The present study aimed to identify the correlations between morphology and quantitative parameters of blastocyst components segmented on a still blastocyst image using the Blast-Net AI model and to determine which quantitative parameters would be the most predictive of implantation and live birth.DESIGN: This is a retrospective cohort study analyzing blastocyst images acquired at 116 AE 2 hours post-insemination between 2012 and 2017, a sixyear period. 415 blastocysts from 315 single or double embryo transfers from patients at mean age of 33.3 AE 3.3 years (range: 23 -38) with a receptive uterine environment were included. All blastocysts were assessed with the Gardner 1 blastocyst scoring system at image acquisition and they were later confirmed by two experienced embryologists. All images analyzed have known implantation and live birth outcomes associated.MATERIALS AND METHODS: All 415 images were analyzed with the Blast-Net automatic segmentation model as previously described. 2 Each image was segmented into the background, Zona Pellucida (ZP), Inner Cell Mass (ICM), Blastocoel, and Trophectoderm (TE). Each component was then quantitively measured using mathematical algorithms and converted into mm or mm 2 for data analysis. The Blast-Net model provides a confidence score on the segmentation of each component. 22 images (5.3%) failed to meet the 90% confidence inclusion cutoff for ICM and 2 images (0.5%) for ZP. 391 images were included for the determination of average blastocyst radius (BR), average ZP thickness, ZP area, TE area, ICM area and ratio of ZP thickness relative to BR (ZP/BR). The correlations of these parameters and morphology grading were assessed and logistical regression analysis was used to determine the Odds Ratio for each parameter in predicting implantation and live birth.RESULTS: Among ZP area, average ZP thickness, average BR, and ZP/ BR, ZP/BR has the best goodness of fit to the blastocyst expansion. The mean ZP/BR for size 2 expansion is 0.197, 95%CI [0.191, 0.203], size 3 is 0.153, 95%CI [0.149, 0.158], and size 4 is 0.080, 95%CI [0.075, 0.084], and they are significantly different (p<0.0001). ICM area and TE area are both significantly different between grade A and B (p<0.0001 and p<0.005, respectively), but not against grade C, due to the small number of grade C ICM or TE transferred. Mean ICM area for grade A ICM is 3170.4 mm 2 , 95%CI [3080.6, 3260.2] and grade B is 2560.9 mm 2 , 95%CI [2403.4, 2718.3]. Mean TE area for grade A TE is 4265.8 mm 2 , 95%CI [4171.6, 4360.0] and grade B is 4052.6 mm 2 , 95%CI [3940.5, 4164.7]. Among ZP/BR, ICM area, and TE area, only ZP/BR is indicative of implantation (OR ¼ 2.59, 95%CI [1.10, 6.10]) and live birth (OR ¼ 2.39, 95%CI [1.02, 5.61]). Blastocysts with a ZP/BR below 0.144 have significantly higher implantation rate (59% vs 48%, p<0.04) and live birth rate (53% vs 43%, p<0.05) than ones above this threshold.CONCLUSIONS: Automatic segmentation and quantification of various components of a blastocyst image using the Blast-Net model can potent...
