Brain Atlases

Overview

Atlases reach TI-Toolbox two different ways, and the distinction matters when you define an ROI:

  • MNI-space atlases are shipped with the toolbox as ready-to-use NIfTI volumes in resources/atlas/. They are the same seven volumes for every user, and SimNIBS transforms a selected label into your subject’s space at run time.
  • Subject-space atlases are generated per subject during preprocessing by SimNIBS charm/subject_atlas and FastSurfer or optional FreeSurfer. Existing FreeSurfer outputs are also read. Nothing is shipped — they only exist once you have run the pipeline on a head model. The viewer above uses the ernie example subject so you can see what they look like.

Both families are queried through the same ROI picker used in flex-search, ex-search, and the analyzer.

  MNI-space Subject-space
Shipped with the toolbox Yes, in resources/atlas/ No — generated per subject
Same for every user Yes No
Produced by Bundled with TI-Toolbox charm, FastSurfer, optional FreeSurfer

MNI-Space Atlases

TI-Toolbox ships seven atlases as MNI-space NIfTI volumes. With Target space set to MNI, the 3D pane draws them on the MNI152 template and clicking a region adds it to the target: Harvard-Oxford cortical, Schaefer and Glasser as regions on the template’s central surfaces (the pane opens on Harvard-Oxford cortical), and the sub-cortical atlases as surfaces of their own volumes, which Explode pulls apart. A region’s outline on the surface comes from 1 mm voxels. Whichever way it is drawn, the target is the label volume.

Atlas Regions Native space
CIT168 Subcortical (left/right) 32 MNI152NLin2009cAsym
Glasser HCP-MMP1.0 360 FreeSurfer-conformed 256x256x256 1mm grid
MASSP Subcortical 31 ICBM152 2009b nonlinear asymmetric, hi-res 0.5mm
Harvard-Oxford cortical, lateralized 96 MNI152NLin6Asym, FSL 182x218x182 1mm grid
Harvard-Oxford subcortical 21 MNI152NLin6Asym, FSL 182x218x182 1mm grid
Cerebellum-MNIfnirt (Diedrichsen 2009) 28 MNI152NLin6Asym, FSL 182x218x182 1mm grid
Schaefer 2018, 400 parcels / 7 networks 400 MNI152NLin6Asym, FSL 182x218x182 1mm grid

CIT168 Subcortical Atlas

Pauli WM, Nili AN, Tyszka JM. A high-resolution probabilistic in vivo atlas of human subcortical brain nuclei. Scientific Data 5:180063 (2018). doi:10.1038/sdata.2018.63.

The shipped volume is a deterministic label map, derived locally from the paper’s probabilistic masks by winner-takes-highest-probability at a 0.05 minimum-probability threshold. The probability maps themselves are not shipped — only the resulting hard segmentation.

The published labeling is bilateral (16 labels, each covering a structure in both hemispheres). TI-Toolbox ships it split at the midline (x = 0 mm of its own template): 32 lateralized labels, where the paper’s label k is 2k-1 (Left-…, x < 0) and 2k (Right-…, x ≥ 0), made by dev/build_lateralized_atlases.py. Nothing else about the map changes. The viewer’s region table lists all 32.

Glasser HCP-MMP1.0 Atlas

Glasser MF, Coalson TS, Robinson EC, et al. A multi-modal parcellation of human cerebral cortex. Nature 536(7615):171-178 (2016). doi:10.1038/nature18933.

360 labels (180 per hemisphere), using the atlas’s native sparse indexing scheme: left-hemisphere regions are numbered 1-180, right-hemisphere regions 1001-1180. The shipped volume is natively on a FreeSurfer-conformed 256x256x256 1mm grid, not the 182x218x182 MNI152 grid, so it was resampled onto the template grid for the viewer.

MASSP Subcortical Parcellation

31 labels, natively at 0.5mm resolution in ICBM152 2009b nonlinear-asymmetric hi-res space — the highest native resolution of the shipped atlases. It was resampled onto the 1mm template grid for the viewer.

Other shipped MNI atlases

These three are not in the viewer.

Harvard-Oxford cortical and subcortical (HarvardOxford-cortl-maxprob-thr25-1mm.nii.gz: FSL’s lateralized cortical map, 96 labels = 48 structures x left/right, odd = left, even = right; HarvardOxford-sub-maxprob-thr25-1mm.nii.gz: 21 subcortical labels). FSL’s own files, unmodified (NeuroDebian fsl-harvard-oxford-cortical-lateralized-atlas and fsl-harvard-oxford-atlases, both 5.0.7-2); ids = XML index + 1 as FSL’s maxprob images encode them. The subcortical map labels thalamus, caudate, putamen, pallidum, hippocampus, amygdala, accumbens and brain stem per hemisphere, plus whole-hemisphere cortex / white matter and the ventricles (kept so the file is unmodified; they are not TI targets). Licence CC BY-SA 4.0: the FSL licence page states that “The Cerebellum and Harvard-Oxford atlases, whilst not being the property of Oxford, are released under the CC BY-SA 4.0 licence”.

Cerebellum-MNIfnirt (Cerebellum-MNIfnirt-maxprob-thr25-1mm.nii.gz, 28 labels: lobules I-IV to X, left / right / vermis; FNIRT-normalised maximum-probability map at 25 %, 1 mm). Diedrichsen J, Balsters JH, Flavell J, Cussans E, Ramnani N. A probabilistic MR atlas of the human cerebellum. NeuroImage 46(1):39-46 (2009). doi:10.1016/j.neuroimage.2009.01.045. CC BY-SA 4.0 by the same FSL sentence.

