An automated method for micro-state segmentation of evoked potentials

We present a method for segmenting evoked potentials into functional micro-states. The method is based on measuring the similarity between all the topographic maps in the evoked potential and grouping them into functional micro-states based on minimizing an error function. The similarity is measured as the normalized cross-correlation coefficient. The method was validated on simulated data and tested on its ability to segment a visual evoked potential. On simulated data the method missed from 1% to 8.5% of the micro-state boundaries for evoked potentials with a signal-to-noise ratio of 20-1 dB, respectively. The proposed segmentation method was compared with segmentation based on K-mean clustering. It was found that the proposed method was better at detecting the correct number of micro-states and was computationally more efficient. The automatic segmentation of the visual evoked potential was compared to the manual segmentation performed by eleven EEG specialists. No significant difference in the deviation of micro-state boundaries was observed between two random EEG specialists and between a random EEG specialist and the automatic method. Thus it was found that the method could reliably segment evoked potentials into their functional micro-states.