Quantitative image assessment of embryos to predict pregnancy in embryo transfer programs

Summary

Female-focussed reproductive technologies have modest adoption in the Australian cattle industry. In part, this is caused by low and /or variable pregnancy rates.  Here we assessed the predictive capacity for pregnancy establishment from microscopic imaging of routinely selected transferable cattle embryos prior to transfer. This incorporated analysis of features built from custom-algorithms.  A total of 476 multiple ovulation and embryo transfer (MOET) embryos (266 Pregnant [P] and 100 Non pregnant [NP]) at Day 7 and juvenile in vitro fertilised embryo transfer (JIVET) embryos (same stage: 62 P, 63 NP) were imaged and analysed.

Mid-section and bottom images of embryos were collected under white, or red, blue and green filtered differential interference microscopy. Images were corrected to ensure consistency between different days of collection. Supervised analysis specifically targeting P/NP outcomes were conducted.  At present, the probability of an accurate prediction for MOET embryos is 88% and for JIVET embryos is 96%.  A probability of over 95% is considered a highly predictive assay.  

For images from JIVET embryos, a cluster in an unsupervised analysis was highly predictive of pregnancy failure, suggesting that there are common features in some JIVET embryos that will not form a pregnancy.

Cross-validation and further images taken across breeds, management and regions is now required.  With further data, we are confident that an accurate image-based assessment for pregnancy for each embryo is achievable.  Direct value to breeders is estimated at an additional $30,000 for every 1000 embryos transferred with a modest 5% pregnancy rate improvement.