229 Machine learning identifies differences in morphokinetics of in vivo-derived bovine embryos between hot and cool seasons

Heat-stressed donors have decreased viable blastocyst production and lower multiple ovulation embryo transfer (MOET) pregnancy outcomes when compared with non-heat-stressed contemporaries. Many of these embryonic changes are nonidentifiable to the human eye after considerable magnification, highlighting the need to find alternative solutions to identify compromised embryos. Thus, the objective of this study was to use machine learning techniques to detect morphokinetic activity of embryos from donors undergoing MOET based on seasonal exposure to heat stress. A preliminary study was performed on 237 bovine embryos (n = 237) from cross-bred Bos taurus primiparous and multiparous beef donors that underwent MOET in the Texas panhandle during routine practice field conditions. All embryos were staged and quality graded according to IETS standards at 90× magnification. The average external temperature was 32.8°C during warm season (June to August) and 12.2°C during cool season (November to February) collection periods. A 30-s video was recorded of all embryos using standard video equipment and microscopy was done in a controlled laboratory at 20°C. Each 30-s video was recorded at 30 frames per second, leading to a total of 900 frames for analysis (30 fps × 30 s video = 900 total frames). Computer vision object recognition identified individual embryos, followed by an image subtraction process that identified pixel changes from one frame to the next. The subtracted frames provided the absolute difference between pixel values, and delivered a quantifiable, objective measurement of embryo morphokinetics. Data were analysed using a t-test with a P-value of 0.05 for significance. Results from the analysis are presented as normalized activity (standard deviation/mean) pixel change per embryo. Preliminary data suggest that there is evidence for machine learning models to identify differences in real-time morphokinetics on in vivo-derived Bos taurus embryos, regardless of stage and quality grade, whereby embryos collected from donors during the hot season had greater real-time morphokinetic activity (2.46 ± 0.54), when compared with those collected during the cool season (1.98 ± 0.63; P < 0.05). It is documented that heat-stressed embryos have an increased metabolism due to an attempt to repair denature proteins and damaged DNA, and from this, it is hypothesised that machine learning is depicting this change in embryo metabolism. All embryos from this study were morphologically graded acceptable to transfer, yet the damages due to heat stress, combined with the increased metabolic activity, suggest embryos from donors collected with seasonal exposure to heat stress might have lower pregnancy outcomes when transferred. Future work aims to advance results by collecting data on pregnancy outcomes to improve reproductive performance of donors undergoing embryo transfer in at-risk environmental conditions.