Abstracts

Rationale: Deep brain stimulation (DBS) as a means for seizure control has recently gained much attention. In parallel, methods from nonlinear dynamics have been employed in seizure prediction algorithms; however, merging of the two ideas in closed loop seizure control is at its infancy. The goal of this study was to close the loop between seizure prediction and deep brain stimulation. An automated seizure prediction and control system, that utilized measures of chaos of intracranial EEG (Lyapunov exponents) coupled with DBS, was designed and implemented in chronically epileptic rats. We compared the performance of the automated system in controlling seizures to periodic DBS with identical stimulation parameters.Methods: Six adult male Sprague-Dawley rats were used in this study. Each rat was made chronically epileptic following a prolonged episode of status epilepticus using the lithium-pilocarpine model. Rats were implanted with an array of six Tungsten microwires along with two twisted bipolar stimulating electrodes targeted at the centromedial thalamic nucleus. Five weeks of testing were performed on each rat with EEG recorded continuously with online real-time nonlinear dynamical analysis. During Week 1, a range of frequency and amplitude stimulation parameters were tested for effects on EEG dynamics. During Week 2, optimization of the seizure prediction algorithm in terms of its sensitivity and specificity was performed. During Week 3, automated closed loop control was employed for each rat in which one-minute biphasic DBS trains were delivered to the thalamus at seizure warning time points determined by the prediction algorithm. Stimulation was off during Week 4. During Week 5, one-minute periodic biphasic DBS trains with identical stimulation parameters and same mean period as the mean inter-warning period from Week 3 were utilized. Seizure frequency per week was retrospectively estimated per rat in the study.Results: Seizure predictability during Week 2 yielded sensitivities and specificities larger than 60% and less than one false prediction every five hours respectively across rats. The automated control scheme reduced seizure frequency by >50% in 33% of the rats tested, with no significant effect in the rest. The periodic control scheme exacerbated seizures in 50% of the rats, and only reduced the seizure frequency in one. Due to the use of identical stimulation parameters, these results show the importance of the timing of stimulation for seizure control. Furthermore, a retrospective dynamical analysis of the EEG in Weeks 3 and 5 showed that success in seizure control was highly correlated with decreases in global brain entrainment, while a lack of seizure control showed no such correlation.Conclusions: Closing the loop between seizure prediction algorithms using nonlinear dynamics and deep brain stimulation is possible and, although very early in development, may prove more useful as a seizure control device than the simple periodic DBS. (This research was supported by the US Epilepsy Research Foundation and the Ali Paris Fund for LKS Research and Education, and the US grant NSF 0601740)