[Sleep EEG as a nonlinear dynamic process: a comparison of global correlation dimension of human EEG and measures of...

The goal of this work was to study (1) whether the estimation of correlation dimension (D2) using spatial embedding distinguishes between sleep stages and (2) whether information gained from the application of global D2 is redundant to measures of linear interdependence between channels. Twenty one-channel EEG segments of 12 healthy male subjects recorded during waking and sleep stages REM, I, II, and III-IV (according to the Rechtshaffen and Kales criteria) were analyzed with global (multichannel) D2, mean square correlation coefficients (MS) and proportion of variance accounted for by the first principal component (PC1). D2 was found to decrease progressively from stage I to stage III-IV with D2 values of waking and REM being close to those of stages I and II. MS and PC1 did not distinguish among sleep stages but yielded significant differences between waking and sleep. The results suggest that global D2 extracts information from human EEG. That sort of evidence cannot be obtained with measures of linear interdependence between channels.