Detailed connectomic cluster resource for white matter mapping from ultra-high-field diffusion MRI

Large-scale brain mapping initiatives have underscored the necessity for white matter atlases that extend beyond the currently identified pathways, particularly in underexplored regions such as the superficial and cerebellar white matter. To address this gap, we develop a data-driven fiber-cluster atlas using ultra-high-field 7T structural and diffusion MRI data from 171 participants in the Human Connectome Project (HCP). Following preprocessing, we construct the whole-brain tractogram, comprising probabilistic and deterministic tractography from multi-tissue fiber orientation dispersion functions to mitigate streamline-tracking bias. Data from multiple algorithms are registered to the MNI space and subsequently aggregated. We cluster streamlines connecting seven cortical networks and nine subcortical regions using cosine k-means clustering along with two-level consensus filtering. The resulting atlas comprises 33,256 clusters for a seven-network scheme and 65,184 clusters for a seventeen-network scheme, encompassing both deep and superficial white matter. Across participants, the overlap between individual and population clusters exceeds 97%, and the median Davies-Bouldin scores are below 0.35, indicating high reproducibility and anatomical compactness. Importantly, classical tracts such as the arcuate fasciculus and corticospinal tract are subdivided into anatomically coherent subclusters, and numerous previously uncharacterized U-fibers are also identified. This open-access 7T resource aims to facilitate research on structure-function relationships, algorithm benchmarking, and precision connectomics.