EEG Reference Reliability: Referential vs Source Estimation
For EEG recordings, the Reference Electrode Standardization Technique (REST/rREST) is more reliable than traditional referential (RE) montages, as REST demonstrates the lowest relative error in estimating true scalp potentials across all tested conditions. 1, 2, 3
Understanding the Reference Problem
The choice of EEG reference is a fundamental issue affecting the accuracy of recorded potentials:
- Traditional referential (RE) montages using monopolar references (such as linked mastoids, earlobe, Fz, Pz, Oz, or Cz) typically produce large distortions in the measured EEG potentials 1, 2
- The ideal reference is theoretically at infinity, which is implicitly determined by forward theory but cannot be physically achieved 1, 2
- A re-reference procedure after online monopolar recording should be adopted to mitigate distortion effects 1, 2
Comparative Performance of Reference Methods
REST (Reference Electrode Standardization Technique)
REST is generally superior to average reference (AR) for all factors compared, including channel number, scalp regions, electrode layout, dipole source position and orientation, sensor noise, and head model perturbations 1, 2:
- REST shows the lowest relative error in estimating EEG potentials at infinity across all conditions tested 3
- REST is insensitive to head model perturbations, making it robust in clinical applications 1, 2
- The regularized version (rREST) allows for simultaneous denoising and reference estimation, further improving accuracy 3
- REST utilizes prior information based on the EEG generative model, providing biophysically informed estimates 3
Average Reference (AR)
- AR may serve as an alternative option primarily in high sensor noise cases 1, 2
- AR is subject to electrode coverage and dipole orientation but shows no close relation with channel number 1, 2
- AR results from biophysically non-informative prior assumptions 3
Linked Mastoids (LM)
Practical Implementation Considerations
For clinical applications, an average lead field provides results comparable to individual lead fields, making REST practical without requiring individual MRI-based head models 3:
- The regularization parameter for rREST can be selected using Generalized Cross-Validation (GCV), which approximates the optimal "oracle" choice 3
- When evaluated with 89 real resting state EEGs, rREST consistently yielded the lowest GCV values 3
- Realistic volume conductor models improve the performance of REST and rREST 3
Clinical Bottom Line
REST should be the first choice for re-referencing EEG data to obtain the most accurate representation of scalp potentials 1, 2. This recommendation is based on systematic evaluation showing REST's superiority across multiple factors affecting EEG accuracy, its robustness to technical variations, and its foundation in biophysically informed modeling of EEG generation.