Analyse multi-échelles par ondelettes complexes : application aux signaux électrophysiologiques intracrâniens chez les patients épileptiques

Abstract

The main aim of this research is to design a computing environment software to develop analytical tools dedicated to the characterization of intracranial signals in epileptic patients. More especially, this study aims at discrimining interictal (between seizures) and pre-ictal (before seizures) periods. During the study of EEG signals, three features are noticed. The first is the presence of a power-law behavior of the power spectral density (PSD). This behavior of the PSD presented in the work of Yamaguchi (2003) could be explained by the presence of scale invariance in the signal. The second one is the increasing number of high frequency oscillations (HFO) when approaching the seizure. This result presented by Zijlmans et al. (2011) implies changes in the shape of the PSD when approaching the seizure. The third one is the use of intracranial EEG signals which allow to see high frequencies thanks to a good signal to noise ratio (SNR). These three characteristics make the study of scale invariance properties a potential candidate for the discrimination of pre-ictal and interictal periods. One contribution of this thesis was to develop an estimator of the scale invariance properties based on symmetrical complex Daubechies wavelets introduced by Lina et Mayrand (1993). These wavelets improve performance of the estimator and the temporal precision when calculating wavelet. Another significant contribution is the application of the estimator to intracranial EEG signals using a sliding window method based on hypothesis testing. The use of this methodology on epileptic patients intracranial EEG signals during the interictal and preictale period confirmed the presence of scale invariance properties. This application also showed significant variations in these properties at the transition from the préictale to interictal period. These results led to a collaboration with the Montreal Neurological Institute in the framework of epileptic seizure prediction. The results of this collaboration corroborate the diagnosis established by the neurologist and improve the performance of the prediction algorithm.

Date	19 Nov 2014
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Jean-Marc Lina (Supervisor)