Data-determined window size and space-oriented segmentation of spontaneous EEG map series.

For the segmentation of series of momentary potential distribution maps into epochs of quasi-stable landscape (brain electric microstates), the maps are reduced to extracted landscape descriptors. Changes of the descriptors over time are recognized as segment terminators. The selection of the descriptors' tolerated variance (the window size) determines the result. We present a window-determining function which allows a data-driven determination of the optimal window size, based on equal weight given to the recognition of similarity and dissimilarity between maps. Segmentations based on two map descriptors (locations of extreme potentials and centroids) were used on 211 two-second map epochs from 8 normal subjects for validation of the window-determining function and to establish normative data. Using the data-determined window sizes for segmentation, the mean duration of the obtained microstates across subjects did not differ between descriptors (144 and 143 msec, respectively). Random permutation of the maps in time produced significantly shorter segments, ensuring that the segmentation disclosed real properties of the original data and not artifacts of the procedure.