Abstracts

Rationale: Recent advancements in neuroimaging and computational neuroscience have facilitated the assessment of individual connectomes, the organization of a brain’s structural network. Mounting data from the connectome of patients with localization-related epilepsy have changed our understanding of the disease, revealing abnormalities throughout the brain network, well beyond the area of seizure focus. Whether these changes are cause or consequence of seizure onset and propagation is still matter of debate, but deriving quantitative measures to characterize different features of a brain’s network may help identify personalized biomarkers useful in predicting clinical outcomes. Here, we focused on betweenness centrality (BC), a measure of a node’s influence on the network, and its association with post-surgical seizure control among patients with non-lesional temporal lobe epilepsy (TLE). Methods: We reconstructed the individual brain connectome of 50 patients with TLE who underwent anterior temporal lobectomy (ATL) based on their presurgical 3T diffusion MRI. For each patient, we computed the BC of each of the 384 nodes in the network, based on the AICHA atlas, and compared the values between patients who were seizure-free (SF, Engel I, n = 36) and non-seizure-free (NSF, Engel II-IV, n = 14). We identified the top 10 nodes whose BC most reliably distinguished SF from NSF and these were entered into a discriminant analysis to determine their ability to classify patients based on their post-surgical outcomes. Results: Of the nodes identified, four were stronger influencers among NSF patients, of which three were ipsilateral to the clinically-identified seizure focus. Six nodes (five ipsilateral) were more central to the network of SF patients (Figure 1). The discriminant analysis of the model composed of these specific nodes was significant (Wilks’ λ = 0.500, χ2 = 29.8, df  = 10, p = .001) and correctly identified 90% of the cases (Table 1A). Leave-one-out approach as a method of cross-validation revealed that the BC of these ten nodes correctly classified 82% of patients based on their outcomes (Table 1B). In contrast, discriminant analysis on a model with age of onset, seizure frequency and seizure burden, duration of epilepsy, seizure risk factors, presurgical interictal EEG pattern, and age of surgery was not significant (Wilks’ λ = 0.96, χ2 = 1.89, df  = 8, p = .98) and only classified 32.7% of cases correctly on cross-validation. Conclusions: From the whole-brain connectome of patients with TLE, we identified regions acting as stronger influencers differently among SF and NSF patients. These nodes correctly identified post-ATL seizure outcomes in 90% of cases and, when cross-validated by means of U-method, reliably classified post-surgical outcome with 82% accuracy. These preliminary findings demonstrate that studying the centrality of certain brain regions in the structural connectome of patients with TLE may help identify biomarkers for prediction of post-surgical seizure outcome.  Funding: None to report