FULL AUTOMATION OF EMBRYO EVALUATION MODELS BENEFITS FROM TRAINING ON BOTH TRANSFERRED AND DISCARDED EMBRYOS

Is overall performance of an AI-based embryo evaluation model improved by including discarded embryos during training compared to using only transferred with known implantation data (KID)? A set 14,643 KID fetal heart (4,337 FH+ and 10,307 FH-) incubated for at least 4 days was obtained from 18 different clinics. The also included 23,674 either cryopreserved or unknown FH 101,153 that were thus not transferred. Two 3D convolutional deep learning models trained on 85% the set. first (KID-o) embryos. second (KID-d) both embryos, pseudo-labelled as FH-. KID-d used oversampling ensure equal distribution FH- training. evaluated internal validation based remaining 15% data. In addition, external 1,125 6,327 data, 9,728 American Hospital, Turkey. main goal is rank determine order transfer. This task calculating area under curve (AUC) Evaluation solely clinical outcome. categorize into useable (i.e. cryopreserved) finding threshold where 95% classified useable. At this sensitivity threshold, specificity calculated. Thus, concerned if related ranking AUCs two nearly identical. KID-o model, respectively, 0.667 0.670 validation, 0.715 0.714 validation. However, specificities categorization significantly (p<0.001 – McNemar’s test) higher than model. 0.48 0.82 0.50 0.84 results showed pre-selected negatively affected in categorizing performed better inclusion overall, best observed