One of the main goals of recent large collaborative studies such as the Human Connectome Project (HCP) [1] has been to develop a reliable map of brain connectivity. To most accurately map brain networks, there has been an increasing demand for improvements in the achievable angular and spatial resolution of in-vivo diffusion Magnetic Resonance Images (dMRI). While high angular resolution dMRI can greatly help resolve crossing fibers, it is often not effective in delineating short association fibers of very high curvature (i.e. high turning angle) within the voxel in images with low spatial resolution. Recent studies [2] have shown that it is possible to achieve sub-millimeter isotropic spatial resolutions in-vivo, thereby recovering and resolving short association fibers (i.e. u-fibers connecting adjacent gyri). As these fibers play a critical role in constructing the human connectome, this report aims to assess the connectivity gains at high spatial resolution. Initial evidences are presented that demonstrate both the recovery of lost connections (edges), as well as improved accuracy of mutual connectivities (i.e. thickness of edges) between nodes (ROIs), when the spatial resolution increases. Effort is also made to potentially identify an optimal spatial resolution for an accurate construction of a human brain connectome.To assess the impact of spatial resolution over a wide range, eight data sets with 0.9, 1.0, 1.1, 1.2, 1.4, 1.6, 1.8 and 2.0 mm isotropic voxels were generated by resampling an in-vivo dMRI data set acquired with 15 diffusion directions and true 0.85 mm isotropic voxels. After resampling, fiber tractography was carried out for each resolution at every turning angle threshold between 10° and 82° using the FACT tracking algorithm [3]. The optimal tracking angles for each resolution were identified as the angle that produced the maximum number of short association fibers (3-30 mm) [4], illustrated in Figure 1. Using the FreeSurfer gray matter ROI segmentation, a connectome matrix was calculated for each resolution based on the fibers tracked with the optimal turning angle at that resolution.

To quantify improvements in the connectivities at each node in the connectome gained by increasing spatial resolutions, the total number of nodes connected with each node (the degree) was measured for all resolutions. As presented in Figure 2, every node in the segmentation exhibits a variation in degree across resolutions, with the greatest degree at each node observed by the high-resolution data and the lowest degree observed in the data with 2.0 mm voxels. Although some nodes, such as the pre and post-central gyri, only display a degree variation of 4 nodes between the high and low resolution data sets, the thickness of the fiber bundles that make up the edge between these gyri varies greatly across the resolutions. Figure 3 illustrates that voxels on the order of 0.85-1.0 mm can accurately represent the short association u-fibers that form the connection between the sensory and motor cortices, while 2.0 mm voxels are unable to resolve these fibers, reducing the connection to only a handful of fibers.

While the degree presents a measure of connectivity improvements on a node-by-node basis, a quantification of improvements to the connectome as a whole is achieved by calculating the correlation coefficient between the connectome of each resampled resolution and that of the original 0.85 mm data set. As shown in Figure 4, the strong correlation observed for spatial resolutions from 0.9 to 1.2 mm suggests that the overall connectome structure of the 0.85 mm data set is well preserved. However, for images with voxels larger than 1.2 mm, there is a steady decrease in correlation that indicates a significant reduction (i.e. loss of connectivity) in the connectome map. This is caused when short association fibers with high curvature are inadequately resolved at low spatial resolutions, resulting in either a reduction of connectivities within existing edges, as previously illustrated in Figure 3, or a loss of edges altogether.

This study experimentally and systematically demonstrates that it is possible to construct a consistent and accurate human brain connectome at very high spatial resolution, after recovering lost edges and refining thicknesses of existing edges and the radii of nodes. Moreover, it is shown that a minimal spatial resolution of 1.2 mm is needed before significant connectivity information is lost. Such a loss is primarily due to the missing cortical u-fibers, as well as the inability to resolve crossing fibers, at low spatial resolution. The results presented in this study will help our continued effort to improve the accuracy of the human brain connectome.