Identification of Functional Clusters in the Striatum Using Infinite Relational Modeling

In this paper we investigate how the Infinite Relational Model can be used to infer functional groupings of the human striatum using resting state fMRI data from 30 healthy subjects. The Infinite Relational Model is a non-parametric Bayesian method for infering community structure in complex networks. We visualize the solution found by performing evidence accumulation clustering on the maximum a posterior solutions found in 100 runs of the sampling scheme. The striatal groupings found are symmetric between hemispheres indicating that the model is able to group voxels across hemispheres, which are involved in the same neural computations. The reproducibility of the groupings found are assessed by calculating mutual information between half splits of the subject sample for various hyperparameter values. Finally, the model’s ability to predict unobserved links is assessed by randomly treating links and non-links in the graphs as missing. We find that the model is performing well above chance for all subjects.