Irreversible functional network deformation induced by preprocessing scheme order in multivariate EEG recordings.

Summary

This methodological study demonstrated that applying re-referencing before filtering can generate spurious cross-correlations in multivariate EEG recordings. The distortion was strongest and irreversible in signals with steep power spectra, including slow-wave sleep, anesthesia, postictal states, neonatal EEG, and vegetative states; filtering followed by re-referencing with surrogate correction was recommended.

Key Findings

• Re-referencing followed by filtering produced spurious cross-correlations that increased with frequency band and depended on the signal power spectral exponent.
• For signals with power spectral exponent β ≥2, a pattern relevant to anesthesia and slow-wave sleep, the artificial correlations were reported to be irreversible.
• Filtering followed by re-referencing, combined with surrogate-corrected cross-correlations, was recommended for multivariate signals with β ≥2.

Clinical Implications

EEG studies used for anesthesia depth assessment, emergence research, epilepsy monitoring, and disorders of consciousness should prespecify preprocessing order and use surrogate-based correction when estimating functional connectivity. Existing datasets may require reanalysis before biological conclusions are accepted.

Limitations

• The abstract does not provide the number of human participants or detailed effect estimates for all signal conditions.
• The findings were demonstrated in EEG and require validation across additional recording modalities, acquisition systems, and preprocessing implementations.

Future Directions

Future studies should establish community-wide preprocessing benchmarks, reanalyze published anesthesia and consciousness datasets, and determine how preprocessing artifacts affect clinical EEG biomarkers and machine-learning models.