Abstracts

Rationale: Localizing epileptic activity independent of EEG is especially useful when the shape of hemodynamic response function (HRF) may vary across Interictal Epileptic Discharges (IEDs), or scalp EEG can detect just part of the IEDs, or even nothing. In this research, a data-driven method, independent component analysis (ICA), integrated with general linear model (GLM), was used to localize the epileptic activity in patients with left temporal lobe epilepsy. In this study, based on previous findings that there are deactivations in default mode network (DMN) in temporal lobe epilepsy (TLE) simultaneous with IEDs [1], we hypothesized that the time courses of IED-related activated voxels should be anti-correlated with the time courses of DMN-related components. Methods: Five patients with left mesial temporal sclerosis or hippocampal structural abnormality who were part of a larger investigation [2] were the subjects of this study. All underwent left selective hippocampectomy and became free of disabling epileptic seizures. Subjects were scanned on a 3.0T MRI scanner with structural and fMRI scans acquired at rest with eyes closed (64x64, FOV=24cm, TE/TR=35/2000ms, 200 volumes). ICA was performed for each patient using GIFT [3], and the DMN-related component was determined with a DMN template defined in MNI space [4]. The time course corresponding to the DMN was used as a regressor in a GLM to localize the deactivations, which were expected especially in left hippocampus as epileptic activity for the patients with left TLE. Results: ICA produced one best-fit component corresponding to DMN for each subject with GIFT. In GLM analysis, we used the same threshold (p<0.005 uncorrected, minimum 10 voxels) for all subjects, except for subject 3 (p<0.01 uncorrected, minimum 10 voxels). We detected deactivations in both left and right hippocampus for subject 1; in left hippocampus for subject 2 (Fig. 1); in left hippocampus for subject 3; in both left and right insula for subject 4 (no deactivations in either left or right hippocampus); in both right and right hippocampus for subject 5. In addition, for subject 4, when we lowered the threshold (p<0.01 uncorrected, minimum 10 voxels), we did not detect deactivations in left hippocampus, but found the deactivations were adjacent to the left hippocampus. It should be noted that we also find some activations in other brain regions in addition to abovementioned activations, where relevance is unclear. Conclusions: The preliminary results show that we may localize epileptic activity in GLM with the time course corresponding to DMN component as a regressor of interest, and they are promising, although the method still needs to be developed to increase specificity. The method may be more useful in localizing epileptic activity than the method using scalp EEG, especially in the cases that the spikes cannot be easily detected. 1. Laufs H., et al., Human Brain Mapping 2007. 28: 1023-1032. 2. Morgan V., et al., Epilepsy Research 2007. 76: 22. 3. Calhoun V., et al., Human Brain Mapping 2001. 14: 140-151. 4. Greicius M., et al., Proc. Natl. Acad. Sci. USA 2004. 100:253-258.