The possibility of acquiring fMRI at ultra-high field (UHF) offers the opportunity to greatly enhance BOLD contrast sensitivity and to subsequently improve the spatial resolution or decrease the number of events required to obtain a significant effect^1,2. Furthermore, the intravascular signal contribution from draining veins decreases with magnetic field strength³ allowing a more accurate localization and a better understanding of negative BOLD responses⁴. In patients who are candidates for epilepsy surgery, this allows better characterization of epileptic networks using simultaneous EEG recordings of interictal epileptiform discharges (IEDs) as well as functional eloquent cortex. However, EEG acquisition during fMRI, especially at UHF, suffers from various artifacts compromising data quality. Our aim is to demonstrate that eloquent cortex and epileptic-related hemodynamic changes can be safely and reliably detected using simultaneous EEG recording at 7T for clinical evaluation of epileptic patients.

Nine patients with refractory lesional epilepsy were selected to have a simultaneous EEG-fMRI recording at 7T. According to the localization of the lesion, one patient also performed a language fMRI and one patient a motor fMRI for mapping of functional eloquent cortex to be preserved during surgery. All patients gave written informed consent and this study was approved by the local ethics committee.

Acquisition. Simultaneous EEG-fMRI acquisitions were performed in a 7T head-only scanner (Siemens Magnetom, Erlangen, Germany) equipped with an 8-channel transmit/receive head coil (Rapid Biomedical, Rimpar, Germany) during 20 minutes at rest with eyes closed. All functional images were acquired using a T₂^*-weighed GE-EPI sequence (TR=2000ms, TE=25ms, α=78°, voxel size=1.5x1.5x1.5mm³, 32 axial slices with 1.5mm interslice gaps). EEG was acquired at 5kHz using two MR-compatible amplifiers (Brain Products, Gilching, Germany) synchronized with the MR clock. An optimized setup, with a customized 64-channel cap (EasyCap, Herrsching, Germany) connected to the amplifiers via two ultra-short bundled cables, was used⁵. A 0.6x0.6x0.6mm³ resolution MP2RAGE⁶ and a 0.4x0.4x1mm³ resolution susceptibility-weighted imaging sequences were acquired for structural localization purposes using a 32-channel head coil (Nova Medical, MA, USA).

EEG preprocessing. Gradient artifacts were corrected using a hybrid mean and median artifact template subtraction. The EEG was then down-sampled to 1kHz and pulse artifacts were detected using an estimated ballistocardiogram extracted from a subset of temporal electrodes⁷. To deal with the high variability between successive artifacts worsened by the magnetic field intensity, we used a non local mean (NLM) averaging technique.

fMRI preprocessing. Functional MRI images were motion-corrected, co-registered onto the structural images and spatially smoothed with an isotropic Gaussian kernel of 4 mm full-width at half-maximum. Functional time-series were analyzed voxel by voxel with a general linear model (SPM8, Wellcome Trust Centre for Neurosciences, UCL).

Spike-related and topography-related analyses. An experienced neurophysiologist manually detected IEDs in the corrected EEG. If no IEDs were recorded during the EEG-fMRI, a patient-specific epileptic topographic map was built by averaging IEDs detected in the clinical EEG acquired outside MRI. The presence of this epileptic topographic map in the intra-MRI EEG was quantified by means of correlation-based fitting. The IEDs timing or the time course of the topography-based correlation was then convolved with the canonical hemodynamic response function and used as a regressor for the fMRI analysis⁸. IED-related or patient-specific topography-related hemodynamic changes were detected using a t-test (p<0.001, 20 voxels extent threshold).

Mapping of functional eloquent cortex. For motor and language task⁹, the block design was convolved with the canonical hemodynamic response function and activated areas were detected (p<0.05, FWE correction).

After optimized gradient and pulse artifact removal, IEDs were successfully detected on the corrected EEG (Fig. 1). Topography-related hemodynamic changes were obtained and their localizations were comparable with the EEG-fMRI at 3T (Siemens, Prisma) using a high-resolution 256 channels EEG (EGI, Eugene, OR)(Fig. 2) attesting the reproducibility of EEG-fMRI at different fields and the excellent quality of corrected EEG. Language functions were successfully located in one patient (Fig. 3) whereas primary motor cortex localization was complicated by B₁ inhomogeneities in fronto-central regions using this EEG cap configuration (Fig.4).Epileptic network and functional eloquent cortex localizations using an optimized EEG-fMRI setup and appropriate artifact removal algorithms is feasible at UHF. The EEG quality allows noise-sensitive analyses such as EEG topography spatial correlation, and yielding precise localization of correlated hemodynamic changes. B₁⁺ inhomogeneities are highly dependent of the layout of the EEG cap, and could potentially be mitigated in regions of interest by adapting the layout accordingly. These results open new perspectives to better characterize epileptic networks at higher field with greater spatial resolution and better BOLD sensitivity than at 3T.