Abstracts

Although both EEG and structural MRI provide diagnostic information for epilepsy, these modalities are not expressly integrated in routine clinical practice. In part, this is because the spatial resolution for solving [italic]global[/italic] EEG inverse problems (e.g., via dipole localization or cortical current density) depends both on modeling assumptions and on variable EEG data composition, and thus is not quantifiable with high confidence. Regional activity estimation (REGAE) is a new method for solving a [italic]local[/italic] inverse problem, namely, how to detect activity emitted from a specified brain region of interest (ROI) while unknown activity simultaneously is emitted from outside the ROI ([italic]NeuroImage[/italic] 13(6):S219). Signal detectability, as quantified by area under the ROC curve (AUROC), depends on both the size and the situation of the ROI, and can be estimated before recording EEG based on geometry from MRI, electrode locations, and conservative assumptions. REGAE localization requires scanning numerous ROIs that cover brain gray matter. This in turn requires calibrated setup of one estimator per ROI. Until recently, setup time was too slow for practical scanning purposes, except by making a crude approximation to AUROC (an approach presented at AAN 2002, [italic]Neurology[/italic] 58(Suppl 3):A141). Using a new method for fast calculation of AUROC, this study aims (a) to demostrate the practical feasibility of REGAE scanning via application to an actual EEG+MRI data set, and (b) to quantify the setup time improvement.
Whole-head structural MRI for a patient with intractable TLE was parcellated as white matter, gray matter, CSF, bone, and skin tissue types. A full source space model was made by placing three orthogonal dipole elements at about 16,000 locations inside the gray matter region. A tetrahedral mesh finite element model (FEM) was made using all regions, and a transfer matrix to 57 digitized electrode locations was computed via an efficient FEM solver. About 100 ROIs covering gray matter, with Gaussian taper, were adjusted in size to achieve calibrated REGAE detectability of AUROC=0.75 (using 50 random simulations per ROI). This process was timed and performed twice, using source space (previous method) versus signal space (new method) simulations. The resulting estimators were applied to 24 interictal spike segments and 2 early ictal discharges.
Interictal spikes and early ictal discharges localized predominantly to left temporal lobe regions. Overall (i.e., including common routines), the new method was about 15 times faster for calibrated setup than the previous direct method (39 s/ROI vs. 579 s/ROI using a 900 MHz processor). In the critical routine, the new method was more than 100 times faster.
REGAE scanning is, theoretically and practically, a feasible alternative for localizing epileptiform EEG sources. Direct comparison with dipole localization and cortical current density methods remains to be studied.
[Supported by: NIH grant 2 R44 MH64343-02]