Effects of eye artifact removal methods on single trial P300 detection, a comparative study
Foad Ghaderi1, Su Kyoung Kim, Elsa Andrea Kirchner
1Robotics Group, University of Bremen, Bremen, Germany; Robotics Innovation Center, German Research Center for Artificial Intelligence (DFKI GmbH), Bremen, Germany.
Journal of Neuroscience Methods
|September 24, 2013
Summary
Electroencephalography (EEG) artifact removal is crucial for accurate analysis. The Infomax and ADJUST methods combined offer superior eye artifact removal, improving P300 classification accuracy in EEG data.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Electroencephalographic (EEG) signals are frequently contaminated by eye artifacts, impacting data quality even in controlled settings.
- Effective artifact removal is essential for reliable analysis of EEG data, particularly for event-related potentials like P300.
- Various automated methods exist for artifact removal, but their comparative performance requires quantitative evaluation.
Purpose of the Study:
- To quantitatively compare standard EEG artifact removal techniques.
- To evaluate two independent component analysis (ICA) artifact identification approaches: ADJUST and correlation.
- To assess the impact of artifact removal methods on P300 single-trial classification accuracy.
Main Methods:
- Compared regression, filtered regression, Infomax, and second-order blind identification (SOBI) artifact removal methods.
- Utilized ADJUST and correlation for artifact identification within ICA.
- Cleaned EEG datasets by removing eye artifacts and subsequently performed P300 single-trial classification.
Main Results:
- The combination of Infomax and ADJUST demonstrated superior performance, yielding an average 0.6% improvement in classification accuracy across subjects.
- The SOBI and correlation combination performed the worst among the evaluated methods.
- Low-pass filtering EEG data at lower cutoffs (e.g., 4 Hz) also enhanced classification accuracy.
Conclusions:
- The Infomax ICA method combined with the ADJUST artifact identification technique provides a robust and effective approach for eye artifact removal in EEG.
- This combination outperforms other tested methods, including SOBI and correlation, without needing an artifact reference channel.
- Optimizing artifact removal strategies, potentially with low-pass filtering, is critical for improving the reliability of EEG-based analyses, such as P300 detection.
