Fine granularity parcellation of gyrus via fiber shape and connectivity based features

In vivo parcellation of the cerebral cortex via non-invasive neuroimaging techniques has been in active research for over a decade. A variety of model-driven or data-driven computational approaches have been proposed to parcellate the cortex. A fundamental issue in these parcellation methodologies is the features or attributes used to define boundaries between cortical regions. This paper proposes a novel DTI-derived fiber shape and connectivity based feature for the parcellation of cortical gyrus into fine granularity segments. The gyrus parcellation is formulated as a surface vertex clustering problem, in which both of feature similarity and vertex adjacency are used to define the distance between vertices. The affinity propagation algorithm is employed for the vertices clustering. This methodology is developed and evaluated using the precentral gyrus (primary motor cortex) as a test bed, and motor task-based fMRI is used to validate the parcellation results. The experimental results show that the precentral gyrus can be consistently parcellated into 3 broad segments on both hemispheres across different subjects using the proposed method, which is reasonable according to neuroscience knowledge and motor task-based fMRI activations.