Adaptive segmentation of spontaneous EEG map series into spatially defined microstates

Abstract

Space-oriented segmentation can decompose multi-channel EEG map series into time segments characterized by quasi-stationary field map configurations. This assesses the dynamics of the underlying processes as activities of different neural generator ensembles. Our method of space-oriented segmentation describes the scalp field at times of maximal field strength (Global Field Power) by the locations of the centroids of positive and negative map areas. A quantitative measure of the simultaneous distance of the centroid locations evaluates the similarity between consecutive maps. A segment is defined as a sequence of maps that do not differ from each other by more than a present value. Finally, the average centroid locations for each segment are entered into an agglomerative clustering procedure to obtain a set of distinct classes of field configurations. Four records of 16 s of 42-channel resting EEG (band-pass filtered 2-16 Hz) from six subjects were analyzed. Average segment duration was 157.9 ms. Most segments belonged to a small number of classes (from 2 to 6, mean 3.7 classes for 90% of analysis time). The most frequent class showed an anterior-posterior field orientation and covered from 45 to 74% (mean 55% across subjects) of total time, with an average duration of 265 ms. The procedure was also tested using temporally and spatially unstructured data (white noise and randomly shuffled EEG) to ascertain that the methods reflect the spatio-temporal structure of the EEG processes.