EEG is better left alone.

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Large publicly available electroencephalography (EEG) databases require automated preprocessing methods.
• The optimal automated approach for EEG data preprocessing remains undetermined due to a lack of robust data quality metrics.
• Existing methods for analyzing event-related potentials (ERPs) are sensitive to preprocessing choices.

Purpose of the Study:

• To develop and validate a novel EEG data quality metric.
• To compare the efficacy of various automated preprocessing pipelines in maximizing ERP significance.
• To identify optimal preprocessing strategies for large-scale EEG data analysis.

Main Methods:

• A simple, robust EEG data quality metric was designed, assessing the percentage of significant channels within a 100 ms post-stimulus window.
• Three public EEG datasets were utilized to evaluate automated preprocessing techniques.
• Optimized preprocessing pipelines were implemented and compared across leading open-source software (EEGLAB, FieldTrip, MNE, Brainstorm).

Main Results:

• Automated data corrections, excluding high-pass filtering and bad channel interpolation, generally had no effect or decreased the percentage of significant channels.
• Referencing and advanced baseline correction methods significantly reduced performance.
• Automated Independent Component Analysis (ICA) for artifact rejection did not reliably improve performance.
• Only one optimized pipeline demonstrated significantly better performance than simple high-pass filtering.

Conclusions:

• Simple high-pass filtering and bad channel interpolation are effective preprocessing steps for EEG data analysis.
• Many advanced automated preprocessing methods do not enhance, and can even degrade, the quality of EEG data for ERP analysis.
• The developed data quality metric provides a valuable tool for optimizing EEG preprocessing pipelines.