Multi-channel analysis of the EEG signals and statistic particularities for epileptic seizure forecast

In France, 5 to 8 people in 1000 suffer from epilepsy. An epileptic seizure is sudden, impressive, and is often followed by a loss of consciousness by the patient. The clinical studies have demonstrated that neuronal activity is responsible for these seizures. The electroencephalograms recorded by the doctors allow the visualization of the very beginning of the crises. The aim of our previous studies was to determine characteristics of the EEG signals that will allow us to forecast the seizures (Peyrodie et al., 2001). We reached the conclusion that using that using principal components analysis, we were able to find out some grapho-elements in the 2 first principal components that where leading us to highlight a state were patients are likely to make a seizure. A new challenge consists in trying to provide an interpretation of these grapho-elements. The first part is concerned in the explanation of the principal components analysis and the results it gave us. The second part is concerned with a statistical study of the principal components using a log-likelihood method. The third part is concerned with the use of independent component analysis to filter the EEG signals.