Parcellation of functional sub-regions from fMRI: A graph clustering based approach

Abstract We propose a framework for parcellating a single brain region-of-interest (ROI) into spatially-contiguous functional sub-regions (subROIs) – each consisting of one or more voxels – based on fMRI connectivity patterns between subROIs and other brain ROIs. First, a functional connectivity network between the voxels in the primary ROI is generated by taking into account the connectivity pattern within the primary ROI and all other ROIs, with a spatial constraint to ensure the spatial continuity of the final subROIs. A community detection algorithm is then applied to the associated adjacency matrix of the connectivity network to parcellate it into functional subROIs. As an illustrative example, the framework was applied to resting state fMRI data from nine healthy subjects to parcellate the putaminal region into two functional subROIs. Training on odd and even time points resulted in more than 98% concurrence of voxels assigned to the same cluster. The relative fraction of voxels assigned to each subROIs was also robust across subjects. As a general tool, the proposed framework has the potential to be integrated into studies investigating subROI alterations in neurological disorders.