Abstracts

RATIONALE: In previous work we have shown that the multichannel autoregressive (AR) model can be used to describe the linear portion of the dynamics of the EEG signal during both ictal and interictal periods. We have also demonstrated that time-frequency analysis using the Matching Pursuit (MP) method provides a detailed description of the dynamics of the seizure, even during the most rapidly changing periods. In these Matching Pursuit analyses, however, the low amplitude components tend to be obscured by dominant rhythmic activity of greater energy. We now combine both approaches to take advantage of the strengths of each method. METHODS: The multichannel autoregressive model was applied to analyze 5 second, quasi-stationary epochs of preictal, ictal, and postictal portions of the intracranial EEG recordings of complex partial seizures from patients undergoing monitoring with subdural grid arrays. The residuals remaining after fitting the AR model were then analyzed using the MP time-frequency analysis. Two to three seizures were analyzed from each of five patients. RESULTS: The MP decompositions of the residuals after fitting the AR model show greater detail of the time-frequency structure of the EEG signal than if only the MP method is applied. The previously obscured low amplitude transient components can be more readily identified. CONCLUSIONS: Applying the AR model to EEG signals allows analysis of residuals with the MP method to yield enhanced time-frequency decompositions of low amplitude activity, providing for better descriptions of seizure dynamics. This low amplitude activity (often of high frequency) may be important, particularly in early seizure evolution. Improved analysis of this activity can provide additional information aiding detection and localization of seizure onset. Supported by NIH grant NS 33732