Follicle Identification in Primate Ovaries Via Machine Learning†

Area of Science:

• Reproductive biology
• Computational biology
• Veterinary medicine

Background:

• Ovarian reserve estimation is crucial for understanding ovarian aging and reproductive health.
• Accurate follicle counting is essential for studies on fertility, drug effects, and environmental toxicants.
• Nonhuman primates serve as valuable translational models for human reproductive studies.

Purpose of the Study:

• To develop and validate a machine learning pipeline for classifying preantral ovarian follicles in histology images.
• To enable automated and accurate assessment of ovarian reserve in nonhuman primate models.
• To provide a tool for research on ovarian aging, fertility preservation, and menopause.

Main Methods:

• A ResNet34 convolutional neural network was retrained using a transfer learning approach.
• 7770 preantral follicles from 11 nonhuman primates were manually annotated in histology images.
• The model was trained and validated on classified follicle images, achieving high accuracy.

Main Results:

• The machine learning model achieved >98% accuracy during training and 75.3% on the validation set.
• The model demonstrated good performance on the test set (72.4%), with errors primarily between adjacent follicle stages.
• The results suggest the model learned an implicit developmental ordering of ovarian follicles.

Conclusions:

• The developed machine learning pipeline accurately classifies preantral ovarian follicles in histology images.
• This tool facilitates the study of ovarian aging, follicle depletion, and drug effects in primate models.
• The publicly available model and data support research in reproductive health, fertility preservation, and menopause delay.