Abstracts

Rationale: Seizure detection and prediction are critical for the implementation of real time seizure intervention therapies. These algorithms often make use of the analysis of a single channel of electrophysiological data to explore various aspects of the dynamics of the brain. We hypothesize that effective connectivity algorithms, such as Granger causality that delineates the strength and direction of temporal and spatial patterns of interdependency, can be used to detect abnormal patterns of activity between hippocampal subfields that would be predictive of seizure. Methods: Fifty-day-old Sprague-Dawley rats, weighing between 200 - 250 g were chronically implanted bilaterally in the hippocampus CA1 and dentate gyrus subfields with dual 16-channel 50-micron microelectrode arrays and a 330-micron bipolar twist electrode in the ventral hippocampus. Two to six weeks after electrical self-sustaining status epilepticus using the bipolar twist electrodes these animals experience intermittent spontaneous Racine grade 5 seizures. In previous analysis of these seizures using Granger causality (a multichannel signal processing tool that delineates the strength and direction of dependencies between brain areas) we described the dynamic patterns of interdependency between bilateral hippocampal subfields during seizure. During the initial analysis, it was unclear when a change from normal interdependency patterns occur prior seizure onset. Using these same methods, we explore the period immediately before the seizure onset. The seizure onset was delineated by previously developed seizure detection methods that were verified by an expert epileptologist. Results: A general progressive strengthening of directional interdependencies was shown from electrodes placed in the left CA1 to electrodes placed in the left dentate gyrus prior to seizure onset beginning up to 3 minutes prior to the behavioral onset of seizure in 15 Racine grade seizures from 3 animals. Generally, this sequence begins after an interictal population spike and gradually the interdependencies between these regions nearly double. While these initial findings were encouraging, results were found to vary somewhat between animals. Specifically, in one animal this pattern was observed before every Racine grade 5 seizure, where seizures from two other animals demonstrate this pattern but less reliably. Conclusions: Gradual increases in interdependencies from the CA1 to the dentate gyrus were found to precede seizure onset in the animal model of temporal lobe epilepsy. Variability between animals may have been due to subtle differences in electrode placement and potentially multiple seizure foci. However, we suggest that these preliminary results indicate the possible efficacy of combination of strategically positioning electrodes and effective connectivity measures for early seizure detection. Future work will focus on optimal positioning of electrodes in areas including the entorhinal cortex to capture this sequence for early seizure detection.