A collaborative resource platform for non-human primate neuroimaging

Abstract

Neuroimaging non-human primates (NHPs) is a growing, yet highly specialized field of neuroscience. Resources that were primarily developed for human neuroimaging often need to be significantly adapted for use with NHPs or other animals, which has led to an abundance of custom, in-house solutions. In recent years, the global NHP neuroimaging community has made significant efforts to transform the field towards more open and collaborative practices. Here we present the PRIMatE Resource Exchange (PRIME-RE), a new collaborative online platform for NHP neuroimaging. PRIME-RE is a dynamic community-driven hub for the exchange of practical knowledge, specialized analytical tools, and open data repositories, specifically related to NHP neuroimaging. PRIME-RE caters to both researchers and developers who are either new to the field, looking to stay abreast of the latest developments, or seeking to collaboratively advance the field .

Keywords: Diffusion; Functional; Open science; Pipeline; Resource sharing; Structural; Toolbox.

Fig. 1.. Organization of PRIME-RE and common NHP-neuroimaging analysis workflows.

A) The main components of the PRIME-RE website (https://prime-re.github.io) are the Tutorials and Resources sections. The Tutorials section has a community-driven Wiki explaining the basics of NHP neuroimaging as well as links to external tutorials. The Resources section provides descriptions of the various tools, with links to the relevant files, documentation, and manuscripts. Both the wiki and the Resources section are organized thematically (e.g., structural, functional, diffusion), shown here in different colors. The website also includes a Community section, where NHP neuroimaging researchers and developers can connect and find relevant literature, and a Contribute section that explains how people can get involved in the PRIME-RE initiative through GitHub, the Wiki, or the Mattermost channel. B-D) Example workflows for the analysis of structural (B), functional (C), and diffusion (D) NHP neuroimaging data, indicating common analysis steps. The colored numbers refer to section numbers in this paper that describe tools and pipelines developed to address these analysis steps. Note that there is not a standardized workflow for these types of analyses and that individual researchers may include or omit certain steps and proceed in an order that is fitting for their own needs and circumstances.

Fig. 2.. Macaque Templates and Atlases.

The symmetric variant of the NIMH Macaque Template v2 (NMT v2), with tissue segmentation (top row) and anatomical regions (middle and bottom rows) depicted in color on a coronal slice (11 mm anterior to ear bar zero) and the pial surface (right column). The NMT v2 includes A) the population average volume; B) a manually refined brain mask that indicates which voxels contain brain tissue; C) a 5-class tissue segmentation that differentiates cerebrospinal fluid (green), gray matter (dark blue), white matter (light blue), subcortex (purple), and vasculature (red); and D) rendering of surfaces. The middle row shows the Cortical Hierarchy Atlas of the Rhesus Macaque (CHARM) and the bottom row shows the Subcortical Atlas of the Rhesus Macaque (SARM). The atlases provide anatomical labeling of cortical and subcortical regions, respectively, at six progressively finer spatial scales. Parcellations for E,I) Level 2, F,J) Level 4, and G,K) Level 6 are shown. H,L) Surface representation of the finest (Level 6) parcellations.

Fig. 3.. Marmoset atlases.

A-B) Example visualizations available on the website of the Marmoset Brain Mapping Project. A) 3D-viewer that demonstrates the connectivity profile of an area selected by clicking (here V4). B) Comparison with the weighted and directed connectivity graph based on the results of monosynaptic retrograde fluorescent tracer injections from the Marmoset Brain Connectivity Atlas (https://analytics.marmosetbrain.org/graph.html). The graph view highlights the spatial relations between connected areas. Clicking a node reveals further information, including an average connectivity profile, interactive visualizations of the data for each injection, and metadata. C-D) Marmoset Connectivity Atlas. C) Locations of 143 tracer injections registered to a stereotaxic template illustrated in a 2-dimensional flat-map of the marmoset cortex (C). D) The main data layers of the histology-based average morphology of the adult marmoset brain showcasing the interoperability of this dataset with neuroimaging research (T2* MR signal).

Fig. 4.. Segmentation tools.

A) Thresholdmann is a brain extraction tool that creates a mask from those voxels that are brighter than a spatially varying threshold. The threshold is constrained by control points (blue dots flagged with number arrows) that are added by clicking at locations in the web viewer (left). Adjustable sliders to the right control the threshold in the vicinity of each control point. In the examples shown to the right, the mask initially (top) excludes ventral brain areas but is filled out (bottom) by the addition of ventral control points. B) Brainbox is a web-based tool for collaborative brain segmentation. In the interface, each MRI (left) has a page where information on all projects utilizing this dataset is centralized. Under project settings (right), users can assemble datasets into projects, add collaborators, manage access rights, and add annotation layers. C) The UNet tool uses a neural network model approach for brain extraction. Local accuracy at each voxel is estimated by the regional Dice coefficient (top). Bottom panels show results for three example subjects from the PRIME-DE repository, with regions assessed to be brain tissue shown in red.

Fig. 5.. Anatomical MRI pipelines.

A) Six sagittal slices illustrating the D99 atlas transformed to the native space of a macaque subject using @animal_warper. B) CIVET-macaque results for the D99 subject. Top row (left to right): anatomical scan, brain mask (red) from brain extraction, tissue classification (CSF in dark blue, cortical GM in green, portions of non-cortical GM in white, WM and other non-cortical GM in yellow) from segmentation, and generated cortical surfaces overlaid on the anatomical scan (WM surface in red, pial surface in green). Bottom row (left to right): 3D-renderings of the white and pial surfaces; cortical thickness indication from morphometric analysis (1.0 mm in light blue to 3.5 mm in red) and D99 surface parcellation (atlas) viewed on the pial surface. C) PREEMACS results. Top row: brainmask (red) created by a Deep Learning convolutional neural network model (left) and volumetric tissue segmentation (right). Middle row: white matter and pial surface estimation (left) and rendered white matter surface (right). Bottom row left: surface registration to the PREEMACS Rhesus parameterization template to obtain vertex correspondence (i.e., surface coregistration) between subjects. Bottom row right: individual cortical thickness estimation from morphometric analysis. The resulting surfaces can be analyzed in the geometric space of the PREEMACS average surface.

Fig. 6.. Functional MRI pipeline Pypreclin.

The Pypreclin pipeline has been validated with a range of input data (left), a broad set of template brains (middle) and auditory task data. Preprocessing steps incorporated in Pypreclin include slice-timing correction, susceptibility distortions correction, motion correction, reorientation, bias field correction, brain extraction, normalization, coregistration and smoothing (for more details see: Tasserie et al., 2020).

Fig. 7.. Diffusion-weighted MRI.

A) Result from the Diffusion-MRI resource using macaque data. Fiber orientation distributions (left) and tractography (right). B) Examples of DTI-based tractographic output in a high-resolution macaque dataset in AFNI, following distortion correction with TORTOISE. Both images show a (frontal) coronal view of whole brain tracking using AFNI’s 3dTrackID: the left panel shows the results of deterministic tracking (as tract fibers), and the right panel shows the results of probabilistic tracking (as WM volume surfaces). Coloration reflects the orientation of the local structure relative to the coordinate frame, where red, blue and green are parallel to the x-, y- and z-axes, respectively. In the probabilistic panel, the FA value modulates the brightness (higher FA is brighter). The images are displayed using SUMA.