SEGMENTATION OF BRAIN ELECTRICAL-ACTIVITY INTO MICROSTATES - MODEL ESTIMATION AND VALIDATION

A brain microstate is defined as a functional/physiological state of t he brain during which specific neural computations are performed. It i s characterized uniquely by a fixed spatial distribution of active neu ronal generators with time varying intensity, Brain electrical activit y is modeled as being composed of a time sequence of nonoverlapping mi crostates with variable duration, A precise mathematical formulation o f the model for evoked potential recordings is presented, where the mi crostates are represented as normalized vectors constituted by scalp e lectric potentials due to the underlying generators, An algorithm is d eveloped for estimating the microstates, based on a modified version o f the classical k-means clustering method, in which cluster orientatio ns are estimated, Consequently, each instantaneous multichannel evoked potential measurement is classified as belonging to some microstate, thus producing a natural segmentation of brain activity, Use is made o f statistical image segmentation techniques for obtaining smooth conti nuous segments, Time varying intensities are estimated by projecting t he measurements onto their corresponding microstates, A goodness of fi t statistic for the model is presented, Finally, a method is introduce d for estimating the number of microstates, based on nonparametric dat a-driven statistical resampling techniques.