Automatic morphological human embryo assessment using convolutional neural network and resampling technique.

Morphological human embryo assessment is a critical process to assess human embryo quality for in vitro fertilization (IVF) treatment. Embryo morphology assessments are conventionally performed through manual microscopic analysis by embryologists. However, this process leads to a subjective result that depends on how they visually inspect the morphological structures of the embryo under microscopes. Deep learning can help with the problem of assessing the image of the human embryo. But the availability of human embryo image data is not always sufficient. This result causes a limited amount of data and imbalanced data. In this study, we propose a 2D Convolutional Neural Network (CNN) for embryo grading. In addition, we use Resampling methods and Generative Adversarial Networks (GAN) to overcome imbalanced dataset problems and improve embryo grading performance. We conducted a 10-fold cross-validation to measure model performance. Our proposed method achieved an F1-Score of 0.95, 0.81, and 0.95 for classifying embryo expansion, ICM quality, and Trophectoderm quality, respectively. [ABSTRACT FROM AUTHOR]