Research On Algorithms Of Brain Connectivity In Epilepsy Recorded In EEG Signals

In China, epilepsy has become the second most common disease in Neurology after headache. It is a neurological disorder characterized by repetitive seizures. In 30% of cases, seizures remain drug-resistant. For these patients with drug-resistant epilepsy, the treatment is used to remove the epileptogenic zone (EZ) by surgical procedures. The key of the operation is to locate the EZ position accurately. By studying the algorithms of brain effective connectivity, we can analyze the recorded EEG signal, which provides effective help for the accurate positioning of the preoperative assessment of the EZ. Therefore, this thesis focuses on the study of the model and the algorithm of the brain effective connectivity.Wiener-Granger Causality Index (WGCI) is a typical data-driven method of effective connectivity. It is a method in the context of linear autoregressive (AR) models of stochastic processes, which can detect the causal influence between the multiple time series. The partial directed coherence (PDC) and nonlinear partial directed coherence (NPDC) methods are extended from WGCI, and these two methods can detect the linear and nonlinear causality between signals in frequency domain. These methods are dependent on the linear or nonlinear autoregressive model. Therefore, the parameter estimation of autoregressive model is a very important part in our work. Optimal parameter search (OPS) and forward regression orthogonal least squares (FROLS) are the methods of parameter estimation for linear and nonlinear autoregressive models respectively. But there are two shortcomings in the OPS method:1.OPS is not good in noise conditions.2. The process of OPS requires a threshold to be used as a condition for the termination of the algorithm, however, there is no theoretical method to select the threshold value. Therefore, two improved methods based on generalized Akaike information criterion (gAIC) and OPS are proposed in this thesis: 1.Seting up a moving window on the signal to reduce the impact of noise.2. All candidate terms are divided into high part and low part according to the weight value, and the high part is selected as the final candidate part, so as to solve the problem that the threshold is difficult to be selectedIn the experimental part of this thesis, we first apply the OPS method to the linear autoregressive model and the WGCI method is improved, then the FROLS method is applied to the nonlinear autoregressive model and the physiological-based model. We also apply the PDC and NPDC to detect the linear or nonlinear effects between the signals.