OBJECTIVE: To evaluate the performance of an artificial intelligence-based approach, using a deep convolutional neural network (CNN) combined with a genetic algorithm (GA), in selecting top quality day 5 euploid blastocysts compared to those selected by highly trained embryologists.Selection of top-quality final-staged embryos directly impacts the success rate of in-vitro fertilization (IVF) procedures. The success rate of IVF not only increases the pregnancy rates but also helps in conserving time and cost involved. In collaboration with experienced embryologists, we propose a deep learning based system which enhances the selection procedure of top quality embryos for implantation and completion of the IVF procedure. Our system gained intuition in terms of selection of top-quality embryos and we observed that the proposed methodology shows a significant increase in the success rate in the implantation outcome of the blastocysts when compared to the embryologists’ selection.

MATERIALS AND METHODS: Using a dataset of 3,469 embryos, the CNN model was trained and tested to primarily classify images of embryos captured at 113 hours post insemination (hpi). A non-overlapping set of 97 euploid embryo images with known implantation outcomes was then used to compare the embryo predicting accuracy of 15 highly trained embryologists from multiple centers in the US to that of the CNN. Only euploid embryos that had undergone preimplantation genetic testing for aneuploidies (PGT-A) were included to remove the bias introduced by chromosomal abnormalities.

RESULTS: The CNN performed with an accuracy of 75.3% while the embryologists performed with an average accuracy of 67.4% in differentiating euploid embryos based on their implantation outcome. The CNN performed with a sensitivity and specificity of 84.2% and 62.5%, respectively. One sample t-test revealed that the CNN significantly outperformed embryologists in predicting embryo implantation of euploid embryos.