An atlas of white matter anatomy, its variability, and reproducibility based on Constrained Spherical Deconvolution of diffusion MRI

Abstract Virtual dissection of white matter (WM) using diffusion MRI tractography is confounded by its poor reproducibility. Despite the increased adoption of advanced reconstruction models, early region-of-interest driven protocols based on diffusion tensor imaging (DTI) remain the dominant reference for virtual dissection protocols. Here we bridge this gap by providing a comprehensive description of typical WM anatomy reconstructed using a reproducible automated subject-specific parcellation-based approach based on probabilistic constrained-spherical deconvolution (CSD) tractography. We complement this with a WM template in MNI space comprising 68 bundles, including all associated anatomical tract selection labels and associated automated workflows. Additionally, we demonstrate bundle inter- and intra-subject variability using 40 (20 test-retest) datasets from the human connectome project (HCP) and 5 sessions with varying b-values and number of b-shells from the single-subject Multiple Acquisitions for Standardization of Structural Imaging Validation and Evaluation (MASSIVE) dataset. The most reliably reconstructed bundles were the whole pyramidal tracts, primary corticospinal tracts, whole superior longitudinal fasciculi, frontal, parietal and occipital segments of the corpus callosum and middle cerebellar peduncles. More variability was found in less dense bundles, e.g., the first segment of the superior longitudinal fasciculus, fornix, dentato-rubro-thalamic tract (DRTT), and premotor pyramidal tract. Using the DRTT as an example, we show that this variability can be reduced by using a higher number of seeding attempts. Overall inter-session similarity was high for HCP test-retest data (median weighted-dice = 0.963, stdev = 0.201 and IQR = 0.099). Compared to the HCP-template bundles there was a high level of agreement for the HCP test-retest data (median weighted-dice = 0.747, stdev = 0.220 and IQR = 0.277) and for the MASSIVE data (median weighted-dice = 0.767, stdev = 0.255 and IQR = 0.338). In summary, this WM atlas provides an overview of the capabilities and limitations of automated subject-specific probabilistic CSD tractography for mapping white matter fasciculi in healthy adults. It will be most useful in applications requiring a highly reproducible parcellation-based dissection protocol, as well as being an educational resource for applied neuroimaging and clinical professionals. Graphical abstract (Top) shows the FWT pipeline for both CSTs, AF, and motor CC bundles. (Left to right) show the required input structural parcellation maps and a priori atlases for FWT and the resulting virtual dissection include/exclude VOIs. FWT provides two approaches to virtual dissection: (1) is a bundle-specific approach where streamlines are only seeded for the bundle of interest, (2) is a whole brain tractography followed by streamlines segmentation, (top right) shows output tractograms. (Middle) Group-averaged T1 and fODF images are generated from the HCP test-retest data, and FWT is applied to generate the HCP-atlas using the bundle-specific approach (1*). FWT’s whole brain tracking and segmentation approach (2*) was applied to the HCP and MASSIVE dataset (right and left) and conducted model-based, and pair-wise similarity analyses and generated voxel-wise cumulative maps per bundle. FWT= Fun With Tracts, FS= FreeSurfer, MSBP= MultiScaleBrainParcellator, PD25= NIST Parkinson’s histological, JHU= John’s Hopkins university, Juelich= Juelich university histological atlas, AC/PC= anterior commissure/posterior commissure) UKBB= UK Biobank, SUIT (spatially unbiased cerebellar atlas template), dMRI= diffusion magnetic resonance imaging, CSD= constrained spherical deconvolution, fODF= fiber orientation distribution function, CST= corticospinal tract, AF= arcuate fasciculus, CC= corpus callosum, HCP= human connectome project, MASSIVE= Multiple acquisitions for standardization of structural imaging validation and evaluation.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.