Embryonic Quality Assessment using Advanced Deep Learning Architectures utilizing Microscopic Images of Blastocysts.

Area of Science:

• Embryology
• Artificial Intelligence in Medicine
• Assisted Reproductive Technology

Background:

• Accurate embryonic quality evaluation is crucial for successful in vitro fertilization (IVF).
• Current morphological scoring methods have limitations in accuracy and latency.
• Advanced deep learning offers potential for enhanced embryo assessment.

Purpose of the Study:

• To develop and compare deep learning models for assessing microscopic embryo images.
• To improve the accuracy and efficiency of embryonic quality evaluation at critical developmental stages (Day 3 and Day 5).
• To investigate the performance of Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT) for this task.

Main Methods:

• Microscopic embryo images were pre-processed using histogram equalization.
• Various deep learning models, including GCN and GAT, were trained and evaluated.
• Performance metrics such as accuracy, sensitivity, and specificity were compared.

Main Results:

• Graph Convolutional Networks (GCN) achieved 96.1% accuracy.
• Graph Attention Networks (GAT) further improved accuracy to 98.8%, with 96.4% sensitivity and 97.5% specificity.
• The attention mechanism in GAT dynamically weighted feature importance for enhanced performance.

Conclusions:

• Deep learning, particularly GAT, offers a highly accurate and efficient method for embryonic quality assessment.
• This AI-driven approach can significantly aid in improving IVF success rates by optimizing embryo selection.
• The study highlights the potential of advanced AI in revolutionizing assisted reproductive technologies.