Human Embryo Quality Assessment with Deep Learning Models.

Area of Science:

• Assisted Reproductive Technology (ART)
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Embryo quality assessment is crucial for successful assisted reproductive technology (ART) outcomes.
• Subjective visual grading of embryos by embryologists can lead to inconsistencies.
• Deep learning offers a path toward objective and reproducible embryo evaluation.

Purpose of the Study:

• To investigate the efficacy of deep learning models for classifying embryo quality.
• To compare the performance of various convolutional neural network (CNN) architectures.
• To identify the optimal model for accurate embryo assessment at day-3 and day-5 stages.

Main Methods:

• Utilized a dataset of embryo images from Hung Vuong Hospital.
• Trained and evaluated four CNN architectures: VGG-19, ResNet-50, InceptionV3, and EfficientNetV2.
• Assessed model performance using accuracy, precision, and recall metrics.

Main Results:

• EfficientNetV2 achieved the highest performance among the tested models.
• EfficientNetV2 demonstrated 95.26% accuracy, 96.30% precision, and 97.25% recall.
• Deep learning models, especially EfficientNetV2, show potential for consistent embryo quality assessments.

Conclusions:

• EfficientNetV2 is a highly accurate tool for objective embryo quality assessment.
• This AI-driven approach can enhance fertility treatment efficiency.
• Objective embryo evaluation supports specialists and prospective parents in the reproductive journey.