Seizure prediction using a hippocampal circuitry model developed based on a tripartite synapse structure

This study presents a new approach to develop a model based seizure prediction system. In this system, a neural mass model is used to simulate the macro-scale dynamics of a cortical micro-zone and the local field potentials (LFP) generated by its activity. The LFP is modeling with an acceptable approximation of the intracranial EEG (iEEG) data. The model includes one population of pyramidal cells, two populations of inhibitory interneurons and a functional block describing basic modulatory and regulatory functions of astrocyte. All of these populations are described through nonlinear state equations. Seventeen parameters of the model are estimated using a parameter estimation method that utilizes optimization algorithms to minimize statistical distances between real recorded iEEG and output of the model signals. Then, by investigating the temporal changes in the estimated parameters within pre-ictal period in comparison with the reference values estimated from inter-ictal period, the upcoming seizures can be predicted. iEEG of three dogs with temporal lobe epilepsy (TLE) is used in this study. Results demonstrate that the system can predict the seizures at least 20 minutes before the occurrence of seizure. The presented findings show that the proposed model-based algorithm might be able to predict effectively epileptic seizures in patients suffering from TLE.