Critères spatial et spectral pour la détection des fuseaux du sommeil en EEG

Abstract

Sleep spindles are short bursts of relatively high frequency oscillations occurring mainly during stage 2 sleep as observed in EEG. This sporadic activity is thought to have a role in sleep integrity protection, memory processes and plasticity. Many automatic detectors have been proposed to assist or replace the expert in the task of identifying sleep spindles. The persistent problem is that these algorithms usually detect too many events and that the compromise between sensitivity (Se) and specificity (Sp) is difficult to achieve. In this work, we propose a semi-automatic and supervised detector which adds a specificity phase, using spatial and frequency criteria, to a sensitivity phase based on proven criteria in the literature. In the sensitivity phase, selected candidate events (10Hz-16Hz band) are those whose amplitude and spectral ratio characteristics reject a null hypothesis (p <0.1), which is that the considered event is not a spindle. This null hypothesis is constructed from events occurring during REM stages identified by an expert. In the specificity phase, a hierarchical clustering of the candidates is done on the frequency and spatial position (anterior-posterior)characteristics. The selected class is the one grouping the majority of a set of spindles marked by an expert. In the first phase, we obtain Se = 93.2% and Sp = 89.0%. In the second phase, we obtain Se = 85.4% and Sp = 95.5%. Results suggest that spatio-frequency criteria are characteristic to spindles and can help improve automatic detection methods.

Date	15 May 2013
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Jean-Marc Lina (Supervisor) & Julie Carrier (Co-supervisor)