There is overwhelming clinical evidence of the interactions between sleep and epilepsy, but the literature on seizure prediction has largely disregarded the effects of state of vigilance (SOV) on seizure generation. Existing seizure prediction algorithms (SPAs) are not accurate enough for use in epilepsy care due to high false prediction rates. Recent SPA assessments reveal that false seizure predictions are more likely during certain stages of sleep, arousal, or times of day. These observations suggest that (1) SPAs confuse certain normal states of vigilance with truly preictal states (should they exist), or (2) SPA measures merely track transitions to normal states of vigilance that are more susceptible to seizure. In order to investigate the mechanism of seizure generation from different brain states, methods are needed to accurately track changes in SOV from continuous physiological measurements. We have developed SOV discrimination methods that utilize acceleration measurements from head-mounted accelerometers to significantly improve on methods that rely on EEG alone (Sunderam et al., J. Neurosci. Methods, 2007). We propose to develop a methodology for tracking SOV in real time by (1) Applying a combination of supervised Fisher discrimination and unsupervised Bayesian discrimination to establish ground truth labels for SOV;and (2) Constructing state-space models (hidden Markov models and switching Kalman filters) to track the noisy, nonstationary SOV dynamics. In the current project, we will test the ability of these methods to perform accurate, real-time brain state tracking, using chronic EEG and kinematic measurements from a rodent model of temporal lobe epilepsy as well as normal controls. The output will be used to quantify the relative probability of seizure onset from different states of vigilance. The long term goal of this project is to track state of vigilance in epilepsy patients and investigate behavioral states with a high probability of seizure generation. The formulation of a behavioral index of seizure susceptibility will help improve the performance of SPAs, and serve as the basis for state-dependent seizure detection, anticipation and preventive control using low frequency electric field modulation. It will also be useful for testing whether candidate SPA measures truly discriminate between the preseizure period and normal behavioral states.
Our long term goal is to develop, for use in implantable seizure control devices, methods for real-time tracking of state of vigilance in epilepsy patients. Certain states, particularly related to sleep and arousal, are more likely to lead to seizure. The availability of a behavioral index of seizure susceptibility will lead to improved performance of existing seizure prediction and control algorithms.
| Yaghouby, Farid; Donohue, Kevin D; O'Hara, Bruce F et al. (2016) Noninvasive dissection of mouse sleep using a piezoelectric motion sensor. J Neurosci Methods 259:90-100 |
| Yaghouby, Farid; Sunderam, Sridhar (2015) Quasi-supervised scoring of human sleep in polysomnograms using augmented input variables. Comput Biol Med 59:54-63 |
| Sedigh-Sarvestani, Madineh; Thuku, Godfrey I; Sunderam, Sridhar et al. (2014) Rapid eye movement sleep and hippocampal theta oscillations precede seizure onset in the tetanus toxin model of temporal lobe epilepsy. J Neurosci 34:1105-14 |
| Yaghouby, Farid; Modur, Pradeep; Sunderam, Sridhar (2014) Naive scoring of human sleep based on a hidden Markov model of the electroencephalogram. Conf Proc IEEE Eng Med Biol Soc 2014:5028-31 |
| Sunderam, Sridhar; Gluckman, Bruce; Reato, Davide et al. (2010) Toward rational design of electrical stimulation strategies for epilepsy control. Epilepsy Behav 17:6-22 |
