Sleep stages classification based on feature extraction from music of brain

Area of Science:

• Neuroscience and Biomedical Engineering
• Signal Processing and Machine Learning

Background:

• Sleep stage classification is crucial for diagnosing sleep disorders and preventing cognitive risks.
• Current methods for sleep stage classification can be complex and require multi-channel data.

Purpose of the Study:

• To propose a novel method for classifying sleep stages using electroencephalogram (EEG) signals mapped to music.
• To introduce a new single-channel EEG sonification technique for sleep analysis.

Main Methods:

• EEG signals were mapped to musical parameters (tempo, scale) to generate musical notes.
• Features were extracted from the musical notes and reduced using algorithms.
• A two-stage classification structure was employed for 5-class and 2-class sleep stage identification.

Main Results:

• The method achieved high classification accuracies across multiple databases: 89.5% for 5 classes (Cap sleep) and 90.1% for 2 classes (S1 vs. REM) on the Cap sleep database.
• Overall correct classification for 6 sleep stages reached 88.13% on the Cap sleep database.
• The proposed EEG sonification and classification method demonstrated comparable or superior performance to contemporary techniques.

Conclusions:

• The novel EEG-to-music mapping method is effective for sleep stage classification.
• This approach offers a promising, potentially simpler alternative for sleep disorder diagnosis using single-channel EEG.
• The study highlights the potential of sonification in biomedical signal analysis.