High-throughput visual assessment of sleep stages in mice using machine learning
Brian Geuther1, Mandy Chen1, Raymond J Galante2
1The Jackson Laboratory, Bar Harbor, ME, USA.
Sleep
|October 31, 2021
Summary
This study introduces a computer vision method to classify mouse sleep stages from video, offering a non-invasive alternative to traditional EEG/EMG analysis. This approach enables high-throughput sleep research and reduces barriers for screening genetic mutations affecting sleep.
Area of Science:
- Neuroscience
- Computational Biology
- Animal Models
Background:
- Assessing sleep substages typically requires invasive electroencephalogram/electromyogram (EEG/EMG) analysis, involving surgery and expert scoring.
- This traditional method presents significant cost and throughput limitations for large-scale studies, particularly in preclinical research.
Purpose of the Study:
- To develop and validate a non-invasive method for classifying sleep substages (wake, NREM, REM) in mice using computer vision on video data.
- To bypass the need for surgical implantation of electrodes and expert EEG/EMG scoring.
- To enable high-throughput, cost-effective sleep studies in mouse models.
Main Methods:
- Collected synchronized high-resolution video and EEG/EMG data from 16 male C57BL/6J mice.
- Extracted time- and frequency-based features from video data.
- Trained and investigated various machine learning classifiers and data augmentation techniques using expert-scored EEG/EMG data as ground truth.
Main Results:
- Developed a visual sleep classifier achieving high accuracy (0.92 ± 0.05) in distinguishing wake, NREM, and REM sleep.
- Identified and genetically validated video features correlated with breathing rates, showing distinct variability patterns between NREM and REM sleep.
- Successfully applied the method to detect sleep stage disturbances induced by amphetamine administration.
Conclusions:
- Machine learning-based visual classification of sleep is a viable and accurate alternative to invasive EEG/EMG scoring.
- This non-invasive approach significantly reduces the barrier for high-throughput sleep studies and screening mutant mice for sleep abnormalities.
