Clustering Probabilistic Sleep Microstate Curves: a Functional Data Analysis Approach

We introduced and validated an EEG data-based model of the sleep process with an arbitrary number of different sleep states and a high time resolution allowing modelling of sleep microstructure. The proposed probabilistic sleep model describes sleep via posterior probabilities of a finite number of microstates. Using the model, we extracted objective sleep parameters describing quantitative and qualitative characteristics of the probabilistic sleep microstate curves and proved their usefulness when assessing selected aspects of sleep quality. In the current work we are investigating functional data clustering methods applied to sleep microstate posterior curves. The hierarchical structure of the data given by the repeated visits of subjects in the sleep lab motivates our focus on recently proposed multilevel functional clustering analysis approaches. We are applying the multilevel functional principal component analysis to the sleep posterior curves. Preliminary results show promising potential of the approach to separate age-related sleep profiles and extracting subjects' specific night deviations from the mean sleep profiles.