Conference Paper Published

Abstract

Understanding how interactions between functional components in the brain form a network organization is a fundamental question in neuroscience. Graph theory has been applied to neural data to characterize these networks. However, multiple methodological decisions during network derivation raise questions about the reliability and reproducibility of such studies. In this study, we systematically investigated how frequency of interest, epoch length, and thresholding steps influence the stability and reliability of three key network measures derived from resting-state EEG: clustering coefficient, global efficiency, and Small-World Propensity. To ensure fair comparisons between different bands, we proposed using band-specific epoch lengths determined by the number of effective cycles at the frequency of interest, instead of a fixed length in seconds. Our findings reveal that clustering coefficient requires a specific threshold to achieve reliable estimates, while global efficiency benefits from fewer and stronger connections. Small-World Propensity showed less satisfactory reliability and may require more data for accurate estimation. These results emphasize the importance of examining reliability across different network measures before applying them in studies. The reliability varies across frequency bands, with higher-frequency oscillations needing more effective cycles to ensure accurate estimation. Our findings provide valuable insights for researchers in optimizing their choices of epoch length and thresholds in future EEG network studies, enhancing the reliability and reproducibility of these analyses.