An Automated Blastocyst Grading System Using Convolutional Neural Network and Transfer Learning

In vitro fertilization (IVF) involves collecting several mature egg samples fertilized with sperm in the IVF laboratory through a process of manual grading and selecting the healthiest blastocyst embryo through visual morphology assessment. Nevertheless, this manual process is highly subjective, prone to error and human bias, time-consuming, and can result in multiple pregnancies. Since the grading system for blastocyst embryos remains inconsistent, the successful rate of IVF remains low for achieving a healthy pregnancy. This study proposes an automated system for grading the quality of blastocyst embryos to enhance IVF selection methods using convolutional neural networks (CNN) and VGG-16 with new classification layers. The CNN model has achieved training and validation accuracy of 97% and 95%, respectively, while in the testing stage with accuracy of 90% of average precision, recall, and F1-score of 91.3%, 90%, and 90%, respectively. Additionally, VGG16 with the new classification layers achieved impressive training and validation accuracy of 99.4% and 99.5% respectively, with outstanding testing accuracy of 94% and average precision, recall, and F1-score of 94.5%, 94%, and 94%, respectively. The proposed automated system is helpful for embryologists in providing consistent blastocyst grading and selection methods in IVF process.