Abstracts

Rationale: Resection of the epileptogenic zone is a potential treatment for patients with drug-resistant epilepsy, resulting in seizure freedom for about one-half to two-thirds of patients (Spencer and Huh 2008). Improvement in localization of the epileptogenic zone is needed to increase the success rate. Resting-state magnetoencephalography (rs-MEG) functional connectivity analysis has been investigated for localization of the epileptogenic zone (Nissen, Stam et al. 2018). This study investigated whether the rs-MEG network topology in the source space can be used to lateralize epilepsy. Methods: Rs-MEG data were collected from 22 healthy controls (HCs), 19 patients with right-hemispheric epilepsy (RHPs), and 30 patients with left-hemispheric epilepsy (LHPs). Scalar beamformer weights were used to reconstruct time series for the centroids of 210 cortical regions of interest (ROIs) defined by the Brainnetome atlas (Fan, Li et al. 2016). The debiased weighted phase lag index (dwPLI) between the ROIs was used to construct adjacency matrices (210-by-210) in the delta, theta, alpha, low beta, high beta, low gamma, and high gamma frequency bands. Two global graph measures, global efficiency (GE) and characteristic path length (CPL), were calculated based on three quadrants (105-by-105) of the adjacency matrices, representing dwPLI of right versus right (RvR), left versus left (LvL), and right versus left (RvL) ROIs. The Wilcoxon rank sum test was used to investigate significant differences between the graph measures of the three groups. Based on the results, the CPL of RvR in all frequency bands was chosen as input features to train, cross-validate, and test a Naïve Bayes classifier to classify the three groups. Results: The CPL of RvR was significantly lower for RHPs than for LHPs and HCs in the theta band (P < 0.005) and significantly greater for LHPs than for RHPs and HCs in the delta band (P < 0.01) (Fig. 1). A similar pattern but in opposite direction was observed for the GE of RvR (Fig. 1). This pattern was absent in CPL and GE of RvL and LvL. When used as input features of a Naïve Bayes classifier, the CPL of RvR in all frequency bands was able to classify the three groups with an 80.6% accuracy and an area under the receiver characteristic operating curve of 0.991 for HCs, 0.878 for LHPs, and 0.860 for RHPs. Conclusions: The results revealed that LHPs may be distinguished from RHPs by observing the CPL of RvR in the delta and theta bands. Therefore, an rs-MEG network topology in the source space may be valuable for lateralizing epilepsy and may be incorporated in pre-surgical evaluations. Funding: This study was funded by the Children’s Foundation Research Institute & The Shainberg Neuroscience Fund, Memphis, TN.