Rating by detection: an artifact detection protocol for rating EEG quality with average event duration

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Quantitative evaluation is crucial for optimal electroencephalography (EEG) artifact removal algorithm development.
• Visual inspection of real EEG for artifact removal pipeline selection is impractical.
• Hand-crafted EEG data only allow artifact removal assessment in simulated environments.

Purpose of the Study:

• To propose a novel, principled approach for quantitatively evaluating algorithmically corrected EEG without ground truth in real-world conditions.
• To develop a reliable protocol for comparing artifact removal methods on real EEG data.

Main Methods:

• An offline evaluation protocol utilizing an artifact detector to score artifact presence.
• Calculating the average duration of detected artifacts as a measure of deviation from background EEG activity.
• Employing a configurable Wiener-based artifact removal method to validate the detection protocol's reliability.

Main Results:

• Extensive quantitative experiments comparing multiple Wiener filters demonstrated consistent rankings aligned with theoretical expectations.
• The proposed rating-by-detection protocol proved reliable in evaluating artifact removal effectiveness.
• The average event duration measure provided a single, quantitative score for deviation from background activity.

Conclusions:

• The rating-by-detection protocol with average event duration is valuable for EEG practitioners and developers.
• Reliable comparisons between various artifact filtering configurations are achievable on real EEG data, even without ground truth neural signals.
• This approach facilitates the advancement of artifact removal techniques in electroencephalography.