Deep Neural Network Segmentation of Embryo Inner Cell Mass and Trophectoderm Epithelium

Embryologists commonly assess the viability of embryos for in vitro fertilization using the Gardner blastocyst grading system, which focuses on the inner cell mass (ICM) and trophectoderm epithelium (TE) regions. To overcome the difficulties in classifying these regions due to similarities in texture and appearance, this paper demonstrates the use of ResUNet-50, a deep learning neural network model. With ResUNet-50, the ICM region was segmented with a 99.8% accuracy, 98.7% precision, 98.7% recall, 98.7% Sørensen-Dice coefficient, and 98.6% Jaccard Index. ResUNet-50 also segmented the TE region with a 98.2% accuracy, 92.4% precision, 90.3% recall, 91.2% Sørensen-Dice coefficient, and 91.3% Jaccard Index.