Automatic characterization of human embryos at day 4 post-insemination from time-lapse imaging using supervised...

Area of Science:

• Reproductive medicine and artificial intelligence.
• Computational embryology and machine learning.

Background:

• Embryo morphology is crucial for predicting implantation success and live births.
• Manual embryo assessment is subjective and time-consuming.
• Automated methods offer objective and accurate predictions.

Purpose of the Study:

• To develop a deep learning methodology for automated human embryo evaluation.
• To predict embryo viability and classify embryo quality using time-lapse imaging.

Main Methods:

• Implemented a supervised contrastive learning framework for viability prediction.
• Applied an inductive transfer approach for embryo quality classification.
• Utilized deep learning on time-lapse imaging data for human embryos.

Main Results:

• Achieved high accuracy in predicting embryo viability (0.8103 day 4, 0.9330 day 5).
• Reached significant accuracy in classifying embryo quality (0.7500 day 4, 0.8001 day 5).
• Outperformed conventional methods and aligned with clinical interpretations.

Conclusions:

• Proposed deep learning methods show promise for clinical application in embryology.
• Demonstrated breakthrough potential for embryo selection as early as day 4.
• AI-driven insights support embryologists' decisions, enhancing objectivity.