Estimating sensor-space EEG connectivity PART 1: Identifying best performing methods for functional connectivity in simulated data.

Abstract

Functional brain connectivity (FC) can be estimated using electroencephalography (EEG). However, there is considerable variability across studies in the FC measures used and in data (pre-)processing methods, leading to difficulties comparing and amalgamating results between studies. Thus, standardisation of EEG (pre-)processing for the measurement and reporting of FC is needed.We aimed to assess differences in FC estimates produced by different settings across multiple EEG pre-processing steps, (including re-referencing and epoching) to validate a reliable methodological pipeline for assessing EEG-FC in simulated EEG data.

We simulated EEG-FC data where the 'ground truth' of the connections is known and compared estimates of FC from this ground truth data across multiple FC measures and variations in multiple pre-processing steps.

Our results indicated that pre-processing steps that included segmenting the data into 40 or more epochs that were 6 s or more in length provided the most accurate estimation of the simulated FC. With regards to the data re-referencing, the Reference Electrode Standardization Technique or the common average re-referencing appeared best when used in conjunction with imaginary coherence and weighted phase lag index metrics. However, the magnitude-squared coherence FC measure performed best with the Current Source Density reference free techniques.

Our paper provides an evidence-base for the influence of referencing, epoch length and number, controls for volume conduction, and different FC metrics on EEG-FC measurement. Using this evidence, we present an initial and promising account of the best performing (pre-)processing choices for robust EEG-FC assessment.