Related Experiment Video
Updated: Aug 19, 2025
07:02
Live Imaging of Early Cardiac Progenitors in the Mouse Embryo
Published on: July 12, 2022
1.5K
An explainable deep learning-based algorithm with an attention mechanism for predicting the live birth potential of
Yuta Tokuoka1, Takahiro G Yamada2, Daisuke Mashiko3
1Center for Biosciences and Informatics, Graduate School of Fundamental Science and Technology, Keio University, 3-14-1 Hiyoshi, Kouhoku-ku, Yokohama, 223-8522, Japan.
Artificial Intelligence in Medicine
|December 3, 2022
Summary
A new AI model, the Normalized Multi-View Attention Network (NVAN), accurately predicts live birth potential in embryos using nuclear morphology. This advanced technique improves upon current methods for selecting embryos in assisted reproductive technology (ART).
Area of Science:
- Developmental Biology
- Artificial Intelligence in Medicine
- Assisted Reproductive Technology (ART)
Background:
- Current assisted reproductive technology (ART) embryo grading relies on limited morphological assessments, leading to low live birth rates.
- Existing methods overlook crucial dynamic changes in nuclear structures during early embryogenesis.
- There is a need for more accurate methods to predict embryo viability for successful in vitro fertilization (IVF).
Purpose of the Study:
- To develop an advanced AI model for predicting live birth potential directly from mouse embryo nuclear structures.
- To analyze the contribution of specific nuclear morphological features to live birth prediction accuracy.
- To enhance embryo selection strategies in ART and developmental engineering.
Main Methods:
- Development of a Normalized Multi-View Attention Network (NVAN) utilizing live-cell fluorescence imaging of mouse embryos.
- Extraction of nuclear morphological features as multivariate time-series data using a specialized segmentation algorithm.
- Training and validation of the NVAN model for predicting live birth potential.
Main Results:
- The NVAN model achieved a classification accuracy of 83.87%, significantly outperforming existing machine learning methods and human expert evaluation.
- An attention mechanism identified key features influencing prediction, notably nuclear size and shape at the morula stage and during cell division.
- The model demonstrated the importance of dynamic nuclear morphology over static assessments for predicting embryo viability.
Conclusions:
- The NVAN model offers a novel, data-driven approach to accurately predict embryo live birth potential.
- This technology has the potential to significantly improve IVF success rates by optimizing embryo selection.
- NVAN serves as a foundational technology for advancing ART and developmental engineering applications.
