Abstracts

Rationale: The aim was to define sensitivity and specificity of Blood Oxygen Level Dependent (BOLD) responses to interictal epileptic spikes during EEG-correlated functional MRI (EEG-fMRI) for detecting the ictal onset zone (IOZ). Methods: Twenty-seven consecutive adults with refractory focal epilepsy and a well delineated IOZ after a full presurgical evaluation were studied with EEG-fMRI. After standard EEG artifact removal, concordant interictal spikes could be marked visually in 10 patients. The canonical hemodynamic response to spikes was modeled using an event-related design. We delineated the IOZ as the region of subtraction ictal SPECT co-registered to MRI (SISCOM) hyperperfusion (Z>1.5) within the presumed epileptogenic zone or resection zone in patients who underwent epilepsysurgery. Areas of BOLD activations or deactivations overlapping the IOZ in patients were considered as true positives and if no overlap was found, it was considered as a false negative. To determine true negative and false positive fractions, we included 10 age- and gender-matched healthy case-controls. Each control underwent EEG-fMRI using the same parameters as the corresponding patient. The interictal spike event-related regressor of the patient was used in the design matrix of the case-control normal subject. Suprathreshold BOLD activations or deactivations in any part of the brain in healthy controls were considered as false positives, absence of BOLD responses as true negatives. Sensitivity and specificity for IOZ detection were calculated for different statistical Z-thresholds and cluster sizes (1, 5 and 50 voxels) and illustrated in ROC-curves. The commonly used statistical thresholds uncorrected p < 0.001 and family wise error (FWE) corrected p < 0.05 correspond to a Z-threshold of 3.1 respectively Z in the 4.8 - 5.1 interval (the exact Z-value depends on the search volume and the smoothness of the data). Results: Figure 1 illustrates the ROC-curve corresponding to a minimal cluster size of 5 voxels. When using an uncorrected p < 0.001 (combined with a 5 voxel cluster extent), sensitivity is 100% and specificity 10%. In contrast, when a significance level of FWE corrected p < 0.05 was used, the sensitivity dropped to 70% but the specificity increased to 100%. The latter results were the same when using no additional cluster size or when using a minimal cluster extent of 50 voxels. Interestingly, if only BOLD activations were considered with a minimum cluster extent of 5 voxels, the same sensitivity and specificity was reached at FWE corrected p < 0.05. Conclusions: This study reports that BOLD changes to interictal spikes during EEG-fMRI, corresponding to a voxel based significance level of FWE corrected p < 0.05, resulted in 70% sensitivity and 100% specificity for localizing the IOZ in 10 refractory focal epilepsy patients. Sources of funding: IWT-TBM grant (080658)