Automatic sleep stage classification: A light and efficient deep neural network model based on time, frequency and...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Traditional methods rely on multi-channel EEG, which can be cumbersome.
• Developing efficient and accurate single-channel EEG-based methods is desirable.

Purpose of the Study:

• To propose a novel method for automatic sleep stage classification using single-channel EEG.
• To investigate the utility of fractional Fourier transform (FRFT) domain features for improving classification accuracy.
• To develop a computationally efficient deep learning model for sleep stage classification.

Main Methods:

• Extracted time, frequency, and FRFT domain features from single-channel EEG.
• Utilized a bidirectional long short-term memory (BiLSTM) network for classification.
• Trained the model using the American Academy of Sleep Medicine (AASM) criteria for sleep staging.

Main Results:

• The inclusion of FRFT features improved the performance of sleep stage classification.
• Achieved an overall accuracy of 81.6% for Fpz-Cz EEG data from the Sleep-EDF dataset.
• The proposed model has a small parameter size (0.31 MB), approximately 5% of DeepSleepNet, with comparable performance.

Conclusions:

• FRFT domain features can enhance the accuracy of automatic sleep stage classification from single-channel EEG.
• The proposed BiLSTM model is a light and efficient deep learning approach for sleep staging.
• The model shows promise for on-device machine learning applications in sleep analysis.