Abstracts

Rationale:
Epilepsy is increasingly recognized as a disorder of structural brain networks. Diffusion MRI tractography is a non-invasive technique to model macroscopic structural brain networks, which are commonly assessed by graph theoretical techniques for an analysis of topological properties. In healthy individuals, differences in the regional definition has been shown to influence such topological properties [1]. In this assessment, we evaluated the effect of cortical parcellation schemes and the inclusion/exclusion of subcortical areas on structural network properties derived from graph theoretical analyses in healthy individuals and patients with drug-resistant epilepsy. Moreover, we assessed how case-control differences when comparing epileptic patients to controls are impacted by different parcellation resolutions.
Method:
Participants and image acquisition. We included 35 individuals with drug-resistant temporal lobe epilepsy and 34 age- and sex-matched healthy controls. Whole-brain connectomes were derived from diffusion weighted images. Cortical and subcortical segmentations were obtained from a high resolution T1 weighted image.  Parcellations. Eight different parcellations were evaluated: four included cortical nodes only (Destrieux 148, Schaefer 200, 500, 1000 nodes, and subcortical structures obtained from Volbrain). Connectome topology. For each subject, an anatomically constrained brain tractography with filtering of tractograms was generated, and eight connectomes were calculated based on the streamline counts for each parcellation. Subject- and parcellation-specific topological network parameters were computed (nodal efficiency and degree centrality; clustering coefficient and characteristic path length). Mean absolute percentage error metrics assessed differences in network metrics across parcellations, with or without subcortical structures. Global network properties in individuals with TLE were then compared to controls across the range of parcellation choices using Mann-Whitney tests. Nodal network measures were compared using Cohen's D effect sizes. To highlight robust findings, we calculated the mean, vertex-wise, Cohen’s d value across different parcellation schemes.
Results:
a) Effects of subcortical nodes on network topology (Figure 1). Inclusion of subcortical structures had marked effects on nodal and global path length across all Schaeffer parcellations; more subtle effects were observed on nodal and global (i) degree centrality (at lower resolution parcellation) and (ii) clustering coefficient (at higher resolution). Nodal efficiency was not affected by the inclusion of subcortical nodes.  •b) Effects of parcellation resolution on case-controls analysis (Figure 2A). Frontoparietal path length increases in TLE, relative to controls, varied considerably across parcellations, but remained stable in the temporal lobes. In contrast, clustering coefficient and in bilateral frontal areas was observed with an increasing parcellation granularity. Efficiency was relatively robust across parcellations, except in temporal and frontal areas. Degree centrality maintained similar effect sizes across the parcellation choices. c) Robust network topology across parcellations (Figure 2B). Despite variation across parcellations, the main differences between groups previously described were found with a consensual approach.
Conclusion:
While our results demonstrate variability in structural network topology metrics across parcellation choices, we could nevertheless identify robust network alterations across different methodological choices, suggesting scale invariant network perturbations of drug-resistant TLE.
Funding:
:FRQS-291486