Normal Modes of the Structural Connectome

Division of the human cortex into distinct regions is of high importance to neuroscientific inquiry. Fully-automated, multi-modal schemes of achieving such parcellation on an individual subject basis are particularly advantageous, however difficulties in inter-modal and inter-subject registration of brain images, as well as obstacles in preserving group-level correspondence of individual parcellation maps have slowed progress in this area. In parallel, there exists a relative dearth of data-driven parcellation schemes that incorporate high resolution structural connectivity metrics; the majority of widely-accepted parcellation maps in the literature have primarily used functional connectivity. Here, a fully data-driven, automated routine based on structural geometry and connectivity which achieves subject-specific cortical parcellation maps while maintaining group-level correspondence of maps is presented and optimized. Using high resolution white matter surface meshes and advanced fiber tracking techniques, a novel vertex-wise structural connectivity graph is constructed for each of 10 unrelated subjects, and the first k eigenvectors of the Laplacian Matrix of the graph's adjacency matrix are calculated. These eigenvectors represent the steady-state modes of vibration of the manifold described by this graph, and thus provide subject-specific maps of modes of connectivity in white matter. In order to obtain parcellations at varying levels of coarseness of the cortex from these eigenvectors, hierarchical agglomerative clustering is then performed on the surface mesh, where each vertex's feature vector is its profile in spectral space. Further, a multi-layer graph of all subjects is constructed, and individual parcellations with group level correspondence are obtained by agglomerative clustering of the eigenvectors of the laplacian matrix of the multi-layer graph.Bachelor of Scienc