Figure 2
From: On wakefulness fluctuations as a source of BOLD functional connectivity dynamics
Clustering into four dynamic connectivity states tracks wakefulness and NREM sleep stages. (A) We matched occurrence probabilities over time of the identified (W, N1-N3; based on EEG; red lines) sleep labels with the assigned cluster IDs (CW, CN1, CN2, CN3; based on k-means clustering; black lines), the legends show the linear correlation coefficient r in each case. This allowed a bidirectionally unequivocal mapping of each cluster to one sleep stage. (B) Overlay of the average EEG spectra (PSD: power spectral density) obtained for the time windows identified with each sleep stage (black line) and the matched dynamic connectivity state (red line), respectively. (C) Comparison of the average fMRI correlation matrix obtained for the volumes identified with each sleep stage (left column) and with the matched clustered dynamic connectivity states (right column). (D) We confirmed the high visual similarity between both columns in panel C by computing the correlation between all pairs of correlation matrices, observing the highest values along the diagonal (confirming the best match). The explicit values of the correlation coefficient are shown along the diagonal. (E) Accuracy (defined as percentage of correctly labeled volumes taking EEG-based sleep scoring as the reference) as a function of the window length.
