Abstracts

Rationale: Discordant EEG and imaging findings, in refractory focal epilepsy, can lead to great difficulty in identifying the epileptic focus. To address this common clinical dilemma we considered one specific scenario: focal epilepsy due to a structural lesion, where the interictal scalp EEG discharge is spatially discordant. Using simultaneous EEG-fMRI data and stochastic dynamic causal modelling (sDCM) we asked (i) what is the network relationship between the structural lesion and the region corresponding to the interictal discharges, and (ii) what is the direction of network driving and/or modulation during interictal discharges? Methods: We identified three patients with a solitary focal epileptogenic lesion, discordant scalp EEG, and an EEG-fMRI study showing significant peri-lesional activation: (1) 23M with a cavernous hemangioma of the right precuneus but diffuse right frontal sharp-slow and low-amplitude fast discharges. (2) 10F with a right parahippocampal DNET but diffuse slow-spike-wave and paroxysmal fast activity. (3) 38M with right temporal perinatal traumatic/ischemic changes, but diffuse paroxysmal fast, bifrontal spike-wave and rare right temporal discharges.For each patient, one hour of EEG-fMRI was recorded (3T MRI, echo-planar imaging, TR 3.0/3.2s, TE 30/40ms, voxels 3.0/3.4mm isotropic) and processed as per our standard protocol (Flanagan et al, Clin. Neurophys. 2014;125:21–31). General linear modelling was performed in SPM12 using the timing of all interictal discharges, convolved with the hemodynamic response plus time and dispersion derivatives. From the resulting activation map, we identified one region adjacent to the lesion and one region within the frontal lobe, and extracted the first eigenvariate from each. The DCM model space comprised all two-region fully connected stochastic models, with/without an input at one node, and with/without modulation of between-node connectivity, driven by the timing of interictal discharges (9 models per participant). Bayesian model selection with fixed-effects analysis was used to identify the most plausible model for each patient. Results: Multiple areas of significant discharge-related fMRI activation were found in each patient, both adjacent to the structural lesion (peak t-scores: 5.3 at precuneus, 15.1 at posterior temporal cortex, 3.1 at inferior temporal gyrus) and throughout the frontal lobes corresponding frontal-predominant EEG discharges (peak t-scores: 8.2 at superior frontal gyrus, 13.0 at right frontal pole, 9.0 at anterior cingulate). The global peak t-score did not identify the lesional area.Model selection identified the same model structure in every patient. The most plausible model structure showed reciprocal excitatory connections between the lesion and the frontal-lobe node (strength 0.08 to 0.36Hz, with probability >0.99) and self-inhibition at each node (strength -0.05 to -0.22Hz). For each patient, interictal epileptiform discharges corresponded to a driving input at the lesion (strength 0.009 to 0.017Hz, probability >0.99), and a modulating influence on connectivity from the lesion to the frontal-lobe node. Conclusions: EEG-fMRI can help reconcile discordant data in lesional focal epilepsy, by demonstrating a network of interictal epileptiform activation. Causal modelling in these cases indicated that network fluctuations during discharges were driven from the structural lesion. Principled hypothesis-driven application of stochastic DCM may be a useful approach to identifying the driving node in other apparently discordant cases of focal epilepsy. Funding: Australian NHMRC Programme Grant #628952Victorian Government Operational Infrastructure Support Programme