Comparing Manual and Automatic Artifact Detection in Sleep EEG Recordings

Area of Science:

• Neuroscience
• Signal Processing

Background:

• Sleep electroencephalogram (EEG) data frequently contain artifacts that can distort analysis.
• Both visual inspection and automatic methods are used to identify and remove artifactual EEG segments.

Purpose of the Study:

• To systematically evaluate the impact of artifacts on sleep EEG power spectrum density (PSD).
• To compare the effectiveness of visual artifact detection against a simple automatic method using Hjorth parameters.
• To determine if manual artifact removal is essential for accurate sleep EEG analysis.

Main Methods:

• Systematic exploration of artifact effects on sleep EEG PSD.
• Comparison of gold-standard visual artifact detection with an automatic detector using Hjorth parameters.
• Analysis of all-night average PSD across different artifact detection methods.

Main Results:

• Most distortions in average PSD are caused by a small number of severe artifacts, primarily affecting beta, gamma, and NREM delta frequencies.
• Visual and automatic artifact detection methods showed only moderate agreement.
• Despite detection differences, all methods yielded highly similar all-night average PSDs, preserving known age and sex correlations.
• Accurate PSD estimates can be obtained from a fraction of the data epochs.

Conclusions:

• Artifacts in sleep EEG recordings pose a minor and solvable problem.
• Visual inspection of EEG data for artifact removal is not strictly necessary.
• Automatic artifact detection methods are sufficient for accurate sleep EEG analysis, especially for large databases.