Deep Learning-Based Quantitative Blastocyst Assessment

Area of Science:

• Embryology
• Artificial Intelligence
• Medical Imaging

Background:

• Blastocyst morphology assessment is critical for in vitro fertilization (IVF) success.
• Current deep learning methods often provide qualitative, rather than quantitative, blastocyst evaluations.
• Ranking blastocyst quality using qualitative assessments presents significant challenges.

Purpose of the Study:

• To develop a quantitative method for evaluating blastocyst quality using deep learning.
• To improve the precision of blastocyst quality assessment for IVF procedures.
• To introduce a regression network with a soft attention mechanism for this purpose.

Main Methods:

• A regression neural network combined with a soft attention mechanism was proposed.
• The network generates a continuous score for precise blastocyst quality evaluation.
• An attention module identifies and localizes regions of interest (inner cell mass) in blastocyst images.

Main Results:

• The proposed regression network quantitatively evaluates blastocyst quality.
• The soft attention mechanism accurately localizes the inner cell mass (ICM).
• The method demonstrated superior performance compared to traditional classification-based networks.

Conclusions:

• The developed deep learning model provides a more reliable and accurate quantitative assessment of blastocyst quality.
• The visualized attention maps enhance the interpretability and trustworthiness of the network's predictions.
• This approach offers a significant advancement for selecting optimal blastocysts in IVF treatment.