Abstracts

We have previously reported seizure prediction in human temporal lobe epilepsy (TLE) using Short Term Lyapunov Exponent (STL[sub]max[/sub]) and Average Angular Frequency ([Omega]) with a sensitivity of 80% (Iasemidis et al, 2003). These results have prompted us to apply the same techniques to a rat chronic limbic epilepsy (CLE) model described by Lothman et al.(1990). The present study tests the hypothesis that similar dynamical changes exist in the CLE model. Thirty, 2-hr epoch data sets from 6 CLE rats (mean seizure duration 74[plusmn]20 sec), each containing a grade 5 seizure and continuous intracranial EEG beginning 1 hr before the seizure were analyzed. Teager energy is calculated iteratively every 5-second epoch for each electrode and a detection is determined at an adaptive fixed threshold. Coherence is calculated for every 4-sec non-overlapping epoch for each electrode pair. Ictal and interictal periods were compared to estimate the change in synchronization between channels at seizure onset. STL[sub]max[/sub] measures the dynamical chaoticity of a signal and [Omega] measures how fast the local state of a system changes on average and are estimated for every 10.24-second interval. T-index, a statistical measure is used to quantify the convergence or divergence of STL[sub]max[/sub] and [Omega] before and after a seizure. Data analysis showed an abrupt increase in the energy profile of each channel and a significant increase in coherence values in multiple frequency bands between the area of ictal onset and other sites during a seizure. Nonlinear analysis shows multiple transient drops in STL[sub]max[/sub] values during the pre-ictal period followed by a significant drop during the ictal period. [Omega] values show transient peaks during the pre-ictal period followed by a significant peak during the ictal period. The channel corresponding to the stimulated side of the hippocampus had a consistently lower value of STL[sub]max[/sub] and a higher value of [Omega] compared to other channels. A convergence among electrode sites was also observed in both STL[sub]max[/sub] and [Omega] values (low T-index) before a seizure. Results suggest that linear and non linear analysis can detect the dynamical change that precedes and accompanies seizures in rat CLE model. Furthermore, convergence in non-linear measures suggest that it is possible to identify the preictal transition. Thus, the rat CLE model may serve as a tool to further develop seizure prediction algorithms and implement real time closed-loop intervention techniques. (Supported by NIH grant R01EB002089, Children[apos]s Miracle Network, University of Florida Division of Sponsored Research and Department of Veterans Affairs)