Detection of change points in phase data: a Bayesian analysis of habituation processes

Given a time series of data points, as obtained in biosignal monitoring, the change point problem poses the question of identifying times of sudden variations in the parameters of the underlying data distribution. We propose a method for extracting a discrete set of change points from directional data. Our method is based on a combination of the Bayesian change point model (CPM) and the Viterbi algorithm. We apply our method to the instantaneous phase information of single-trial auditory event-related potentials (ERPs) in a long term habituation paradigm. We have seen in previous studies that the phase information enters a phase-locked mode with respect to the repetition of a stimulus in the state of focused attention. With adaptation to an insignificant stimulus, attention tends to trail away (long-term habituation), characterized by changes in the phase signature, becoming more diffuse across trials. We demonstrate that the proposed method is suitable for detecting the effects of long-term habituation on phase information in our experimental setting.