Connectivity-based parcellation of putamen using resting state fMRI data

In this paper, we present a novel framework for parcellation of a brain region into functional sub-regions based on connectivity patterns between brain regions. The proposed method takes the prior neurological information into consideration and aims at finding spatially continuous and functionally consistent sub-regions in a given brain area. The proposed framework relies on 1) a sparse spatially regularized fused lasso regression model for encouraging spatially and functionally adjacent voxels to share similar regression coefficients despite of spatial noise; 2) an iterative voxels (groups) merging and adaptive parameter tuning process; and 3) a Graph-Cut optimization algorithm for assigning overlapped voxels into separate sub-regions. With spatial information incorporated, spatially continuous and functionally consistent sub-regions could be obtained and further used for subsequent brain connectivity analysis.