Research On Effective Connectivity In Epilepsy Recorded In Ieeg Signals Using Dynamic Causal Modeling

In recent years, due to the high incidence of epilepsy, its treatment has become a hot research topic. About 30% patients are drug-resistant and this pathology affects all their life. Drug-resistant epilepsies are often partial with an epileptogenic zone (EZ) located in a particular brain region. In this case, in order to remove the EZ to avoid seizures, a surgical treatment should be considered.This study deals with effective connectivity analysis among distant neural ensembles recorded with intracerebral near field electrodes during seizures (fast onset activity, FOA) in the brains of epileptic patients. Our goal is to analyze the ability of Dynamic Causal Modeling of steady state response (DCM-SSR) approach to detect causal links when the underlying model is a well-known neural population model (Physiology-based Model) dedicated to the simulation of epileptic activities in hippocampus in human brian. The output of the Physiology-based Model can be interpreted as intracranial electroencephalographic (iEEG) signals similar to those recorded with depth electrodes in the brain. Our objective is to determine the causal relationship between iEEG signals. The analysis of such signals remains a difficult task aiming at determining the relationship between the different zones prior to surgery, this problem is usually called effecitive connectivity which is defined as causal (directed) influences between neural populations.DCM-SSR includes two parts:(1) Generative model:we first introduce a physiology-based model to generative model in DCM, from the state-space description of the system obtained by coupling a pair of such neural population models, a linearization around the equilibrium state leads to a transition matrix and a parametrized description of the power spectral densities matrix for the corresponding pair of output iEEG signals in the two-populations model. (2) Bayesian statistic inference:given the observed spectral, the interesting global model parameters’ priors are set, the posterior of parameters are estimated and the object function (Free Energy) is calculated in a Bayesian framework from the observed spectral by the Variationnal Bayes Expectation Maximization (VBEM) algorithm, then Log Bayes Factors are employed to discriminate among the possible effective connectivity hypotheses. Simulation results show that DCM-SSR can identify and distinguish the independence, unidirectional or bidirectional interactions between two epileptic populations in hippocampus both in Simulated Power Density (theoridical senario) and in Sample Power Density (real senario).In the last part of this thesis, our current work is summarized and some perspectives are given.