A machine learning model for predicting fertilization following short-term insemination using embryo images

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Embryology

Background:

• Early prediction of fertilization is crucial for timely intervention in assisted reproductive technologies.
• Manual assessment of fertilization by embryologists can be subjective and time-consuming.
• Machine learning offers a potential solution for objective and efficient embryo assessment.

Purpose of the Study:

• To develop and evaluate a machine learning model (MLM) for predicting fertilization post-insemination using embryo images.
• To compare the predictive performance of the MLM against manual classifications by embryologists.
• To assess the utility of the MLM for early rescue Intracytoplasmic Sperm Injection (ICSI).

Main Methods:

• Embryo images were acquired at 4.5 and 8 hours post-insemination.
• ResNet50 was used for image preprocessing and vector extraction.
• Light Gradient Boosting Machine (Light GBM) was employed for model training.
• Fertilization prediction was assessed by the MLM and compared with senior and junior embryologists' classifications.

Main Results:

• The MLM achieved an accuracy of 0.71 ± 0.01, outperforming junior embryologists (0.61 ± 0.05).
• No significant difference was found between the MLM and senior embryologists (accuracy 0.75 ± 0.05).
• Similar trends were observed for recall, F1-score, and area under the curve, with the MLM performing comparably to senior embryologists.

Conclusions:

• The developed MLM effectively predicts fertilization by analyzing subtle cytoplasmic changes in embryo images.
• The model demonstrates potential to enhance clinical decision-making in early rescue ICSI procedures.
• Integrating AI tools like this MLM can lead to improved patient outcomes in fertility treatments.