Fig. 5: Performance on Segmentation, Witnessing, and Blastulation Time Prediction.
From: A foundational model for in vitro fertilization trained on 18 million time-lapse images
Performances on ES (Embryoscope) and ES+ (Embryoscope + ) microscopes for each task. Performances are aggregated across 4 replicates (four-fold cross validation) (n = 4). Asterisks represent statistical significance with p < 0.05, where statistical significance is determined by performing a one-way ANOVA test followed by a Tukey HSD test. a Segmentation performance for each embryo component via Dice Score. Inner Cell Mass, p = 0.53; Zona Pellucida, p = 0.26; Trophectoderm, p = 0.06. b Embryo witnessing performance via F1 score across all datasets. Weill ES, p = 0.99; 2020 Weill ES + , p = 8.9e-22; Spain ES + , p = 1.7e-10; Florida ES + , p = 1.9e-25; 2021+ Weill ES + , p = 1.7e-25. c Blastulation time prediction performance using mean absolute error for Weill ES, Spain ES + , and 2020 Weill ES+ datasets. Weill ES, p = 0.81; 2020 Weill ES + , p = 1.3e-39; Spain ES + , p = 1.8e-30. For all subplots, error bars show mean values +/- SEM. Source data are provided as a Source Data file.
