On Clustering fMRI Using Potts and Mixture Regression Models

In this paper, we propose a model based clustering method for functional magnetic resonance imaging (fMRI) data to detect the functional connectivity network. The Potts model, which represents spatial interactions of neighboring voxels, is introduced to integrate the temporal mixture regression modeling into one single unified model. The estimation of the parameters is achieved through a restoration maximization (RM) algorithm for computation efficiency and accuracy. Additional features of our method include: the optimal number of clusters can be automatically determined; global trends and informative paradigms of the data are extracted by a dimension reduction algorithm based on principal component analysis (PCA) and a statistical significance test. Experimental results demonstrate that our approach can lead to robust and sensitive detection of functional networks.