes the "raw" images of the embryo to assess embryo quality. A second algorithm is used to perform fully automated annotation of how the embryos develop (this is termed morphokinetic annotation). Using these classifiers, we predict the potential of each embryo to implant within the uterus and generate pregnancy - a property which most embryos fail to perform. Our algorithms are more accurate than existing classifiers, and provide a standardized fully automated evaluation of the developmental potential of each embryo.
Predicting 1st trimester miscarriage
First trimester miscarriage is a major clinical problem accounting for ~one out of nine pregnancies. To date, there are available options for assessing the potential of an embryo during preimplantation development before it is selected to be miscarried. Hence, we trained a machine-learning model to predict the potential of the transferred embryo to reach live birth or to undergo 1st trimester miscarriage. This task was supported by the fact that our database is sufficiently large to include transferred embryos that were miscarried and to learn differential visual markers as compared with live embryos that reached live birth.
Impact on IVF treatments:
With the support of the ERC, we developed both mechanical tools and computational algorithms that assess multiple developmental functions of preimplanted embryos. Furthermore, we developed a device that can combine these tasks. The expected impact of this work is to serve as decision-making algorithms that will allow the clinical staff to better select which embryos optimize live birth rates and to select these embryos for transfer. This advancement is relevant not only to human IVF treatments and to fertility preservation treatments, but also to cattle IVF procedures that underlie the dairy industry and are targeting existing unmet needs.
