Approches non-supervisées et non-linéaires pour l’analyse de signaux de pléthysmographie

Plethysmography groups together a set of non-invasive methods for measuring respiratory volumes. The current techniques of analysis of the obtained signals focus on the extraction of global characteristics which bring little information on the respiratory behavior and its dynamics. To overcome this problem, we propose a new unsupervised method to automatically segment and extract typical respiratory cycles using a variant of Dynamic Time Warping (DTW) and clustering algorithms. These typical sequences are then used to build a symbolic representation in the form of a colored timeline, which allows a visual analysis of the whole respiratory dynamics.