False discovery rate controller for functional brain parcellation

Adrian Wong, Martin J. McKeown, Mehdi Moradi, Z. Jane Wang

Abstract: Parcellation of brain imaging data is desired for proper neurological interpretation in resting-state functional-magnetic resonance imaging (rs-fMRI) data. Some methods require specifying a number of parcels and using model selection to determine the number of parcels on a rs-fMRI dataset. However, this generalization does not fit with all subjects in a given dataset. A method has been proposed using parametric formulas for the distribution of modularity in random networks to determine the statistical significance between parcels. In this paper, we propose an agglomerative clustering algorithm using parametric formulas for the distribution of modularity in random networks, coupled with a false discovery rate (FDR) controller to parcellate rs-fMRI data. The proposed method combines FDR to reduce the number of false positives and incorporates spatial information to ensure the regions are spatially contiguous. Simulations demonstrate that the proposed FDR controlled method yields more accurate results when compared with existing methods. We also applied the proposed method to real a rs-fMRI dataset and found that it obtained higher reproducibility compared to the Ward hierarchical clustering method.