Abstracts

RATIONALE: For patients with medically intractable epilepsy, there have been few effective alternatives to resective surgery. A strategy receiving increased attention is using interictal spike patterns and continuous EEG measurements to localize, predict, and ultimately control seizure activity via chemical or electrical control systems. This work compares results of six linear and nonlinear methods applied toward this purpose and comments on the merits and limitations of each. METHODS: We completed analysis of 12 intracranial seizures using power spectral density, cross-correlation, principal component analysis, phase, mutual prediction, correlation dimension, and wavelets. We devised a method for standardizing results across techniques for the purpose of comparison using a simple conversion to standard deviation units. Times of departure from the baseline mean and standard deviation were compared to a neurologist's judgement. RESULTS: Linear methods showed significant changes (deviations of greater than two standard deviations from the baseline mean) earlier than or at the same time as the nonlinear methods. Thus, for these data sets, the nonlinear methods offered no predictive advantage over the linear methods. Finally, no single method consistently succeeded in detecting changes leading to a seizure before the first changes noted by the neurologist. CONCLUSIONS: Does this mean that nonlinear tools are not useful for seizure analysis? There are many possible reasons why we did not detect a clear advantage of nonlinear over linear methods, despite known nonlinear behavior of neural systems. First, it is possible that the method of obtaining our original measurement excluded relevant information. We might improve our reconstructions by increasing our electrode density and recording frequency. Second, current nonlinear tools are designed to reveal structure from low-dimensional nonlinear systems. Faced with a truly complex system, they may fare worse than a linear model with a stochastic error term. Finally, due to the extreme variability between seizure types, tailoring combinations of these methods to particular seizure types may give optimal results.