DEEP LEARNING FOR EMBRYO SELECTION AND OUTCOME PREDICTION IN IN VITRO FERTILIZATION

Abstract

Selecting the embryo with the highest implantation potential remains one of the most consequential and subjective decisions in assisted reproduction, traditionally guided by morphological grading systems with well-documented inter- and intra-observer variability. Convolutional neural networks trained on time-lapse or static blastocyst images now automate this grading step, with several architectures demonstrating agreement with senior embryologists that exceeds agreement between embryologists themselves [1]. Beyond morphology alone, deep learning models have been extended to ploidy prediction from non-invasive time-lapse video, offering a potential alternative or adjunct to invasive preimplantation genetic testing [2].