Abstracts

Rationale: The spatial resolution of fMRI allows the accurate localization of functional regions in the brain. Simultaneously recorded EEG-fMRI can be used to localize hemodynamic changes corresponding to interictal epileptic discharges (IEDs) observed on EEG. However, as interictal activity is not always visible on scalp EEG due to the presence of artifacts, being generated by deep brain sources, or the limited time of the acquisition, there is an increasing demand for techniques capable of localizing epileptic activity based solely on the fMRI time series. Methods: Our study includes 20 refractory focal epilepsy patients who underwent full presurgical evaluation including ictal single photon emission computed tomography (SPECT). The ictal onset zone (IOZ) was determined by a neurologist as the overlap between the subtraction ictal SPECT co-registered to MRI (SISCOM) hyperperfusion map (Z>1.5) and the effective or intended resection zone. Functional images were acquired using a 3T MR scanner with a whole brain single-shot T2* gradient-echo Echo Planar Imaging sequence (TE = 33ms, TR = 2.5s/2.25s/2.2s, voxel size 2.6/3/2.6 mm). The images were realigned, slice-time corrected, normalized to MNI space and spatially smoothed with an isotropic Gaussian kernel of 6 mm FWHM using SPM8 software (Wellcome Department of Cognitive Neurology). Scalp EEG was simultaneously recorded with a 64 or 32 channel MR-compatible EEG cap (Brain Products) or a 24 channel electrode set (Yves EEG solutions inc). The EEG signals were amplified (BrainAmp amplifier, sampling rate 5000 Hz, resolution 0.5 µV) and transmitted outside the scanning room. The EEG was filtered offline between 1-50Hz, gradient and pulse artifacts were removed. 2 patient groups were defined based on whether clinically concordant interictal spikes were seen on the EEG or not. Group 1 contained 10 patients with clinically concordant spikes, and Group 2 contained 10 patients where no such spikes were seen. Spatial independent component analysis (ICA) was performed on the preprocessed fMRI time series to obtain spatially independent activation maps. The component activation maps were converted to z-scores, thresholded at z>3 and their overlap with the IOZ was calculated. GLM-based IED-correlated fMRI activation maps of Group 1 patients were also obtained using SPM8, and the overlap between these activation maps and the IOZ was calculated. Results: Both the GLM-based activation map and at least one ICA-based component activation map overlapped with the IOZ in all patients in Group 1 (Figure 1). We show that the ICA-based maps overlap more with IOZ in 8 out of 10 patients than the GLM-based maps do. Moreover, at least one ICA-based component activation map overlapped with the IOZ in all patients in Group 2 as well (Figure 2). Conclusions: ICA is capable of deriving activation maps from the fMRI time series which overlap more precisely with the IOZ than the GLM-based activation maps. Furthermore, ICA also finds such activation maps in patients, where the traditional GLM-based analysis can not be performed due to the lack of clinically concordant IEDs on the EEG.