Abstracts

Rationale: Magnetoencephalography (MEG) resting state networks evaluated as band limited amplitude envelope correlations (AECs) overlap with functional Magnetic Resonance Imaging (fMRI) networks in healthy controls (Brookes, PNAS 2011). Using data acquired from epilepsy patients prior to surgery, we studied the correspondence between MEG resting state connectivity and fMRI connector hubs. In brain networks, connector hubs are regions exhibiting denser connections than others and participating in inter-network communications. Our aim is to investigate how patient specific MEG connectivity and fMRI connector hubs could infer surgical outcome in epilepsy patients. Methods: We selected EEG/MEG and EEG/fMRI data, acquired prior to surgery, from 8 patients with refractory focal epilepsy. 5 out of 8 patients were seizure-free (Engel Ia) after surgery.From EEG/MEG, a reliable estimate of the generator of Interictal Epileptic Discharges (IEDs) was first localized with Maximum Entropy on the Mean (MEM)  (Chowdhury, HBM 2018). An homologous contralateral region to this generator was then selected to compare resting state connectivity patterns.MEG resting state analysis consisted in localizing fluctuations in the alpha band (8-13 Hz) over 40s of IEDs free data. First, MEM was applied to localize MEG data in the alpha band. IED generator and the homologous contralateral region were selected as seeds to calculate AEC values with the rest of the cortex. MEG laterality maps were then calculated by subtracting AEC values for seeds in IED generator from AEC values for seeds in homologous region. In resulting laterality maps, a positive value means that the region is more connected to the IED generator than to the homologous contralateral region.  fMRI data was analysed using the SParsity-based Analysis of Reliable K-hubness (SPARK) method (Lee, NIMG 2016). SPARK provides a multivariate analysis of connector hubs, assuming that a voxel can be involved in more than one (k) functional resting state networks. SPARK counts the number of networks involved in each voxel, resulting in a k-hubness map. Results: Fig.1a is presenting a right temporal seizure-free patient, for whom the MEG laterality map was mainly negative, suggesting more MEG connections with the contralateral region, together with low k-hubness within the IED generator (Fig.1b). Fig.1c is presenting a non-seizure-free left frontal patient, for whom the MEG laterality map was positive in several regions together with high k-hubness within the IED generator (Fig.1d), suggesting a less isolated and more widespread epileptic network. For MEG resting state, averaged laterality values were significantly larger for non-seizure-free compared to seizure-free patients (p<0.001) (Fig.2a), whereas fMRI resting state exhibited significantly larger k-hubness within the IED generator for non-seizure-free when compared to seizure-free patients (p<0.001) (Fig.2b). Conclusions: These preliminary findings suggest that resting state MEG connectivity and fMRI k-hubness may provide relevant information to predict surgical outcome, even in absence of IEDs. Low k-hubness within the IED generator and negative MEG laterality map, suggesting an isolated epileptic network, may indicate good surgical outcome. On the other hand, high k-hubness within the IED generator and positive MEG laterality map, suggesting a widespread and infiltrating epileptic network, may be associated with poor surgical outcome. Funding: FRQS and Savoy Foundation postdoctoral fellowships, CIHR MOP 133619 and 93614, NSERC Discovery grant, and FRQNT Research team grant.