Advancing Bovine Embryo Evaluation: Machine Learning to Assess Embryos in Routine Embryo Transfer Practice

Introduction Machine learning (ML) enables rapid, high-precision data processing and holds promise for outperforming human evaluators in livestock embryo transfer (ET), ultimately accelerating genetic progress and improving economic outcomes. Therefore, this study had two primary objectives: (1) to evaluate ML performance in a real-world field trial, and (2) to survey embryologists on their traditional assessments of bovine embryos and compare their evaluations to ML-generated results. Materials and Methods 1. 1) A dataset of 6,900 30s videos of bovine embryos were recorded during routine ET with a smartphone mounted to a microscope and morphology was evaluated. 2. 2) 42 bovine embryologists were surveyed to evaluate ten embryo images. Responses were compared to ML predictions. 3. 3) 573 embryos compared ML stage and grade predictions compared to embryologists’ results. 4. 4) Kruskal–Wallis tests with Bonferroni corrections were used to assess differences in embryo assessments across groups, and independent t-tests were applied where assumptions of normality and equal variance were met. Results Embryologist assessments showed only 59.8% agreement overall, increasing to 74.6% among those with over 5 years of experience. ML demonstrated 70% agreement with all participants and 85% with experts, showing no statistical difference compared to expert evaluations (p>0.05). ML was also proficient in identifying unfertilized oocytes, typically a skill of experienced embryologists. In the broader study, ML reached 81.7% agreement with experts on embryo stage (456/558) and 95.2% on transferability (531/558). Conclusion This study highlights new ML applications to evaluate embryos, offering the reality of an automated and standardized embryo analysis with the potential to improve status quo.