120 High variation in prediction of developmental competence of bovine oocytes based on visual examination

Abstract

The quality of an oocyte is often represented by its morphological characteristics. Stringent selection of bovine cumulus–oocyte complexes (COCs) for in vitro embryo production (IVP) is regularly based on parameters such as colour and granulation of the ooplasm, density of the surrounding cumulus cells, brightness of the zona pellucida, among other less common parameters. The link between developmental competence and morphology of the COC has been demonstrated in literature in group culture system. However, this has not been yet examined in an individual in vitro production system. This study investigates the ability of IVP researchers to predict whether an oocyte will develop into a blastocyst at Day 8 of in vitro culture in an individual production system. A set of 29 pictures of immature bovine COCs was presented in duplicates to eight bovine IVP researchers with different institutional backgrounds. The observers were asked if they would select each oocyte for further IVP processing, assuming that only the oocytes with the highest developmental potential can proceed, and thus a stringent selection is warranted. These 29 immature oocytes were matured, fertilized, and cultured individually in vitro (using conventional methods) so that their ability to develop into a blastocyst at Day 8 post-fertilization was known. The ability to reach the blastocyst stage was stated as the gold standard and compared with the answers of the observers using kappa statistics and sensitivity (Se) and specificity (Sp) tests. Kappa (κ) for interobserver agreement was κ = 0.01 (Se = 50%, Sp = 53%; positive predictive value = 14%, negative predictive value = 87%) with 50% accuracy of prediction. The intraobserver agreement was κ = 0.69, with κ = 1 referring to 100% agreement (Se = 87%, Sp = 83%), with an accuracy of 85%. The ability of trained embryologists to predict the fate of oocytes’ developmental competence was accurate for only half of the presented oocytes. Additionally, the proportion of false positive (1 − Sp) and false negative (1 − Se) predictions was equally distributed. The low positive predictive value might be due to the lower development rates in individual culture systems versus group culture systems, whereas observers are mainly trained to select oocytes for group culture. This study demonstrated the subjective nature of the oocyte selection process that is based on visual examination of COC morphology because interobserver agreement was almost nonexistent. Interpretation of the morphology within an observer remains consequent, represented by the high intraobserver agreement. Consequently, there is a need to develop a new model for the objective determination of oocyte quality for individual IVP. The implementation of machine learning algorithms could objectively enhance oocyte selection and artificial reproductive technologies in extension.