Computer vision for automatic identification of blastocyst structures and blastocyst formation time in In-Vitro Fertilization.

Embryo selection is an indispensable step to ensure the success of In-Vitro Fertilization; however, this decision is a time-consuming, laborious, and highly subjective task for embryologists. In the best scenarios, when implanting an embryo of the best quality, the probability of pregnancy rate is just 34.1%. Automatic segmentation provides detailed, quantitative, and objective information about the embryo, reducing the workload of embryologists and enhancing the success rate of embryo implantation. As such, it represents a valuable resource in the field of assisted reproduction. Towards that aim, we present different computer vision methods that are able to automatically segment the different structures of the blastocyst - the first morphologically differentiated state of the embryo - with a Dice Score up to 0.89. Furthermore, our methods can identify the precise moment of the expanded blastocyst formation with a mean error of less than 4 h. We openly release the code so that anyone can use it and replicate the results. As a summary, this work aims to make the analysis of blastocysts more reliable and comparable, thereby advancing our understanding of embryo implantation.