MLS-Net: An Automatic Sleep Stage Classifier Utilizing Multimodal Physiological Signals in Mice.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Automatic sleep stage classification (ASSC) traditionally uses feature-based methods or deep neural networks (DNNs).
• Feature-based methods offer interpretability but lack spatio-temporal context, while DNNs capture complex patterns but suffer from overfitting and high computational costs.
• Existing methods face challenges in accuracy and efficiency for end-user applications in animal sleep research.

Purpose of the Study:

• To develop a novel neural network model, MLS-Net, for accurate and efficient automated sleep staging in mice.
• To integrate the strengths of feature extraction and deep learning to overcome limitations of existing ASSC approaches.
• To leverage multimodal sleep signals including EEG, EMG, and eye movements for improved classification performance.

Main Methods:

• Developed MLS-Net, a neural network model combining temporal and spectral features from multimodal signals (EEG, EMG, EMs).
• Incorporated a bidirectional Long Short-Term Memory (bi-LSTM) network to capture spatio-temporal nonlinear dynamics in sleep data.
• Evaluated MLS-Net performance against traditional feature-based and other neural network algorithms using a multimodal dataset.

Main Results:

• MLS-Net achieved an overall classification accuracy of 90.4% for mouse sleep stages.
• Achieved high performance for Rapid Eye Movement (REM) sleep classification: 91.1% precision, 84.7% sensitivity, and 87.5% F1-Score.
• Outperformed existing neural network and feature-based algorithms on the multimodal sleep dataset.

Conclusions:

• MLS-Net effectively integrates feature engineering and deep learning for superior automated sleep staging in mice.
• The model demonstrates robust performance and potential for broader application in sleep research.
• MLS-Net offers a promising solution for accurate and computationally efficient analysis of multimodal sleep data.