Abstracts

Rationale: The main objective of this study was to develop a computational method for improving the localization of epileptic seizure foci in the brain based on multi channel EEG recordings. Unlike other EEG source localization methods such as LORETA, we were looking for a solution that specifically localize epileptic activity relative to the background of normal EEG activity and it does at a higher than electrode grid resolution. Methods: Our basic assumption was that seizures manifest in EEG as large amplitude synchronized activity. We further assumed that this synchronized activity can be distinguished from the background EEG activity as zero phase lag (<2 ms) synchrony, as oppose to non-zero phase coherency, typical of traveling or spreading waves in the normal EEG. Therefore, we computed the point-to-point correlation of simultaneously sampled EEG signal between all 24 electrode pairs, placed according to the 10/20 system. The correlation-matrix contained all pair-wise correlation coefficients between the different channels. Theoretically, when the source of seizure is located in the cerebral cortex near the skull, it is detected as a localized synchrony between adjacent channels that is well distinguishable from other channels. In contrast, when the source is located in the deep structures the seizure may project to multiple electrode locations and manifest as synchrony between distant electrodes. In either case, we can identify those channels based on the magnitude of shared synchronized activity. The group of synchronized electrodes is identified as a cluster in the correlation-matrix. The next step was to interpolate between the synchronized channels to obtain a source location beyond electrode grid precision. By limiting the interpolation to the pre-selected channels the accuracy of the interpolation method improves significantly relative to other methods that take the activity over the entire skull into an account. Results: We tested the algorithm on data obtained from two patients diagnosed with partial epilepsy. The EEG data consisted of identified epochs of seizure free EEG (wake state and sleep) and EEG during seizure activity, several minutes each. When the seizure-free recordings were compared with EEG records of seizures we found that the seizure prone areas are clearly distinguishable from non-seizure prone areas even during seizure-free activity based on synchrony. The localization provided by the correlation method was confirmed by MRI as well as by experts' judgments. Conclusions: Based on the results from two patients under different conditions we concluded that the source localization of epileptic focus based on synchrony is not only feasible but may contribute to the practice of EEG diagnosis in two ways: (1) enables to localize the source at a higher than grid resolution and (2) it reveals the pathological region even during seizure-free EEG activity. The results suggest that it is worth to pursue this direction of research in order to develop more sensitive diagnostic methods for partial epilepsy.