Group-wise change point detection in task FMRI data by Bayesian methods

Change point detection has been one of the major objectives in task-based fMRI data analysis. In this paper, we focus on the application the Bayesian magnitude change point model (BMCPM) described in our other paper submitted to NIPS 2013. Firstly, we introduce the basics of BMCPM. The employed BMCPM statistically infers block boundaries separating the time series into quasi-stable temporal blocks that exhibit substantial differences of brain states from each other via a Bayesian framework equipped with a Markov chain Monte Carlo (MCMC) scheme. It applies Metropolis-Hastings algorithm to sample from the posterior distribution of different temporal segmentations (magnitude change points) of a given dataset and to estimate a series of posterior probabilities of time points being magnitude change points. After extensive simulations have been carried out to validate the proposed BMCPM and obtain good results, we focus on the application and analysis of BMCPM on real fMRI data. Then, these detected group-wise consistent magnitude change points are clustered into various patterns of temporal and spatial activities. The methods have been applied on a task-based fMRI dataset and revealed complex and meaningful brain activation patterns.