Automatic sleep stages classification using multi-level fusion.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Sleep efficiency is crucial for overall health and can be assessed through sleep stage classification.
• Automatic sleep stage classification using multi-signal data aims to improve accuracy.
• Fusion methods are employed to integrate data from multiple biological signals.

Purpose of the Study:

• To propose and evaluate a multi-level fusion method for enhanced sleep stage classification.
• To improve the accuracy of classifying sleep stages using electroencephalography (EEG) and electromyography (EMG) signals.
• To specifically enhance the classification of the N1 sleep stage, which is often challenging.

Main Methods:

• Feature-level fusion using a convolutional neural network (CNN) to integrate extracted features from multi-signal data (EEG and EMG).
• Decision-level fusion to combine the classification results obtained from the fused feature data.
• Utilizing the public Sleep-EDF dataset for performance evaluation.

Main Results:

• The proposed multi-level fusion method achieved a classification accuracy of 87.2%.
• This accuracy surpasses that of single-level fusion methods and existing approaches.
• Significant performance improvement was observed in classifying the N1 sleep stage.

Conclusions:

• The multi-level fusion method offers a superior approach for automatic sleep stage classification compared to single-level fusion.
• This method effectively enhances the accuracy of sleep stage classification, particularly for difficult stages like N1.
• The findings contribute to more precise sleep analysis, potentially improving health assessments based on sleep efficiency.