Fig 1. Outline of the processing pipeline.

Both healthy (n = 26) and DOC (n = 23) data sets were processed by estimating signal complexity (left/right panel), next cluster analysis was applied on a group level; healthy, UWS, MCS (upper panel). Finally single subject classifier is trained using healthy previously scored data and tested on DOC, where video recordings are used as a validation proxy (lower panel). Note that first cluster analysis is used as an exploratory step. Next, after confirming that sleep patterns can be identified based on the features (PE), we train a predictive model (classifier).