Enhancing sleep stage classification with 2-class stratification and permutation-based channel selection.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Accurate sleep stage classification from electroencephalographic (EEG) signals is crucial for diagnosing sleep disorders.
• Traditional methods often require extensive computational resources and a large number of EEG channels.

Purpose of the Study:

• To develop and evaluate a computationally efficient method for classifying sleep stages using a convolutional neural network (CNN).
• To identify the optimal minimal set of EEG channels for accurate sleep stage classification.

Main Methods:

• A CNN model named EEGNeX was developed to extract and classify sleep-related waveforms from EEG signals.
• A permutation-based channel selection process was employed to identify the most informative EEG channels.
• Performance was evaluated using accuracy, F-score, precision, recall, AUROC, and kappa values for 2-class sleep stage models.

Main Results:

• The EEGNeX model with 128 channels demonstrated high performance in classifying sleep stages.
• Using only the top 3 permutation-selected channels, performance metrics exceeded 80%, except for N1 versus N2 classification (kappa=0.52).
• Performance was superior to using 3 channels recommended by the American Academy of Sleep Medicine (AASM) or 3 random channels.

Conclusions:

• A 2-class CNN model utilizing 3 permutation-selected EEG channels offers a computationally efficient and effective approach for sleep stage classification.
• This method significantly reduces the number of required EEG channels while maintaining high classification performance.