Abstracts

Rationale: Despite their limitations, routine scalp EEGs are non-invasive, inexpensive, and capable of sampling the entire brain, thus providing the best way to approximately localize an epileptic focus. Related studies of spatial seizure propagation have primarily focused on the dipole source model, as the most adequate representation of an epileptic focus. However, as attractive as this model may be, it does not accurately describe seizure propagation, which is considerably more complex than a series of moving dipoles, and often results in inaccurate localization of the epileptic focus. Thus, more appropriate data-driven and clinically relevant models, that account for the complexity of both the source and the anisotropy of the brain, are needed. Methods: We sequentially estimated source dynamics and propagation patterns of 36 focal seizures from 7 patients with heterogeneous pathologies, recorded using a standard 20-channel scalp EEG system. We initially localized the focus using a signal back-propagation method [Fink, M., 1993] and based on that position, we estimated seizure propagation speed and orientation using a coordinate transformation approach. The initial localization was then corrected for source motion. Based on the new position, we estimated epileptic source strength modulated by the propagation pattern on the surface of the brain. Results: Seizure propagation speed was estimated to be in the range 40-85 mm/s. For patients with multiple focal seizures localized very close to each other, estimated directions of propagation were also similar. The propagation patterns of only 6 analyzed seizures (17\%) were adequately represented by an oriented moving dipole. All other patterns were best described by weighted superpositions of higher order multipoles, e.g, combinations of quadrupoles in the case of multi-focal seizures, or superimposed quadrupoles and dipoles, previously theoretically suggested [Nunez, P., et al, 2006] but here estimated directly from the scalp EEG. Model selection was based on the modified AIC criterion which penalized higher order models. It was also possible to estimate a single discrete focus for a few partial seizures with complex propagation patterns that were visually classified as non-localizable. Conclusions: Scalp EEG recordings of focal seizures can be used to directly estimate seizure propagation patterns and source dynamic parameters directly from the signals, with minimum a priori assumptions regarding propagation characteristics. The resulting models adequately reflect the actual complexity of a seizure, and although the entire brain is sparsely sampled, it is still possible to estimate enough propagation detail using scalp data to select such models.