Abstracts

Rationale: Ample clinical and experimental evidence indicates that there is a link between seizure and sleep dynamics (Dinner 2002). There is now a focused effort in the development and clinical implementation of neural prosthetics for the prediction and control of epilepsy. However, seizure prediction efforts have largely ignored the known and confounding effect of state-of-vigilance (SOV) on seizure states; likely due to the difficulty of classifying SOV using measurements obtained from implanted devices. Our aim is to develop an observer/predictor system, suitable for implantable devices that will track and predict sleep-wake cycles as well as the underlying dynamics to improve seizure prediction. Methods: We have implemented a chronic, behaving animal model of limbic epilepsy in the free-moving rodent. We continuously record and monitor depth and cortical EEG as well as movement information obtained from head-mounted accelerometers which, in the past, we ve demonstrated can be used for accurate SOV classification. We have now implemented biologically-based models of sleep dynamics that embody the cellular networks involved, and have applied nonlinear data assimilation methods (Schiff 2010) to use these models as observers and predictors of state of vigilance. Results: Here we demonstrate that this methodology works well in the case of assimilation of observations from one computational model into an observer model based off the same or different model (Diniz Behn 2007, Tamakawa 2006). We also present results from assimilation of rodent data into one of the models. This assimilation allows us to reconstruct the physiological model using data obtained from our experimental rodents. Thus, we gain access to previously hidden variables that describe sleep dynamics. Finally, we demonstrate that we can use the existing algorithm to make short-term predictions of sleep state and sleep state transitions. Conclusions: It is becoming increasingly clear that in order to predict and control seizure dynamics, we must first be able to grasp the dynamics of sleep and sleep-state transitions. Although several biologically inspired models of sleep have been recently published, their implementation has been limited to computer simulations. Using our data assimilation approach, we are able to combine the advances in computational modeling of sleep together with measurements from our experimental rodents, to access the dynamics of the underlying network which govern both sleep and seizure state transitions. Thus, our work bridges experimental and computational techniques to investigate a crucial missing link which will give us insight into the dynamics of seizures and may drastically improve seizure prediction. Dinner D.S. Journal of Clinical Neurophysiology 19(6): 504-513, 2002. Schiff S.J. Phil. Trans. R. Soc. A 368: 2269-2308, 2010. Diniz Behn C., Brown E., Scammel T., and Kopell N. J Neurophysiol 97:3828- 3840, 2007. Tamakawa Y., Karashima A., Koyamma Y., Katayama N., and Nakao M. J Neurophysiol 95: 2055-2069, 2006.