LG-Sleep: Local and Global Temporal Dependencies for Mice Sleep Scoring.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Manual sleep scoring is time-consuming and requires expertise.
• Automated sleep scoring methods are needed for preclinical and clinical research.
• Accurate sleep stage identification in mice is vital for studying sleep patterns and disorders.

Purpose of the Study:

• To introduce LG-Sleep, a novel deep neural network for subject-independent mice sleep scoring using EEG signals.
• To develop a model that effectively utilizes local and global temporal transitions in EEG data.
• To achieve robust sleep scoring performance with limited training samples.

Main Methods:

• Developed LG-Sleep, a subject-independent deep neural network architecture.
• Employed time-distributed convolutional neural networks for local temporal transition extraction.
• Utilized long short-term memory blocks to capture global temporal transitions.
• Optimized the model using an autoencoder-decoder approach for generalization.

Main Results:

• LG-Sleep demonstrated superior performance compared to conventional deep neural networks.
• The model achieved good performance across wake, REM, and NREM sleep stages.
• LG-Sleep showed adaptability and effectiveness with limited training samples.

Conclusions:

• LG-Sleep offers an efficient and accurate automated solution for mice sleep scoring.
• The subject-independent nature and autoencoder-decoder optimization enhance model generalization.
• This approach facilitates robust sleep analysis in preclinical research, even with data limitations.