Schaefer 2018, 400 parcels / 7 networks (Schaefer2018_400Parcels_7Networks_order_FSLMNI152_1mm.nii.gz from ThomasYeoLab/CBIG, MIT, with CBIG’s own LUT; 1-200 left, 201-400 right, named by Yeo 7-network membership, e.g. 7Networks_LH_Vis_1). Schaefer A, Kong R, Gordon EM, Laumann TO, Zuo X-N, Holmes AJ, Eickhoff SB, Yeo BTT. Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cerebral Cortex 28(9):3095-3114 (2018). doi:10.1093/cercor/bhx179. Like Glasser it is a cortical parcellation distributed as a label volume, so you target it through the picker’s Subcortical mode.

Subject-Space Atlases

These are not shipped. They are produced for each subject during preprocessing, and live under that subject’s own derivatives:

  • SimNIBS charm writes the tissue segmentation m2m_{subject}/segmentation/labeling.nii.gz, plus the surface parcellations lh/rh.{subject}_DK40.annot, _a2009s.annot and _HCP_MMP1.annot.
  • FastSurfer --seg_only (v3) writes aparc.DKTatlas+aseg.deep.mgz. It does not produce thalamic nuclei or hippocampal/amygdala subregion atlases. See FastSurfer / FreeSurfer.
  • Optional FreeSurfer recon-all produces the volumetric parcellations aparc.DKTatlas+aseg.mgz, aparc.a2009s+aseg.mgz, aparc+aseg.mgz and aseg.mgz, with finer thalamic nuclei and hippocampal/amygdala subregions available as additional operations after reconstruction. Existing ThalamicNuclei.v13.T1.mgz and lh/rh.hippoAmygLabels-T1.v22.mgz outputs remain readable.

Read the coordinates as subject space. The centroid column here is in ernie’s own scanner/world coordinates, not MNI. The same structure will sit at different coordinates in your own subject.

CHARM Tissue Labeling

Produced by SimNIBS charm as m2m_{subject}/segmentation/labeling.nii.gz, with its own labeling_LUT.txt alongside. This is the tissue segmentation the head model is built from — grey and white matter, CSF, skull, scalp and the rest — not a cortical parcellation. It is what the analyzer reads when you pick a tissue-level ROI in voxel space.

Desikan-Killiany-Tourville (aparc.DKTatlas+aseg)

The DKT cortical parcellation combined with the aseg subcortical segmentation in one volume. By default FastSurfer writes it as aparc.DKTatlas+aseg.deep; optional FreeSurfer recon-all writes the equivalent aparc.DKTatlas+aseg.mgz, which is the volume shown here for ernie. Klein A, Tourville J. 101 labeled brain images and a consistent human cortical labeling protocol. Frontiers in Neuroscience 6:171 (2012). doi:10.3389/fnins.2012.00171.

Cortical labels use FreeSurfer’s 1000+/2000+ convention (left/right hemisphere); subcortical structures keep their aseg ids below 100. This is one of the voxel atlases the analyzer offers.

Destrieux (aparc.a2009s+aseg)

Optional FreeSurfer recon-all output (FastSurfer’s segmentation-only run does not write it) combining the Destrieux cortical parcellation with aseg. Destrieux C, Fischl B, Dale A, Halgren E. Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature. NeuroImage 53(1):1-15 (2010). doi:10.1016/j.neuroimage.2010.06.010.

The finest of the cortical parcellations here, splitting cortex into gyral (G_) and sulcal (S_) units. Cortical ids run in the 11100+/12100+ range.

Licence, attribution and citation

Atlases are other people’s data. If you publish work that used one, cite it; if you redistribute TI-Toolbox, these are the terms you are redistributing under. Each row is also machine-readable in resources/atlas/manifest.json.

Atlas Licence May be redistributed Cite
CIT168 Subcortical CC BY 4.0 Yes Pauli W. M., Nili A. N., Tyszka J. M. Scientific Data 5:180063 (2018). doi:10.1038/sdata.2018.63
MASSP 2021 Subcortical CC BY 4.0 Yes Bazin P.-L. et al. eLife 9:e59430 (2020); atlas release doi:10.21942/uva.19646328
Glasser HCP-MMP1.0 WU-Minn HCP Open Access Data Use Terms Yes, under those same terms Glasser M. F. et al. Nature 536:171-178 (2016). doi:10.1038/nature18933
Harvard-Oxford cortical / subcortical CC BY-SA 4.0 (FSL licence page) Yes, share-alike Desikan R. S. et al. NeuroImage 31:968-980 (2006); Frazier J. A. et al. Am J Psychiatry 162:1256-1265 (2005); Makris N. et al. Schizophr Res 83:155-171 (2006)
Cerebellum-MNIfnirt CC BY-SA 4.0 (same FSL sentence) Yes, share-alike Diedrichsen J. et al. NeuroImage 46:39-46 (2009). doi:10.1016/j.neuroimage.2009.01.045
Schaefer 2018 400/7 MIT (CBIG) Yes Schaefer A. et al. Cerebral Cortex 28:3095-3114 (2018). doi:10.1093/cercor/bhx179
MNI152 T1 1mm template MNI/McGill permissive (any purpose, without fee, keep the notice) Yes Copyright (C) 1993-2009 Louis Collins, McConnell Brain Imaging Centre, MNI, McGill University

Using Glasser also carries an acknowledgement: “Data were provided [in part] by the Human Connectome Project, WU-Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University.”

The full notices, with copyright holders, are in the repository’s NOTICE file.

Every shipped atlas names one hemisphere per label, apart from true midline structures (cerebellar vermis, brain stem, third and fourth ventricle, fornix). Picking “Left-Putamen” targets the left putamen only; to target both, pick both.