Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Early prediction of the transferability of bovine embryos from videomicroscopy

Yasmine Hachani, Patrick Bouthemy, Elisa Fromont, Sylvie Ruffini, Ludivine Laffont, Alline de Paula Reis

TL;DR

This work tackles the problem of predicting bovine embryo transferability within four days from 2D time-lapse videos. It introduces SFR, a Slow/Fast/Regular three-pathway 3D CNN designed to capture multi-scale temporal dynamics and appearance cues, trained with focal loss to address class imbalance. Across a dataset of 947 videos, SFR achieves state-of-the-art accuracy (up to 75.6% with $\gamma=2$) and demonstrates stability across metrics, outperforming 2D CNN+GRU, 3D-ResNet18, and SlowFast baselines. The study also shows potential for earlier predictions by analyzing shorter video sequences, with significant implications for embryo sorting and cattle breeding by reducing unnecessary inseminations.

Abstract

Videomicroscopy is a promising tool combined with machine learning for studying the early development of in vitro fertilized bovine embryos and assessing its transferability as soon as possible. We aim to predict the embryo transferability within four days at most, taking 2D time-lapse microscopy videos as input. We formulate this problem as a supervised binary classification problem for the classes transferable and not transferable. The challenges are three-fold: 1) poorly discriminating appearance and motion, 2) class ambiguity, 3) small amount of annotated data. We propose a 3D convolutional neural network involving three pathways, which makes it multi-scale in time and able to handle appearance and motion in different ways. For training, we retain the focal loss. Our model, named SFR, compares favorably to other methods. Experiments demonstrate its effectiveness and accuracy for our challenging biological task.

Early prediction of the transferability of bovine embryos from videomicroscopy

TL;DR

This work tackles the problem of predicting bovine embryo transferability within four days from 2D time-lapse videos. It introduces SFR, a Slow/Fast/Regular three-pathway 3D CNN designed to capture multi-scale temporal dynamics and appearance cues, trained with focal loss to address class imbalance. Across a dataset of 947 videos, SFR achieves state-of-the-art accuracy (up to 75.6% with ) and demonstrates stability across metrics, outperforming 2D CNN+GRU, 3D-ResNet18, and SlowFast baselines. The study also shows potential for earlier predictions by analyzing shorter video sequences, with significant implications for embryo sorting and cattle breeding by reducing unnecessary inseminations.

Abstract

Videomicroscopy is a promising tool combined with machine learning for studying the early development of in vitro fertilized bovine embryos and assessing its transferability as soon as possible. We aim to predict the embryo transferability within four days at most, taking 2D time-lapse microscopy videos as input. We formulate this problem as a supervised binary classification problem for the classes transferable and not transferable. The challenges are three-fold: 1) poorly discriminating appearance and motion, 2) class ambiguity, 3) small amount of annotated data. We propose a 3D convolutional neural network involving three pathways, which makes it multi-scale in time and able to handle appearance and motion in different ways. For training, we retain the focal loss. Our model, named SFR, compares favorably to other methods. Experiments demonstrate its effectiveness and accuracy for our challenging biological task.
Paper Structure (15 sections, 2 equations, 4 figures, 4 tables)

This paper contains 15 sections, 2 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Model description
  4. Network architecture
  5. Loss function
  6. Data augmentation
  7. Experimental results
  8. Video capture and video dataset
  9. Implementation details
  10. Ablation study
  11. Comparative experiments
  12. Earlier prediction
  13. Conclusion
  14. Compliance with Ethical Standards
  15. Acknowledgements

Figures (4)

  • Figure 1: The in vitro fertilization (IVF) process for bovine embryos.
  • Figure 2: Two sample videos of IVF bovine embryos, with three images taken at distant time instants (top row: an example from class transferable; bottom row: an example from class non transferable). The bovine embryo (dark grey), surrounded by the zona pellucida, is located in a micro-well (light grey) within the Petri dish (black). The embryo occupies a small part of the image.
  • Figure 3: Our 3D model combines three pathways, Slow and Fast and Regular, all implemented with 3D ResNet18 of different configurations. Input video is given at a lower temporal rate for the Slow pathway. The three outputs are combined before providing the prediction. Our SFR model includes directed lateral connections between the pathways as drawn in the figure.
  • Figure 4: Accuracy of early prediction for our SFR model with focal loss for the following range of video length: from 120 frames (about two days 1/4 of embryo development) to 300 frames (about four days).