SleepEEGNet: Automated sleep stage scoring with sequence to sequence deep learning approach

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Manual sleep stage scoring from electroencephalogram (EEG) is laborious and subjective.
• Inter-rater reliability issues hinder consistent diagnosis of sleep disorders.
• Automating sleep scoring is crucial for efficient and accurate clinical practice.

Purpose of the Study:

• To develop an automated sleep stage annotation method, SleepEEGNet, utilizing single-channel EEG.
• To enhance the accuracy and reliability of sleep disorder diagnosis through artificial intelligence.
• To address the class imbalance problem in sleep datasets for improved model performance.

Main Methods:

• SleepEEGNet employs deep convolutional neural networks (CNNs) for feature extraction.
• A sequence-to-sequence model captures temporal dependencies between sleep epochs.
• Novel loss functions were implemented to mitigate class imbalance issues.
• The model was evaluated on Physionet Sleep-EDF datasets (Fpz-Cz and Pz-Oz channels).

Main Results:

• SleepEEGNet achieved superior annotation performance compared to existing methods.
• The model attained an overall accuracy of 84.26%.
• Macro F1-score reached 79.66% with a Cohen's kappa (κ) of 0.79.

Conclusions:

• SleepEEGNet demonstrates high efficacy in automatic sleep stage annotation using single-channel EEG.
• The developed model offers a reliable tool to assist sleep specialists in diagnosis.
• The approach shows potential for application across diverse sleep EEG datasets.