motilitAI: A machine learning framework for automatic prediction of human sperm motility

Area of Science:

• Andrology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Sperm motility is a critical indicator of male fertility.
• Accurate and automated assessment of sperm quality is essential for clinical diagnostics.
• Existing methods for sperm analysis can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate machine learning models for the automated assessment of human sperm motility.
• To predict the percentage of progressive, non-progressive, and immotile spermatozoa using video analysis.
• To compare the performance of different regression models for sperm quality prediction.

Main Methods:

• Utilized the Visem dataset of human semen samples.
• Employed unsupervised tracking for video analysis.
• Extracted features using custom movement statistics and displacement features.
• Trained and compared neural networks and support vector regression models.

Main Results:

• A linear Support Vector Regressor with aggregated and quantized displacement features yielded the best performance.
• The developed model achieved a mean absolute error (MAE) of 7.31.
• This represents a significant improvement over the previous best submission (MAE of 8.83) in a comparative challenge.

Conclusions:

• Machine learning, particularly linear Support Vector Regression, offers a robust and accurate method for automated sperm motility assessment.
• The proposed approach enhances the efficiency and objectivity of sperm quality evaluation.
• The source code is publicly available, facilitating further research and application.