A sleep staging model on wavelet-based adaptive spectrogram reconstruction and light weight CNN.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Automatic sleep staging is crucial for diagnosing sleep disorders.
• Electroencephalography (EEG) signals present challenges due to weak properties and complex frequency components during sleep stage transitions.

Purpose of the Study:

• To develop an effective and computationally efficient method for automatic sleep staging using EEG.
• To improve the feature representation of EEG signals for more robust sleep stage classification.

Main Methods:

• Wavelet-based adaptive spectrogram reconstruction (WASR) using seed growth to capture time-frequency patterns.
• Integration of Teager operator variant energy into WASR to generate additional spectrograms highlighting hidden EEG dynamics.
• Development of a lightweight Convolutional Neural Network (CNN) with depthwise separable convolution for enhanced feature extraction and classification.

Main Results:

• The proposed model achieved high performance on the Sleep-EDF 20 dataset, with an overall accuracy of 87.6%, F1-score of 82.1%, and Cohen kappa of 0.83.
• The enhanced spectrograms enabled the lightweight CNN to detect finer details of sleep stages, improving feature representation.
• The model demonstrated competitive results compared to baselines while significantly reducing computational cost.

Conclusions:

• The novel WASR method combined with a lightweight CNN offers an effective approach for automatic sleep staging.
• This method improves the accuracy and efficiency of sleep disorder diagnosis.
• The approach provides a robust and computationally inexpensive solution for analyzing sleep EEG data.