Abstracts

To study the earliest stages of experimental epileptogenesis, methodologies that can resolve very low rates of interictal and ictal activity must be employed. Because the lowest resolvable rate is once per observation period, the best resolution entails continuous observation. This in turn necessitates efficient methods for unsupervised detection of ictal and interictal activity from very large datasets.
EEG data was accumulated from weeks of continuous recordings from epidural electrodes implanted in rats, transmitted via radiotelemetry (Data Sciences International), and digitized via routines written in Visual Basic (Microsoft). Status epilepticus was induced with intraperitoneal kainic acid 1 week following onset of telemetry, and rats were observed for subsequent spontaneous seizures. EEG transient detection strategies using both rectangular amplitude-duration windows and spectral characteristics were tested.
The amplitude, polarity, and morphology of epileptic EEG transients changed substantially during the first weeks of epileptogenesis, making it difficult to achieve reliable unsupervised transient detection using rectangular windows. Spike and seizure discrimination based on the spectral characteristics of epileptiform transients were much less sensitive to changes in amplitude, polarity and morphology over the initial weeks of epileptogenesis. Artifact rejection was poor with both techniques.
Unsupervised detection of ictal and interictal activity based on spectral characteristics is feasible during early epileptogenesis, when spike frequency is low and spike morphology varies widely over time. However, artifact rejection with this technique was not reliable, such that unsupervised utilization of this strategy of spike detection is only feasible with low-noise recordings.
[Supported by: NIH (NINDS)]