A Deep Learning Model for Automated Sleep Stages Classification Using PSG Signals.

Area of Science:

• Computational Neuroscience
• Medical Technology
• Artificial Intelligence in Healthcare

Background:

• Sleep disorders are common symptoms of neurological diseases, impacting daily life quality.
• Traditional sleep stage scoring from polysomnogram (PSG) signals is manual, time-consuming, and lab-based.
• Automated sleep stage monitoring offers potential for accurate neurological disorder detection.

Purpose of the Study:

• To propose a flexible deep learning model for automated sleep stage classification using raw PSG signals.
• To develop a one-dimensional convolutional neural network (1D-CNN) for classifying sleep stages.
• To evaluate the model's performance on public sleep datasets.

Main Methods:

• Development of a 1D-CNN model utilizing electroencephalogram (EEG) and electrooculogram (EOG) signals.
• Training and validation of the model using two public databases: sleep-edf and sleep-edfx.
• Classification of sleep stages into two to six distinct classes.

Main Results:

• Achieved high accuracies on the sleep-edf database: up to 98.06% for two classes and 91.00% for six classes.
• Obtained excellent accuracies on the sleep-edfx dataset: up to 97.62% for two classes and 89.54% for six classes.
• Demonstrated superior performance across various sleep class granularities.

Conclusions:

• The developed deep learning model effectively classifies sleep stages from raw PSG data.
• The model shows high accuracy and is suitable for clinical application.
• Further testing with large-scale PSG data is recommended.