Related Experiment Video
Updated: Jul 12, 2025
05:13
Author Spotlight: Advancing Therapeutic Strategies for Improving Pregnancy Rates by Analyzing Embryo-Endometrium Interactions
Published on: June 21, 2024
887
Improved pregnancy prediction performance in an updated deep-learning embryo selection model: a retrospective
Satoshi Ueno1, Jørgen Berntsen2, Tadashi Okimura1
1Kato Ladies Clinic, 7-20-3, Nishi-shinjuku, Shinjuku, Tokyo 160-0023, Japan.
Reproductive Biomedicine Online
|November 1, 2023
Summary
Increasing training data for deep learning models significantly improves pregnancy prediction after single vitrified-warmed blastocyst transfer (SVBT). Enhanced data in AI models boosts accuracy in predicting pregnancy outcomes.
Area of Science:
- Reproductive Medicine
- Artificial Intelligence in Medicine
- Embryology
Background:
- Predicting pregnancy after single vitrified-warmed blastocyst transfer (SVBT) is crucial for optimizing assisted reproductive technologies.
- Deep learning models offer potential for improving embryo assessment and pregnancy prediction accuracy.
Purpose of the Study:
- To evaluate the effect of increased training data on the performance of a deep learning model for predicting ongoing pregnancy after SVBT.
- To compare the predictive accuracy of an enhanced deep learning model (iDAScore v2.0) against its previous version (iDAScore v1.0) and traditional Gardner grading.
Main Methods:
- Retrospective analysis of 3960 SVBT cycles, with embryos stratified by maternal age groups.
- Embryo scoring using deep learning models (iDAScore v1.0 and iDAScore v2.0) and Gardner grading.
- Comparison of model performance using the area under the curve (AUC) of receiver operating characteristic curves.
Main Results:
- The enhanced deep learning model (iDA-V2) demonstrated superior performance (AUC=0.736) compared to iDA-V1 (AUC=0.720) and Gardner grading (AUC=0.702) across all cohorts (P < 0.0001).
- iDA-V2 showed significant improvements in pregnancy prediction accuracy in specific maternal age groups, including >35 years and 41-42 years.
- While improvements were noted, no significant differences were found between models in the 38-40 years and >42 years age groups.
Conclusions:
- Increasing the size of training data enhances the performance of deep learning models for pregnancy prediction in SVBT cycles.
- The iDAScore v2.0 model, trained on more data, offers improved discriminative ability for predicting pregnancy outcomes, particularly in certain maternal age demographics.
