Unsupervised online classifier in sleep scoring for sleep deprivation studies

Area of Science:

• Neuroscience
• Sleep Science
• Computational Biology

Background:

• Automated sleep scoring is crucial for research.
• Existing methods often require manual supervision or lack adaptability.
• Developing unsupervised algorithms for real-time sleep state detection is a significant challenge.

Purpose of the Study:

• To evaluate an unsupervised adaptive algorithm for real-time detection of sleep and wake states in rodents.
• To assess the algorithm's performance in conjunction with a device for selective paradoxical sleep deprivation (PSD).

Main Methods:

• A Bayesian classifier was developed to extract electroencephalogram (EEG) and electromyogram (EMG) features.
• The algorithm automatically categorizes 5-second epochs into sleep/wake states without human supervision.
• The system was coupled with a device for online selective PSD, and its performance was validated against human scoring and a 72-hour PSD protocol.

Main Results:

• The algorithm demonstrated high concordance with human scoring (κ > 70%) and high specificity for paradoxical sleep (PS) detection (92%).
• Real-time selective PSD significantly reduced PS amounts (4.7% vs. 8.9% baseline) and induced a significant PS rebound.
• The system proved effective for precise and efficient PSD in rodents.

Conclusions:

• The fully unsupervised, data-driven algorithm overcomes limitations of other automated sleep scoring methods.
• Coupled with a dedicated device, it offers an improved approach for selective PSD compared to traditional methods.
• This technology facilitates advanced sleep research by enabling precise manipulation and monitoring of sleep states.