A deep learning model for predicting blastocyst formation from cleavage-stage human embryos using time-lapse images

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Embryology

Background:

• Assisted reproductive technology (ART) success rates can be improved by optimizing embryo selection.
• Transferring embryos at Day-3 offers benefits over Day-5 blastocyst transfer, including reduced laboratory exposure and improved uterine environment integration.
• Predicting blastocyst formation from early-stage embryos is crucial for efficient ART clinical workflows.

Purpose of the Study:

• To develop and validate a novel deep learning model for predicting blastocyst formation from early-stage human embryos.
• To assess the model's accuracy, sensitivity, and specificity in predicting blastocyst development at 72 hours post-insemination (HPI).
• To provide an AI-driven tool to assist embryologists in selecting viable embryos for Day-3 transfer.

Main Methods:

• A novel ResNet-GRU deep-learning model was developed.
• The model utilized time-lapse imaging data from Day 0 to Day 3 of human embryo development.
• Model performance was evaluated using validation accuracy, sensitivity, and specificity metrics.

Main Results:

• The ResNet-GRU model achieved a validation accuracy of 93% in predicting blastocyst formation from the cleavage stage.
• The model demonstrated high sensitivity (0.97) and specificity (0.77) in its predictions.
• Early prediction of blastocyst potential was achieved at 72 HPI.

Conclusions:

• The developed deep learning model accurately predicts blastocyst formation from early-stage human embryos.
• This AI tool can assist embryologists in identifying optimal embryos for Day-3 transfer, potentially improving ART outcomes.
• Implementing this model may lead to enhanced patient outcomes and increased pregnancy rates in fertility treatments.