Sleep stage scoring using the neural network model: comparison between visual and automatic analysis in normal...

Area of Science:

• Neuroscience
• Sleep Medicine
• Biomedical Engineering

Background:

• Accurate sleep staging is crucial for diagnosing sleep disorders.
• Visual scoring of polysomnography (PSG) recordings is time-consuming and subject to inter-rater variability.
• Automated sleep analysis offers a potential solution for efficient and consistent sleep staging.

Purpose of the Study:

• To compare the performance of an automated sleep staging system against manual visual scoring.
• To evaluate the impact of expert supervision on the accuracy of automated sleep staging.
• To assess the suitability of automated sleep staging for large-scale sleep research.

Main Methods:

• Validation using a dataset of 60 subjects (20 normal controls, 20 depressed, 20 insomniacs).
• Comparison of automated scoring with visual scoring by two experts on 30-second epochs.
• Calculation of inter-expert agreement (87.5%) and automated vs. expert agreement (82.3%).
• Inclusion of expert supervision for uncertain epochs, improving automated/expert agreement to 90% with supervision on 20% of data.

Main Results:

• The automated system achieved an 82.3% agreement rate with a single expert without supervision.
• Expert supervision on ambiguous epochs increased the agreement rate to 90%.
• Inter-expert variability was measured at 87.5% agreement.

Conclusions:

• The automated sleep staging system demonstrates satisfactory performance for large-scale investigations.
• Automated sleep staging is a useful and potentially more efficient alternative to traditional visual scoring.
• Targeted expert supervision can further enhance the accuracy of automated sleep staging systems.