Microscopic Video-Based Grouped Embryo Segmentation: A Deep Learning Approach.

Area of Science:

• Reproductive Medicine
• Artificial Intelligence
• Biotechnology

Background:

• In vitro fertilization (IVF) relies on accurate embryo assessment.
• Current methods for embryo analysis can be labor-intensive and subjective.
• Advanced computational techniques can potentially improve IVF efficiency and outcomes.

Purpose of the Study:

• To develop and validate a deep learning (DL) approach for precise embryo segmentation and organization in IVF.
• To establish a comprehensive embryo database for research and educational applications.
• To enhance the utilization of DL in microscopic embryo analysis.

Main Methods:

• An advanced DL methodology was developed, involving an Embryo Segmentation Model, Segmented Embryo Image Organization, and Clear and Blur Image Classification.
• The approach was evaluated using 5182 embryos from 362 microscopic embryo videos.
• Key performance metrics included mean average precision (mAP) at various intersection over union (IoU) thresholds.

Main Results:

• The proposed DL method achieved high accuracy in segmenting and organizing embryo images, with mAP of 1.0 at IoU thresholds of 0.5 to 0.95.
• Image segmentation based on developmental day, patient, growth medium, and embryo type facilitates comparative analysis and issue identification.
• Optimal threshold values for classifying clear and blurry images were determined.

Conclusions:

• The developed technique is crucial for data preparation in IVF training and education.
• This study lays a foundation for increased adoption of DL in IVF.
• The findings are expected to positively impact IVF success rates through improved data analysis and insights.