Fig. 1
From: Discovery of key whole-brain transitions and dynamics during human wakefulness and non-REM sleep
Dynamic whole-brain networks from fMRI sleep recordings using a Hidden Markov Model. a ROI timecourses were extracted by averaging BOLD signals across voxels within each of the 90 cortical and subcortical AAL areas for each participant. Each ROI timecourse was demeaned and normalised by its standard deviation. b The data were concatenated across participants, and the dimensionality was reduced using PCA (principal component analysis), such that ~90% of the variance of the ROI timecourses was retained. The HMM was run on the PCA timecourses, resulting in K number of states with associated timecourses, each describing the points in time each state is active and inactive. c Each HMM state was characterised by a multivariate Gaussian distribution comprising a covariance matrix, ΣK, and a mean distribution, μK. The state-specific mean distributions and covariance matrices were back-projected to the MNI space of the AAL by using the mixing matrix, MT from the PCA decomposition, yielding a mean activation map and an FC matrix for each HMM state
