Estimating a neutral reference for electroencephalographic recordings: the importance of using a high-density montage and a realistic head model

Quanying Liu, Joshua H. Balsters, Marc Baechinger, Onno van der Groen, Nicole Wenderoth, Dante MantiniJournal of Neural Engineering 2015 · DOI

TL;DR

This paper investigates how the choice of EEG reference electrode impacts data analysis, finding that a realistic head model and high-density electrode montages are crucial for accurate "neutral" reference estimation, especially for the REST method. The study provides quantitative evidence to guide researchers in selecting the most effective re-referencing strategy for their specific EEG studies.

What Problem Does This Paper Address?

Electroencephalography (EEG) measures brain activity by recording electrical potentials on the scalp. However, EEG is a relative measure, meaning the signal at each electrode is compared to a reference electrode. The choice of this reference is critical, as it can introduce distortions and affect the interpretation of brain activity. Historically, common references like linked-mastoids (LMR) or average reference (AR) have been used, but they have known limitations. The Reference Electrode Standardization Technique (REST) offers a theoretically more neutral reference by estimating it at an infinite distance, but its performance is highly dependent on the accuracy of the head model and electrode placement. This paper addresses the question of how much these factors—specifically, using a high-density electrode montage and a realistic, individualized head model—impact the accuracy of different re-referencing techniques, particularly REST, compared to AR and LMR.

Methods

The researchers employed a multi-pronged approach:

  • Simulated EEG Data: They generated realistic simulated EEG data by projecting known dipole source activity onto a realistic, BEM (Boundary Element Method) based head model derived from an MRI. This allowed them to precisely control the ground truth and quantify the error introduced by different re-referencing methods.
  • Re-referencing Techniques: They systematically applied three common re-referencing techniques:
  • Linked-Mastoid Reference (LMR): A traditional method using mastoid electrodes.
  • Average Reference (AR): Calculates the average of all electrode signals as the reference.
  • Reference Electrode Standardization Technique (REST): A more advanced method that estimates a reference at an infinite distance, requiring a head model and leadfield matrix.
  • Varying Parameters: They systematically varied key parameters to assess their impact:
  • Electrode Density: Simulations were run with different numbers of electrodes (21, 71, 128, and 256 channels).
  • Head Models: REST's performance was evaluated using a single-sphere, a three-concentric-sphere, and the realistic BEM head model.
  • Electrode Position Uncertainty: Random and systematic shifts in electrode positions were introduced to test REST's robustness.
  • Artifacts: Three types of artifactual patterns (localized, homogeneous, and random noise) were added to the simulated data to assess how well each re-referencing method could attenuate them.
  • Real EEG Data Analysis: To validate their simulation findings, they analyzed real EEG data from a target detection experiment using AR and REST.

Datasets

  • Simulated Data: Generated using a realistic BEM head model derived from the Colin27 MRI template.
  • Real EEG Data: Collected from 17 healthy participants performing a visual oddball paradigm.

Key Insights and Ideas

  • LMR is Problematic: LMR consistently introduced significant distortions, altering scalp topography and leading to high errors, especially for certain dipole orientations. This suggests it's less suitable for accurate neurophysiological interpretation.
  • REST's Sensitivity to Head Model: REST's accuracy is highly dependent on the realism of the head model. A realistic BEM model significantly outperforms spherical models, especially with high-density montages.
  • High-Density Montages are Key: Both AR and REST benefit greatly from a higher number of electrodes. This improves the accuracy of the estimated reference, reducing reconstruction errors.
  • Electrode Position Matters for REST: REST's performance degrades with even small uncertainties in electrode positions, highlighting the need for precise localization when using this method.
  • Artifact Handling Varies: REST appears to be more robust to certain types of artifacts (e.g., localized) than AR, although both methods struggle with some patterns. Homogeneous artifacts were effectively removed by all methods.
  • Real Data Corroborates Simulations: The analysis of real EEG data supported the simulation findings, showing REST potentially yielding larger ERP signals with less noise compared to AR in the parietal region.

Key Results

  • Error Reduction: REST with a realistic head model and high-density montage achieved the lowest reconstruction errors (gRE < 1%). LMR showed very high errors (e.g., 79.41% for Y-orientation), while AR also performed reasonably well but generally worse than REST.
  • Montage Impact: With a 256-channel montage, REST with a realistic head model consistently outperformed AR. However, with lower density montages (21-channel), REST showed more reliable reconstruction than AR, even with spherical models.
  • Electrode Position Impact: For REST, electrode position uncertainty of just 2 mm could increase the global relative error to levels comparable to AR (around 5%).
  • Artifact Reduction: REST showed better artifact attenuation than AR for localized and random artifacts, though homogeneous artifacts were removed by all methods.

Limitations and Future Work

  • Template Head Model: The simulations used a template MRI (Colin27) rather than individual subject MRIs. Future work could explore the impact of individual variability.
  • BEM Model Focus: The study focused on a BEM head model for realism. Future research could investigate other realistic models like FEMs or FDMs, which might offer even greater accuracy.
  • Artifact Types: While several artifact types were tested, future studies could explore a wider range of artifacts and their interactions with different re-referencing methods.
  • Volume Conduction: The paper acknowledges that EEG volume conduction is a complex issue that warrants further modeling.
  • Practical Considerations: The authors note that while high-density EEG is beneficial, it comes with practical challenges like preparation time and increased volume conduction effects between neighboring channels.

The authors conclude that while REST offers superior accuracy with a realistic head model and high-density montage, AR might be a more robust choice when precise electrode position information is unavailable, especially with high-density systems. This work provides valuable quantitative guidance for researchers in choosing the most appropriate EEG re-referencing strategy.

Summary updated

Abstract

Our study is the first to quantitatively assess the performance of EEG re-referencing techniques in relation to the use of a high-density montage and a realistic head model. We hope our study will help researchers in the choice of the most effective re-referencing approach for their EEG studies.
BibTeX
@article{liu2015estimating,
  author = {Quanying Liu and Joshua H. Balsters and Marc Baechinger and Onno van der Groen and Nicole Wenderoth and Dante Mantini},
  title = {{Estimating a neutral reference for electroencephalographic recordings: the importance of using a high-density montage and a realistic head model}},
  journal = {Journal of Neural Engineering},
  year = {2015},
  doi = {10.1088/1741-2560/12/5/056012},
}

Loading related content...

Figure