Related Experiment Video
Updated: Jun 29, 2026
10:04
A Neural Network-Based Identification of Developmentally Competent or Incompetent Mouse Fully-Grown Oocytes
Published on: March 3, 2018
6.6K
Application of a methodological framework for the development and multicenter validation of reliable artificial
D Gilboa1, Akhil Garg2, M Shapiro1
1AIVF Ltd, Tel Aviv, Israel.
Reproductive Biology and Endocrinology : RB&E
|January 31, 2025
Summary
This study developed a reliable artificial intelligence (AI) model for evaluating embryos in in vitro fertilization (IVF) using a novel methodology. The AI accurately predicts pregnancy likelihood, enhancing embryo selection for better outcomes.
Area of Science:
- Reproductive medicine
- Artificial intelligence in healthcare
- Embryology
Background:
- Artificial intelligence (AI) models analyze embryo time-lapse images to predict pregnancy likelihood in in vitro fertilization (IVF).
- Ensuring AI consistency and reliability during development and validation in clinical settings remains under-researched.
- A methodology for developing and validating AI models across multiple datasets is presented to demonstrate reliable blastocyst-stage embryo evaluation.
Purpose of the Study:
- To present a robust methodology for developing and validating an AI model for embryo evaluation.
- To demonstrate the AI model's reliable performance across diverse datasets.
- To ensure the AI model's interpretability and explainability in predicting pregnancy probability.
Main Methods:
- A multicenter analysis used time-lapse images, pregnancy outcomes, and morphologic annotations from 10 IVF clinics across 9 countries (2018-2022).
- A four-step methodology involved dataset curation, AI model development/optimization, performance evaluation (discriminative power, pregnancy association), and interpretability analysis.
- Three datasets were used: training/validation (16,935 embryos), blind test (1,708 embryos; 3 clinics), and independent test (7,445 embryos; 7 clinics).
Main Results:
- The AI deep learning classifier ranked embryos by predicted clinical pregnancy likelihood.
- Higher AI scores correlated positively with fetal heartbeat (FH) likelihood across all datasets (ORs ranging from 0.40 to 4.01).
- AI scores showed consistent increases with FH likelihood and correlated with key embryo quality morphologic parameters.
Conclusions:
- The AI model demonstrated strong performance across multiple datasets.
- The presented four-step methodology is valuable for developing and validating AI tools.
- The AI serves as a reliable adjunct for embryo evaluation in clinical practice.
