Deep-learning model for embryo selection using time-lapse imaging of matched high-quality embryos

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Embryology

Background:

• In vitro fertilization (IVF) laboratories face challenges in selecting the most viable embryos for transfer.
• Time-lapse imaging and deep learning offer potential solutions for improving embryo selection accuracy.

Purpose of the Study:

• To develop and validate a deep learning model for predicting embryo implantation success in IVF.
• To leverage self-supervised contrastive learning for comprehensive analysis of embryo morphokinetics.

Main Methods:

• A deep learning model was developed using self-supervised contrastive learning on 1580 embryo videos from 460 patients.
• The model employed convolutional neural networks, Siamese network fine-tuning, and an XGBoost predictor.
• A novel approach utilized matched Known Implantation Data (KID) embryos with differing implantation fates from the same cohort.

Main Results:

• The model predicted subsequent embryo transfer outcomes with an AUC of 0.57 when using prior transfer data from the same cohort.
• Without prior transfer history, the model achieved an AUC of 0.64 in predicting implantation.
• The model demonstrated satisfactory performance in predicting implantation success.

Conclusions:

• The developed deep learning model can serve as an adjunct tool for embryologists in IVF.
• It aids in selecting more viable embryos, potentially reducing the number of unsuccessful transfers.
• This technology can improve IVF success rates by optimizing embryo selection processes.