Test-retest reliability of functional brain network characteristics using resting-state EEG and graph theory

Abstract

Metrics of brain network organization can be derived from neuroimaging data using graph theory. We explored the test-retest reliability of graph metrics of functional networks derived from resting-state electroencephalogram (EEG) recordings. Data were collected within two designs: (1) within sessions (WS) design where EEG data were collected from 18 healthy participants in four trials within a few hours and (2) between sessions (BS) design where EEG data were collected from 19 healthy participants in three trials on three different days at least one week apart. Electrophysiological source activity was reconstructed and functional connectivity between pairs of sensors or brain regions was determined in different frequency bands. We generated undirected binary graphs and used intra-class correlation coefficient (ICC) to estimate reliability. We showed that reliabilities ranged from poor to good. Reliability at the sensor-level was significantly higher than source-level. The most reliable graph metric at the sensor-level was cost efficiency and at the source-level was global efficiency. At the sensor-level: WS reliability was significantly higher than BS reliability; high beta band in WS design had the highest reliability; in WS design reliability in gamma band was significantly lower than reliability in low and high beta bands. At the source-level: low beta band in BS design had the highest reliability; there was no significant main effect of frequency band on reliability; reliabilities of WS and BS designs were not significantly different. These results suggest that these graph metrics can provide reliable outcomes, depending on how the data were collected and analysed.