Automatic artefact detection in single-channel sleep EEG recordings.

Area of Science:

• Neuroscience
• Sleep Medicine
• Biomedical Engineering

Background:

• Quantitative electroencephalogram (EEG) analysis is crucial for sleep research and medicine.
• Manual artifact detection in EEG is time-consuming and limits large-scale studies.
• Developing automated artifact detection is essential for efficient and reliable EEG analysis.

Purpose of the Study:

• To evaluate the performance and generalizability of simple algorithms for detecting artifacts in polysomnographic EEG recordings.
• To optimize parameters for artifact detection algorithms.
• To compare EEG power spectra analyzed with automated artifact exclusion versus manual scoring.

Main Methods:

• Implemented and tested 14 different artifact detection algorithms.
• Optimized algorithm parameters using receiver operator characteristic curves on 32 sleep recordings.
• Validated algorithms on 21 recordings from healthy participants and 10 from patients across different laboratories.
• Compared average power density spectra derived from algorithm-based and expert-based artifact exclusion.

Main Results:

• Most algorithms reliably identified artifact-contaminated EEG epochs with high sensitivity and specificity (0.9).
• Optimal performance was achieved using thresholds for EEG slope, high-frequency power (25-90 Hz or 45-90 Hz), and adaptive autoregressive model residuals.
• Automated artifact exclusion resulted in average EEG power density spectra highly similar to those from expert scoring.

Conclusions:

• Simple algorithms can reliably exclude artifacts in EEG data with good performance.
• Automated artifact detection enables efficient analysis of large EEG datasets for sleep research, genetics, epidemiology, and precision medicine.
• Algorithm-based artifact exclusion provides results comparable to manual scoring, facilitating broader application of quantitative EEG analysis.