Robust, automated sleep scoring by a compact neural network with distributional shift correction.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Manual sleep scoring is time-consuming and a bottleneck in sleep research.
• Deep learning on electroencephalogram (EEG) and electromyogram (EMG) data shows promise for automated sleep scoring.
• Standardization of input data is crucial for machine learning but can remove biologically relevant variability.

Purpose of the Study:

• To develop a standardization method that preserves experimental variability for improved sleep scoring.
• To address the systematic bias introduced by common standardization methods in sleep scoring algorithms.
• To provide robust computational tools for automated sleep scoring.

Main Methods:

• Introduction of mixture z-scoring, a novel standardization technique.
• Validation using simulated data and mouse in vivo electroencephalogram and electromyogram data.
• Development of a free, open-source user interface with a compact neural network.

Main Results:

• Common standardization methods introduce systematic bias in sleep scoring.
• Mixture z-scoring effectively preserves crucial experimental variability and avoids bias.
• The developed tool achieves accuracy comparable to contemporary sleep scoring methods.

Conclusions:

• Mixture z-scoring is a superior standardization method for sleep scoring, particularly in experimental contexts.
• Automated sleep scoring using mixture z-scoring enhances accuracy and efficiency in biological sleep studies.
• The open-source tool facilitates robust and accessible sleep state analysis.