Improving Morphokinetic Classification of Embryo Developmental and Implantation Potential Using Artificial Intelligence Tools
2Department of Cell and Developmental Biology, Institute of Life Sciences, Hebrew University of Jerusalem
3Fertility and IVF Lab, Soroka Medical Center
4Faculty of Health Sciences, Ben-Gurion University of the Negev
5Infertility and IVF Unit, Kaplan Medical Center
6Infertility and IVF Unit, Beilinson Women's Hospital, Rabin Medical Center
7School of Computer Science and Engineering, Hebrew University of Jerusalem
8IVF Unit, Department of Obstetrics and Gynecology, Hadassah Medical Center
Introduction: Current morphokinetic-based classifiers fail to evaluate accurately the developmental competence of cleavage-stage embryos. Therefore, the practice is to transfer multiple embryos to obtain reasonable pregnancy rates. Accurate identification of high-quality embryos will shorten time-to-pregnancy and reduce multiple pregnancy rates.
Aim: Optimizing the accuracy and automation of embryo classification algorithms at cleavage stage (Day-3).
Material & Methods: We generated an expansive database that includes time-lapse videos, clinical protocols, maternal information, transfer / freeze statistics, and implantation rates of tens of thousands of embryos collected from five IVF clinics. We employ state-of-the-art statistical, feature-based machine learning and deep learning methods to perform a quantitative analysis of preimplantation development and to design classification algorithms for predicting embryo blastulation and implantation at early stage.
Results: Based on our large dataset, we define the temporal distributions of morphokinetic events and analyze how these distributions depend on maternal age, number of retrieved oocytes and hospital. While positive and negative implantation embryos show nonsignificant statistical differences, our classification algorithms predict blastulation and implantation rates with area under the curve >0.9 and >0.7, respectively. By integrating deep learning algorithms that identify successive cleave events, we advance towards standardized, automated and accurate classification of embryo developmental potential.
Conclusions: Harnessing advanced artificial intelligence methods improves accuracy and facilitates the transition towards automated, standardized and cost-effective embryo classification algorithms to support single embryo transfers and shorten the time to pregnancy.
Powered by Eventact EMS