Abstracts

Rationale: We have previously shown that the time course of the fMRI signal changes associated with interictal spikes is generally well represented by the standard, ‘canonical’, model derived from activation studies, with additional changes in a few cases, hypothesised to reflect artefacts or epileptiform activity not visible on scalp EEG [Lemieux et al 2007]. In addition, we have also previously shown that data-driven fMRI analysis, namely Independent Components Analysis (ICA), can reveal haemodynamic patterns that match those predicted from scalp EEG plus additional components suspected of representing epileptic activity [Rodionov et al 2007]. In this work, we have examined the temporal relationship between those additional components and epileptiform activity inferred from scalp EEG.Methods: 8 patients with well-characterised focal epilepsy were studied using simultaneous EEG-fMRI at 1.5T in the resting state (TR=3s). The resulting fMRI data was analysed in two ways: EEG-derived General Linear Model (GLM) and data-driven. In the data-driven approach, an automated classification scheme was used to help identify ‘candidate’ components, i.e. potentially of epileptic origin, in addition to the components found to match the patterns revealed using the GLM and artefacts. The candidate components were identified based on their spatial proximity with the irritative zone. For each of candidate component, correlation as a function time-lag relative to the spike-derived model (lag range: +/- 10 TR) was plotted. A correlation threshold level of p<0.01 was applied.Results: Candidate components were identified in 7/8 cases. In 5 of those 7 cases, between 1 and 4 candidate components (total: 10) were identified that correlated significantly with the spike-derived model as a function of time lag. The phase (lag) difference at the peak correlation value ranged between –TR and +2TR (−1TR < lag < +1TR for 8/10 components; others: 0 < lag <= +2TR), with a mean of +0.3s.Conclusions: In this series of cases, we have revealed BOLD signals that were time locked with the activity recorded on the scalp, but with varying phase differences. These differences were generally small and may represent normal variation in the hemodynamic response, but could also represent pathological activity not reflected on scalp EEG. This highlights the potential of combined data- and model-driven analyses. References: Lemieux et al., Human Brain Mapping, 2007. Rodionov et al., Proc OHBM 2007. Acknowledgements: Work funded by the Medical Research Council and the Wellcome Trust.