An interpretable and versatile machine learning approach for oocyte phenotyping

Area of Science:

• Reproductive Biology
• Computational Biology
• Biotechnology

Background:

• Oocyte meiotic maturation is vital for fertilization and embryo development, impacting fundamental research and assisted reproductive technologies.
• Existing computational tools for characterizing oocyte maturation using non-invasive measurements are limited.
• Developing objective, non-invasive methods to assess oocyte quality is crucial for improving reproductive outcomes.

Purpose of the Study:

• To develop and validate a computational framework for phenotyping oocytes using transmitted light imaging.
• To create a machine learning pipeline for recognizing oocyte populations and identifying morphological differences.
• To assess the potential of this framework in predicting oocyte maturation and developmental potential.

Main Methods:

• Development of a computational framework using neural networks for oocyte and zona pellucida segmentation from transmitted light images.
• Definition and extraction of a comprehensive set of morphological features describing oocytes.
• Implementation of a feature-based machine learning pipeline within an open-source Fiji plugin.
• Application of the pipeline for screening oocytes, identifying morphological characteristics, and predicting maturation potential.

Main Results:

• Successfully trained neural networks to segment oocytes and zona pellucida across diverse species.
• Identified key morphological features, including zona pellucida texture and cytoplasmic particle size, for assessing mouse oocyte maturation potential.
• Demonstrated the framework's ability to screen different oocyte strains and automatically characterize their morphology.
• Validated the applicability of identified features for assessing the developmental potential of human oocytes.

Conclusions:

• The developed computational framework provides a novel, non-invasive method for characterizing oocyte meiotic maturation.
• The feature-based machine learning pipeline effectively identifies morphological differences and predicts oocyte maturation and developmental potential.
• This tool has significant implications for both fundamental research in reproductive biology and clinical applications in assisted reproductive technology.