An automatic method using MFCC features for sleep stage classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Medicine

Background:

• Sleep stage classification is crucial for diagnosing sleep disorders.
• Traditional methods rely on manual analysis of biological signals over 30-second intervals.
• Deep learning models show promise for improving sleep scoring efficiency and accuracy.

Purpose of the Study:

• To propose a novel deep learning model for automated sleep stage classification.
• To leverage Mel-frequency Cepstral Coefficients (MFCCs) as a key feature for sleep scoring.
• To evaluate the model's performance on established sleep datasets.

Main Methods:

• A deep convolutional neural network (CNN) integrated with a long short-term memory (LSTM) model was developed.
• Two-dimensional (2D) MFCC features were extracted from electroencephalogram (EEG) and electromyogram (EMG) signals.
• The model architecture involved convolutional layers, an LSTM layer, a fully connected layer, and a softmax classifier.

Main Results:

• The proposed CNN-LSTM model achieved high accuracy in sleep stage classification.
• Performance metrics included 82.35% accuracy and 0.75 Cohen's kappa on the SHHS dataset.
• The model demonstrated effectiveness on the UCDDB dataset with 73.07% accuracy and 0.63 Cohen's kappa.

Conclusions:

• The novel deep learning approach using 2D MFCC features provides an effective method for sleep stage classification.
• The model's efficiency is enhanced by reducing the need for deep layers, leading to faster training times.
• This approach offers a promising advancement for automated sleep disorder diagnosis and analysis.