Development of an intelligent system for the automatic classification of embryo in In Vitro Fertilization

Abstract

This Master thesis addresses the challenge of embryo selection in IVF, aiming to improve pregnancy outcomes through more objective methods. It proposes an intelligent system for automatically classifying human embryo developmental stages using time-lapse imaging. The study compares 2D and 3D CNNs to assess spatial information and explores temporal models like TimeSformer to capture embryo dynamics. A hierarchical classification strategy is also introduced to handle 15 developmental phases. Results show that while 2D CNNs provide a solid baseline, temporal models significantly outperform them by leveraging morphokinetic features, highlighting the potential of AI to enhance accuracy and consistency in embryo selection.