Related Experiment Video
Updated: May 10, 2025
08:28
Morphometric Protocol for the Objective Assessment of Blastocyst Behavior During Vitrification and Warming Steps
Published on: February 28, 2019
8.6K
An interpretable artificial intelligence approach to differentiate between blastocysts with similar or same
Hang Liu1, Longbin Chen2, Guanqiao Shan1
1Department of Mechanical and Industrial Engineering, University of Toronto, Toronto, ON, Canada.
Human Reproduction (Oxford, England)
|April 21, 2025
Summary
A new deep-learning model, BlastScoringNet, quantifies blastocyst inner cell mass (ICM) and trophectoderm (TE) morphology, providing continuous scores that correlate with live birth rates. This tool aids embryologists in selecting viable embryos more effectively.
Area of Science:
- Reproductive Medicine and Embryology
- Artificial Intelligence in Healthcare
- In Vitro Fertilization (IVF) Technology
Background:
- The Gardner grading system, while standard, presents challenges in differentiating blastocysts with similar inner cell mass (ICM) and trophectoderm (TE) grades.
- Human assessment of blastocyst potential for live birth is subjective and inconsistent, impacting clinical decision-making.
- There is a need for a quantitative, objective method to assess blastocyst quality and predict live birth outcomes.
Purpose of the Study:
- To develop a quantitative, interpretable deep-learning model for differentiating blastocysts with similar ICM and TE grades.
- To create a method that accurately reflects blastocyst potential for live birth.
- To enhance the objectivity and consistency of blastocyst assessment in IVF.
Main Methods:
- Developed BlastScoringNet, a deep-learning model, using a dataset of 2760 blastocysts with Gardner grades.
- Applied the model to a live birth dataset of 15,228 blastocysts to generate continuous ICM and TE scores.
- Validated the model's generalizability on an external dataset of 1455 blastocysts and 476 live birth outcomes.
Main Results:
- BlastScoringNet achieved high accuracy in grading blastocyst expansion degree, ICM, and TE morphology (AUCs: 0.997, 0.903, 0.943 respectively).
- Higher continuous ICM and TE scores generated by BlastScoringNet significantly correlated with increased live birth rates (P < 0.0001).
- The model demonstrated consistent performance and correlation with live birth rates across two independent institutions.
Conclusions:
- BlastScoringNet offers an interpretable, quantitative method for blastocyst evaluation, complementing the Gardner grading system.
- The model's continuous scores for ICM and TE morphology are significant predictors of live birth potential.
- BlastScoringNet shows clinical utility and generalizability, potentially improving blastocyst selection in IVF.
