#37 : Improved Pregnancy Prediction Performance in an Updated Deep-Learning-Based Embryo Selection Model: A Retrospective Independent Validation Study Using 3,960 Single Blastocyst Transfer Cycles

Background and Aims: Generally, the performance of deep learning models is improved by increasing the size of the training dataset. However, to the best of our knowledge, there has been no study on the effect of increasing training data on the performance of pregnancy prediction using the same deep-learning model.

Method: A total of 3,960 blastocyst transfer cycles (1 patient, 1 cycle) were retrospectively analyzed. Embryos were stratified according to SART age groups. The quality and scoring of embryos were assessed by iDAScore v2.0 (iDA-V2; Vitrolife, Sweden), v1.0 (iDA-V1; Vitrolife, Sweden), and Gardner grading (GG). The discriminative performance of pregnancy prediction for each embryo scoring model was compared using the area under the curve (AUC) of the receiver operating characteristic curve for each maternal age group.

Results: In the AUCs of the < 35 years age group (n = 757) for pregnancy prediction, the AUC of iDAV2 was significantly higher than that of GG (0.718 vs. 0.694, p = 0.015). For the 35-37 years age group (n = 821), the AUC of iDA-V2 was significantly higher than that of iDA-V1 (0.712 vs. 0.696, p = 0.035). In the AUCs of the 41–42 years age group (n = 715), the AUC of iDA-V2 was significantly different from that of GG (0.745 vs. 0.696, P = 0.007). In the > 42 years age group (n = 660) and 38–40 years age group (n = 1,007), there were no significant difference among the groups.

Conclusion: The present study suggests that the performance of deep learning models for pregnancy prediction will be improved by increasing the size of the training data. Therefore, in a deep-learning model for pregnancy prediction, continuously collecting training data is an important step for improving the performance of updated AI models.