Related Experiment Video
Updated: Aug 1, 2025
05:52
Drosophila Embryo Preparation and Microinjection for Live Cell Microscopy Performed using an Automated High Content Analyzer
Published on: January 19, 2021
8.7K
Deep learning for embryo evaluation using time-lapse: a systematic review of diagnostic test accuracy
Aya Berman1, Roi Anteby2, Orly Efros3
1Faculty of Health Sciences, Ben-Gurion University of the Negev, Be'er Sheva, Israel.
American Journal of Obstetrics and Gynecology
|April 28, 2023
Summary
Convolutional neural network models show high accuracy in embryo assessment using time-lapse monitoring, often outperforming embryologists. Further research should standardize reporting for clinical application.
Area of Science:
- Reproductive medicine and artificial intelligence
- Embryology and computational analysis
- In vitro fertilization and predictive modeling
Background:
- Time-lapse monitoring (TLM) is increasingly used in in vitro fertilization (IVF) for embryo assessment.
- Artificial intelligence (AI), specifically convolutional neural networks (CNNs), offers potential for objective and efficient embryo evaluation.
- The accuracy of CNN models in TLM-based embryo assessment requires systematic investigation.
Purpose of the Study:
- To systematically review and assess the accuracy of CNN models for embryo evaluation using TLM.
- To identify key outcomes evaluated by CNN models in assisted reproductive technology.
- To evaluate the risk of bias and applicability of included studies.
Main Methods:
- Systematic literature search of PubMed and Web of Science (January 2016-December 2022).
- Inclusion criteria focused on studies reporting CNN accuracy for embryo evaluation via TLM.
- Risk of bias and applicability assessed using QUADAS-2 and JBI checklists.
Main Results:
- Twenty-two retrospective studies involving over 522,000 images of 222,000 embryos were included.
- CNN models demonstrated high accuracy (>80%) in blastocyst stage classification and quality assessment.
- Some CNN models outperformed human embryologists, with models predicting live birth showing lower bias and external validation.
Conclusions:
- CNNs applied to TLM offer a promising avenue for more accurate, efficient, and objective embryo assessment in IVF.
- Models for blastocyst stage classification demonstrated the strongest predictive capabilities.
- Standardization of reporting and data sharing are crucial for advancing clinical application of AI in reproductive medicine.
