Abstracts

Rationale: Connectomic analysis of temporal lobe epilepsy (TLE) using graph theoretical methods is increasingly found to be a powerful quantitative method for investigating epileptic brain networks on the whole-brain level. Increasingly, studies are demonstrating the utility of graph theory measures of functional connectivity for identifying network abnormalities in TLE, and for serving as diagnostic markers of TLE laterality or disease extent. However, the majority of graph theory investigations of functional connectivity have relied on the assumption of temporal stationarity. Indeed, current analyses have largely been conducted under the assumption that the level of spontaneous fluctuations is constant over the length of the scan. In this study, we estimate the dynamic nature of graph theoretical measures of whole-brain functional connectivity, with the aim of identifying which graph theory measures of brain exhibit greater temporal stability and are more robust to static functional connectivity analysis in TLE.Methods: Interictal resting-state fMRI was performed in 32 temporal lobe epilepsy (TLE) patients and 24 healthy controls. Sliding window analysis was used to extract time-varying graph theory metrics across the length of the scan. Dynamic changes in graph theory metrics quantified through Bayesian hidden Markov modeling. Temporal stability was estimated for various graph theory measures of network connectivity, including small-world index, global integration measures (global efficiency, characteristic path length), local segregation measures (clustering coefficient, local efficiency), and centrality measures (betweenness centrality, eigenvector centrality).Results: Several graph theory measures, including small-world index, betweenness centrality, and global integration measures, were more robust to the assumption of temporal stationarity than other measures. The exception was clustering coefficient for TLE patients, which was the least temporally stable network measure for healthy controls but most stable for TLE patients. For centrality measures, betweenness centrality was consistently more temporally stable than eigenvector centrality. For global integration measures, global efficiency was consistently more temporally stable than characteristic path length. Subject-level differences in the level of temporal stability of clustering coefficient were found to be potentially useful as a diagnostic marker for TLE.Conclusions: Our results confirm and extend findings from other studies regarding the dynamic nature of functional connectivity, and suggest that the robustness of static functional connectivity analysis may depend on the graph theory measure investigated. Temporal stability of network topology may itself serve as a marker for TLE. Development of advanced statistical methods which explicitly account for the dynamic nature of functional connectivity in graph theory analyses may improve the sensitivity of connectomic investigations in TLE.