Year of Graduation
Change-Point Detection in EEG Records
Segmentation of EEG records is an essential step of the human brain analysis based on its electrical activity. It can be used as a preprocessing technique for the further analysis or as an analyzing tool itself showing the number and lengths of the statistically stationary segments which can reveal the nature of processes in the brain. Change-point detection is NP-hard problem since we do not know beforehand the number of change points. Thus due to the huge size of the EEG sequences we use a heuristic method to detect the change points. Our method is based on the recursive Jensen-Shannon segmentation scheme proposed by Bernaola-Galvan et al. in 1996. Our method provides better change-points detection comparing with other techniques and performs faster either on generated time series or on real EEG records because of our speed improvements. In the work we also describe other ways to improve the quality of found change points.