Online Bayesian change point detection algorithms for segmentation of epileptic activity

Epilepsy is a dynamic disease in which the brain transitions between different states. In this paper, we focus on the problem of identifying the time points, referred to as change points, where the transitions between these different states happen. A Bayesian change point detection algorithm that does not require the knowledge of the total number of states or the parameters of the probability distribution modeling the activity of epileptic brain in each of these states is developed in this paper. This algorithm works in online mode making it amenable for real-time monitoring. To reduce the quadratic complexity of this algorithm, an approximate algorithm with linear complexity in the number of data points is also developed. Finally, we use these algorithms on ECoG recordings of an epileptic patient to locate the change points and determine segments corresponding to different brain states.