The current morphologic embryo grading is highly subjective grading. This leads to not only a weak disposition decision making but also reduces overall efficacy outcome of selection in an embryo laboratory. In this study, we propose a system based on the concepts of artificial intelligence to improve the consistency of morphologic embryo grading and compare our results to that of the manual grading. We see an improvement in the decision-making process and in conclusion, endorse the scope of the study.
To evaluate whether an artificial intelligence (AI) network could improve the consistency of morphologic embryo grading at the blastocyst stage and aid embryologists in embryo disposition decision (DD) making.
MATERIALS AND METHODS:
A dataset comprising 3,469 embryos was used to train and test a deep convolutional neural network (CNN) model to primarily classify between non-blastocysts and blastocysts using images of embryos captured at 113 hours post insemination (hpi). Using a blinded 742 images, grading tendencies and coefficient of variation (%CV) of 7 embryologists qualitatively classifying day 5 blastocysts on a 5-grade system was evaluated. We used a blinded 56 images to evaluate the DD of 10 embryologists after rotating the embryo image 90 and 180 degrees. Consistency is the percentage of cases where the disposition decision was unaffected by the rotation. We compared the assessments of the embryologists’ with CNN.
When qualitatively classifying day 5 blastocysts into a 5-grade system, embryologists exhibited a high degree of variability. When selecting day 5 blastocysts for biopsy or cryopreservation, embryologists had an average consistency of 52.14% and 57.68% respectively. CNN outperformed the embryologists with a consistency of 83.95% and 83.92% (P<0.05 for both), respectively.