Détection automatique des fuseaux de sommeils en magnétoencéphalographie et correspondance avec l’électroencéphalographie

Abstract

Sleep spindles are flashes of transient oscillation. They occur in the brain at a frequency between 10 and 16 Hz, last between 0.5 and 3 seconds and are generated during stage 2 of slow wave sleep. The detection of the flares is done visually on the EEG signals by sleep experts. Automatic detectors have been proposed to assist the experts, but visual identification remains the reference for marking. However, recent research indicates that spindle like activity is detectable on MEG signals. However, there is no consensus among sleep experts on the visual marking of the MEG signal spindles. The goal of this paper is to implement an automatic sleep spindle detector using the EEG-MEG multimodality. Some information contained by the MEG signal is not visible on the EEG signal and vice versa. Thanks to the fusion of the modalities, it becomes possible to use intermodal matching criteria to categorize the sleep spindles. In this work we are interested in three categories: the concordant sleep zones on both modalities (MEEG) and the non-concordant sleep zones on the EEG (L-EEG) or on the MEG (L-MEG). The multimodal analysis was performed on 13 EEG channels and 269 MEG sensors. The density topography of the detected MEG spindles shows that the MEEG spindles are located in the frontal region, while the L-MEG spindles are diffused through the cortex. The spindles with a high amplitude peak are located in the temporal, occipital and parietal regions. The L-MEG spindles, despite the lack of temporal agreement with the EEG spindles, are linked in amplitude to the EEG signals with an activity that differs from the ambient signal. But this result is not reciprocal for L-EEG spindles which show, on average, a very low activity on MEG signals. Both modalities, EEG and MEG, have their own spindles. In conclusion, the use of a spindle detector that works with both EEG and MEG signals contributes to the production of new results specific to the use of multimodality.

Date	21 Dec 2020
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Jean-Marc Lina (Supervisor)