Functional parcellation of the hippocampus by clustering resting state fMRI signals

In this study, we propose a semi-supervised clustering method for parcellating the hippocampus into functionally homogeneous subregions based on resting state fMRI data. Particularly, the semi-supervised clustering is implemented as a graph partition problem by modeling each voxel as one node of the graph and connecting each pair of voxels with an edge weighted by a similarity measure between their functional signals. A geometric parcellation result of the hippocampus is adopted as prior information and a spatial consistent constraint is adopted as a regularization term to achieve spatially contiguous clustering. The graph partition problem is solved using an efficient algorithm similar to the well-known weighted kernel k-means algorithm. Our method has been validated based on resting state fMRI data of 28 subjects for the hippocampus parcellation with three subregions. The experiment results have demonstrated that the proposed method could parcellate the hippocampus into its head, body and tail parts. The distinctive functional and structural connectivity patterns of these subregions, derived from resting state fMRI and dMRI data respectively, have further demonstrated the validity of the parcellation results.