Abstracts

Rationale: Combined fMRI-EEG is being developed for localizing areas of the brain involved in seizure initiation and propagation. A growing literature describes the immediate hemodynamic response seen in the BOLD signal following epileptiform discharges and brief generalized seizures. A few reports describe slower changes in the BOLD signal preceding and following a seizure. We postulate that there may be slow fluctuations in the subthreshold activity of seizure networks which may be undetectable by standard fMRI-EEG approaches. We describe an approach for the identification of such networks adapted from recent fMRI studies of the 'default mode' or resting state network and demonstrate its use.Methods: We performed resting state fMRI-EEG on a patient with intractable left temporal lobe epilepsy confirmed by intracranial EEG. Data were collected on a 1.5T GE scanner with simultaneous 64 channel scalp EEG recording. We applied probabilistic independent component analysis to each run of the fMRI data using FSL and identified components with low frequency fluctuations. From these components, regions of interest (ROIs) were constructed from the subset of voxels present in at least one component in every run. ROI time courses were compared. For each ROI, functional connectivity was assessed across the whole brain by performing a regression using the ROI's time course as the effect of interest, with motion parameters and global signal included as regressors of no interest.Results: Areas present in low frequency components from each run included the posterior cingulate (PC) and bilateral posterior parietal cortex in concordance with standard descriptions of the resting state network (RSN). Bilateral lateral temporal regions also exhibited low frequency BOLD signal fluctuations. The ROI in the left lateral temporal lobe (LT) was noted to correspond to this patient's presumed seizure focus. In one run, a typical seizure was captured and confirmed by simultaneous EEG recording. The LT ROI's time course was highly correlated with the right temporal lobe and the RSN during the interictal runs. This correlation was significantly reduced in the time period surrounding her seizure.Conclusions: This method identified distinct brain regions with correlated low frequency fluctuations, including areas previously identified as part of the 'default mode' resting state network. The method also identified temporal areas which may be involved in this patient's seizure network. We believe that this approach to analyzing resting state fMRI data may be broadly applicable to the identification of seizure networks in the absence of clinical or electrographic events.