Dataset: A multi-scale probabilistic atlas of the human connectome

This repository complements the paper submitted to Scientific Data named A multi-scale probabilistic atlas of the human connectome.

Introduction

The assessment of the networks underlying brain processes is key to understand brain-related disorders. However, groundbreaking connectomics research is highly demanding in terms of equipment and expertise. The aim of this work is to create a multiscale probabilistic atlas of the human white matter (WM) to carry out network analyses in the context of clinical research, particularly when diffusion data is not directly available.

Methods

Sixty six subjects from the Human Connectome Project (HCP) database (29 males, age: 22-36 years old) were used to build the WM probabilistic atlas. MRI acquisition protocols are described in (Van Essen et al, 2012). Besides T1-, T2- and diffusion-weighted (DW) images, the HCP database provides the FreeSurfer outputs (Glasser et al, 2016) namely cortical surfaces (pial and white), subcortical segmentation and a cortical surface parcellation containing 34 structures for each hemisphere (Desikan et al, 2006).

For each subject, the DWIs were employed to segment each thalamus in seven nuclei (Battistella et al, 2016) and to estimate the WM streamlines distribution. The constrained spherical deconvolution (Tournier et al, 2007) algorithm was used to compute the intravoxel fiber distribution functions for the anatomically-constrained particle-filter tractography approach (Descoteaux et al, 2009) to compute the WM streamlines.

Subcortical, thalamic and multiscale cortical (Cammoun et al, 2012) parcellations were gathered to obtain four individual gray matter (GM) parcellations. Finally, for each scale, individual fiber bundles, were created by selecting the streamlines connecting each pair of GM regions (Figure 1a).

Atlas construction

The T1 and T2 images were non-linearly warped to their corresponding MNI templates (Evans et al, 2012, mni_icbm152_tal_nlin_asym_09c version) using ANTs (Avants et al, 2010). The resulting spatial transformations were applied to warp the individual fiber bundles to stereotactic space and the normalized tract density images (TDIs) were created. In these images, each voxel contains the number of streamlines passing through it. Finally, the spatial probability map for each bundle was obtained by binarizing and averaging the bundle TDIs across the subjects (Figure. 1b).

Different views of the developed probabilistic multi-scale connectome atlas are shown in Figure 2.