Assessing the Consistency of EEG Connectome Topology Across Different Parcellation Approaches

Network graph theory (NGT) provides a powerful framework for characterizing the topological organization of brain networks and understanding functional integration and segregation within the brain. However, the estimation of NGT metrics is inherently dependent on the definition of network nodes, which is also determined by the choice of brain parcellation atlas. This dependency raises critical concerns about the comparability, reliability, and interpretability of graph-derived metrics across studies, particularly in clinical populations such as patients with Parkinson's Disease (PD). The present study aimed to systematically evaluate the influence of parcellation scheme on NGT metrics computed from EEG-based functional connectivity (FC) data in a cohort of PD patients. Specifically, the objective was to identify which metrics are robust to parcellation choice and thus potentially more reliable for use as biomarkers in clinical neuroscience. EEG recordings were acquired from 18 PD patients using a 64-channel wearable EEG cap during resting-state conditions. Source reconstruction was performed using a Boundary Element Model and cortical time series were extracted using two parcellation atlases: the anatomically defined Desikan-Killiany 56-region atlas obtained from Desikan Killiany 68 and the functionally defined Schaefer 100-region atlas. FC matrices were computed using the Phase Locking Value, and seven key NGT metrics were extracted from each matrix: clustering coefficient, participation coefficient, characteristic path length, global efficiency, strength, transitivity, and assortativity. The agreement between metrics derived from the two parcellations was assessed first using the Wilcoxon signed-rank test, followed by Bland-Altman analysis to evaluate systematic bias and/or other types of errors. Statistical analysis revealed significant differences across most metrics between the two atlases, indicating that parcellation choice can meaningfully affect the topological characterization of functional brain networks. However, two metrics characteristic path length (CPL) and participation coefficient (P) exhibited strong agreement between parcellations, with non-significant Wilcoxon test results (p = 0.913 for CPL; p = 0.055 for P) and Bland-Altman plots showing no trends and 95% confidence intervals of the bias including the zero line. These findings suggest that CPL and P are robust to parcellation scheme and may serve as stable indicators of global network integration and inter-modular connectivity, respectively underscoring the methodological sensitivity of NGT metrics to the choice of brain atlas. Nevertheless, the observed robustness of CPL and P suggests that certain global integration metrics may provide reliable markers for assessing brain network organization in PD, regardless of parcellation strategy. These findings have important implications for the standardization of brain network analyses and the development of reproducible, generalizable biomarkers. Future studies with larger cohorts and matched healthy control groups are warranted to validate these observations and extend them to broader clinical applications.