Estimating sensor-space EEG connectivity: Identifying optimal artifact reduction techniques for functional connectivity in real data

Objectives: Electroencephalography (EEG) can be used to assess functional brain connectivity (FC). However, there is considerable variability in the methods used for FC measurement across different studies, which may contribute to heterogeneity in research outcomes. We aimed to assess how different EEG pre-processing steps impact EEG-FC measurement when applied to real EEG data. Methods: Using the BrainClinics.com open-source EEG data repository we investigated how different pre-processing steps impacted the ability to detect age-related differences in alpha band FC and the test–retest reliability of FC measures. The pre-processing steps tested included artifact reduction techniques (Independent Component Analysis (ICA), wavelet-enhanced ICA (wICA), and Multi-channel Wiener Filters (MWF)), different epoch lengths (epochs that were 2 s versus 6 s in length), and different re-referencing montages (the common average reference (CAR) versus current source density (CSD) re-referencing). We also assessed different FC metrics including imaginary coherence (iCOH), real magnitude squared coherence (rMSC), and weighted phase lag index (wPLI) metrics. Results: The best performing pipeline at detecting age-related differences in alpha FC and providing high test–retest reliability included artifact reduction by ICA or wICA, data re-referenced using the CSD method, and FC measured by rMSC. Conclusion & significance: This paper presents evidence for an EEG pre-processing pipeline that provides good detection of meaningful effects and high test–retest reliability for sensor space EEG alpha frequency FC.