Automated Sleep Stage Classification in Home Environments: An Evaluation of Seven Deep Neural Network Architectures

Area of Science:

• Neuroscience and Biomedical Engineering
• Focus on sleep science and computational methods.

Background:

• Sleep disorders are a growing public health concern.
• There is a need for accessible, reliable sleep monitoring outside clinical settings.
• Home-based sleep monitoring requires computationally efficient automated classifiers.

Purpose of the Study:

• To develop and assess pragmatic, computationally efficient automated classifiers for home-based sleep stage classification.
• To compare the performance of seven prominent deep learning neural network architectures for this application.

Main Methods:

• Rigorous assessment of seven neural network architectures (LeNet, ResNet, VGG, MLP, LSTM-CNN, LSTM, BLSTM).
• Utilized sleep recordings from a cohort of 20 subjects for evaluation.
• Conducted a comprehensive architectural analysis focusing on strengths and limitations for home use.

Main Results:

• LeNet, VGG, and ResNet demonstrated superior performance compared to recent literature.
• LeNet was identified as the most suitable architecture for home-based sleep monitoring.
• LSTM and BLSTM architectures showed relatively lesser compatibility for this specific application.

Conclusions:

• Automated sleep stage classification is feasible using lightweight neural networks.
• This approach is suitable for scenarios with constrained computational resources.
• The findings support the development of accessible and reliable home-based sleep monitoring solutions.