Predicting Live birth from a transferred embryo using specific time-points across development. All ROC AUC scores shown here are the average ROC AUC scores on the test set after 50 training iterations each with a different randomly allocated train/validation/test split with 25 embryos in the successful class and 48 embryos in the unsuccessful class for both the test and validation set. The accompanying error bars are the standard error across these 50 training attempts. (A) The methodology for developing the live birth prediction models. (B) The ROC AUC scores for models trained to predict live birth using images from precisely defined moments of development. ‘PN’ is 1 h before completion of NEBD, ‘2 cell’ is first frame with 2 cells, ‘4 cell’ is first frame with 4 cells, ‘8–16’ cell is the first fourth cleavage event, and ‘blastocyst’ is the last frame of the video. (C) The ROC AUC scores for models trained to predict live birth using images at various time intervals before and after NEBD. (D) The ROC AUC scores for models trained to predict live birth using images at various time intervals after first division (FD). (E) The ROC AUC scores for models trained to predict live birth using images at various time intervals after the first frame with 4 cells. (F) The ROC AUC scores of the test set for models trained to predict live birth using images at various time intervals after the first fourth cleavage event. (G) The ROC AUC scores for models trained to predict transfer live birth using images at various time intervals before the last frame of the time-lapse video.