Abstracts

Rationale: Seizures can have onset, middle and terminal stages with distinctive dynamics (Schiff et al., Neuroimage 2005), but are usually treated as monolithic events. In the rodent hippocampal tetanus toxin model of temporal lobe epilepsy, seizures have been described qualitatively (Finnerty et al., J Neurophysiol 2000) as having up to five distinct stages characterized by the frequency range of high amplitude field postsynaptic potential (FPSP) 'spikes'. The fourth stage, with ~9-16Hz activity, is strongly correlated with secondary generalization, identified by motor behavior in the form of rearing followed by myoclonic convulsions. We are employing this model to devise closed-loop seizure control algorithms using intracranial low frequency electrical field stimulation. Assessment of treatment effect requires detailed knowledge of the seizure dynamics and associated motor behavior. We tested the ability to describe seizure progression using a Markov source or hidden Markov model (HMM), in which the observed time series (EEG) reflects transitions between discrete underlying states.Methods: All procedures were carried out with IACUC approval. Tetanus toxin (5ng/ul) was injected into the right ventral hippocampus of adult male rats (n=5). Cortical and hippocampal EEG were continuously recorded along with digital video for 3-16 days. Seizures started 2-3 days after implantation and reached a peak rate of ~30/day after about a week. Associated motor activity was quantified from MEMS accelerometers in the head-stage. Over 3-16 days of monitoring, there were 20-254 spontaneous seizures per rat at a mean rate of 1.75±0.5/h. A HMM is defined by a set of unobservable or 'hidden' states with fixed probabilities, transition probabilities between states, and an observed random variable uniquely distributed for each state. With only the number of states prespecified, a HMM was trained using the Baum-Welch algorithm on the FPSP spike frequency time series (1/4s bins) derived from a small sample of seizure segments including baseline (10/rat). For each of the remaining seizures, this HMM was used to determine the most likely seizure state sequence given the observed FPSP spike time series.Results: We used a HMM based on FPSP spike frequency for automatic demarcation of seizure stages. The sequence of stages up to generalization is consistent with that described in the literature. Notably, a state identified by the model and characterized by 9-16 Hz discharges often preceded clonic behavior.Conclusions: This is the first study of the tetanus toxin model of temporal lobe epilepsy using continuous, long-term intracranial recordings with a high sampling frequency (2kHz). The temporal progression of a seizure in terms of transitions through discrete stages with distinctive FPSP spike rates was well characterized by a Markov source model. Because seizure patterns vary over time, between seizures, and between animals, such a state transition map will be useful for statistical comparison especially when evaluating different treatment protocols. (Support: NIH R01EB001507, K02MH01493 and R01MH50006.)