Automatic grading of human blastocysts from time-lapse imaging

Area of Science:

• Embryology
• Artificial Intelligence
• Reproductive Medicine

Background:

• Blastocyst morphology is crucial for predicting in vitro fertilization (IVF) implantation success.
• Manual embryo grading by embryologists is subjective, leading to inter- and intra-observer variability.
• Current grading systems have unclear boundaries between categories, impacting embryo selection.

Purpose of the Study:

• To develop a deep learning method for automated grading of human blastocyst morphology using time-lapse imaging.
• To improve the objectivity and accuracy of embryo quality assessment in IVF.

Main Methods:

• A convolutional neural network (CNN) was trained to predict inner cell mass (ICM) and trophectoderm (TE) grades from single image frames.
• A recurrent neural network (RNN) was employed to integrate temporal information from multiple frames of blastocyst expansion.
• The model processed time-lapse imaging data for automated morphological assessment.

Main Results:

• The deep learning method achieved accuracy surpassing human embryologists on an independent test set.
• The correlation between predicted embryo quality and pregnancy outcome was comparable to that of human experts.
• The AI model demonstrated superior performance in predicting ICM and TE grades compared to average human assessment.

Conclusions:

• Deep learning utilizing time-lapse imaging significantly enhances human blastocyst grading accuracy.
• The automated method is at least on par with, and in some aspects superior to, human embryologists in quality estimation.
• This AI-driven approach offers a more objective and potentially more accurate tool for selecting viable embryos in IVF procedures.