Automated Prediction of Blastulation and Implantation Potentials using Deep-Learning

Yoav Kan Tor 1 Ity Erlich 1 Tamar Amitai 1 Yuval or 2 Zeev Shoham 2 Arieh Horowitz 3 Iris Har-Vardi 4 Matan Gavish 1 Assaf Ben-Meir 3 Amnon Buxboim 1,5
1Hebrew University of Jerusalem, Israel
2Hebrew University of Jerusalem, Israel
3Hebrew University Hadassah Medical Center, Israel
4Ben-Gurion University of the Negev, Israel
5Hebrew University of Jerusalem, Israel

In IVF treatments, early identification of embryos with high implantation potential is required for avoiding clinical complications to the newborn and to the mother that are associated with multiple-embryo pregnancy and for shortening time to pregnancy. The incorporation of time-lapse incubators in IVF clinics provides continuous visual monitoring of the embryos while maintaining them in optimal culture conditions. Here we employ deep learning classification algorithms, which are trained on the time-lapse recordings of embryo preimplantation development, for predicting the developmental potentials to undergo blastulation and to implant within the uterus. To facilitate deep learning, we compiled an expansive database that consists of video files and associated clinical metadata of 4,800 blastulation-labeled and 4,500 implantation-labeled embryos obtained from five medical centers across Israel. Our prediction of embryo implantation outperforms the current state-of-the-art classifier. Our deep learning classification algorithms offer an automated, standardized and accurate substitute to human-based evaluation of embryonic developmental competence. We anticipate that these algorithms will advance IVF performance by facilitating single-embryo transfers and by shortening time to pregnancy.









Powered by Eventact EMS