Allen Human Reference Atlas - 3D, 2020 (new!)


Version 1.0.0

The Allen Human Reference Atlas - 3D is a parcellation of the adult human brain in 3D, labeling every voxel with a brain structure spanning 141 structures. These parcellations were drawn by Song-Lin Ding, and adapted from his prior 2D version of an adult human brain atlas.

These parcellations were drawn on the MRI reference brain volume “ICBM 2009b Nonlinear Symmetric”, a non-linear average of the MNI152 database of 152 normal brain images. The iterative procedure results in an average that combines both high-spatial resolution and signal-to-noise and not subjected to any particularity of any single brain. To obtain the reference volume, please refer to the McConnell Brain Imaging Center website for download and terms of use available from; Copyright © 1993–2004 Louis Collins, McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University.

A 2D sagittal cross-section overlaid on MNI space and 3D medial and lateral rendering of the new human parcellation volume visualized using ITK-SNAP application.


Example python scripts have been included to demonstrate:

  • how to download structure information and ontology from the Allen Brain API
  • use of the annotation volume in context of a hierarchical ontology
  • creation of ITK-SNAP compatible files to visualize the parcellation in 2D and 3D


These materials are provided under the Allen Institute Terms of Use, which are available at


Citation of these materials should conform to the Allen Institute Citation Policy:

Allen Human Reference Atlas - 3D, 2020 Citation information:

Resource Name: Allen Human Reference Atlas – 3D, 2020
Version: 1.0.0
Research Resource Identifier (RRID): RRID:SCR_017764
Copyright notice: © 2019 Allen Institute for Brain Science
Dataset citation: Song‐Lin Ding, Joshua J. Royall, Susan M. Sunkin, Benjamin A.C. Facer, Phil Lesnar, Amy Bernard, Lydia Ng, Ed S. Lein (2020). “Allen Human Reference Atlas – 3D, 2020," RRID:SCR_017764, version 1.0.0.
Available from:

The anatomic structure ontology adapted for use in the 3D atlas was based on the 2016 version of the 2D atlas, published in final form here.

Publication Citation for structure ontology: Ding, S.L., Royall, J.J., […], Lein, E.S. Comprehensive cellular‐resolution atlas of the adult human brain. Journal of Comparative Neurology, Volume 524:16, pages 3127–3481, 1 November 2016, DOI 10.1002/cne.24080.


These materials are provided as-is, without direct support. Community discussion around this resource is available here at


Creation of the 3D parcellation volume was supported by the Allen Institute for Brain Science, and by the National Institute of Mental Health under Award Number 1U01MH114812-01 (PI: Ed Lein, Allen Institute for Brain Science).

The 2D atlas and anatomic structural ontology upon which it was based was supported by the Allen Institute for Brain Science, and by the National Institute of Mental Health under Award Number RC2MH089921 (PIs: Ed Lein & Michael Hawrylycz, Allen Institute for Brain Science).


That’s awesome!

Could you share a link to the example python script please, I don’t seem to able to find them on the API and SDK support pages.

Thank you,

Hi Federico, thanks for your interest!
The scripts are in the download directory at the moment - check the “examples” folder:

Hi Carol. my name is Gaston Zanitti. I am a PhD student/researcher working for the “Institut national de recherche en sciences et technologies du numérique” (INRIA) of France.

This work looks very interesting! I would like to use this information in a project that we are developing, but I would like to know if there is a way to relate the voxels with the refinements of the regions that the ontology offers. As far as I could see in the examples, it’s only possible to relate 140 regions per hemisphere.

Is there any way to distinguish the sub-regions of the ontology? Or did you make some other work in this direction?

Thank you very much!

Hi Gaston,

Hope you are safe and well, and thank you for your question!

This release represents a very deep anatomical dive relative to the resolution of this template volume (0.5mm/voxel). Further delineation is in theory possible, but the returns would be starkly diminishing. Averaging artifacts, modest intrinsic anatomical contrast, and a lack of directly registered supporting data are among other reasons our natural inclination to delineate deeper were tempered. Reaching our totals after an original survey suggested a ~100 structure maximum was a challenge.

To meaningfully relate sub-regions of the ontology not present in the new 3D atlas, we suggest utilizing the interactive adult human reference atlase in parallel. It carries all the classic features of a high-resolution atlas (and more) while employing the same ontology. If there is a particular area or analysis or region that you had in mind that requires consultation, please let us know.


Are there files for the other hemisphere as well?


A double-sided version of the annotation is now available for download.

1 Like

is there a specific reason why the structure IDs refer to the developing brain atlas ontology (graph id 16) and not to the adult human ontology (graph id 10)? Is there an easy liftover from graph 16 to graph 10 ids, or can we just match acronyms/names?

Hi @gdagstn,

The BrainSpan references atlases is a cohesive set references across multiple developmental timepoints including adult using multi-modal data under the guidance of Drs Song-Lin Ding and Gulgun Sengul. This the same annotation framework that Dr Song-Lin Ding carried over to the 3D. See whitepaper for details

The “adult human ontology” is associated with the Allen Human Brain Atlas (a project which preceded the BrainSpan project). The “adult human ontology” was developed Dr Angela Guillozet-Bongaarts to be a practical neuroanatomical guide for sampling and was not intended to be view as a definitive reference atlas.

Is there an easy liftover from graph 16 to graph 10 ids, or can we just match acronyms/names?

In cases like this, I feel the best practice is to abide each anatomist’s intent.
Both ontologies were annotated on the same set of Nissl to support this analysis. So it is possible to compare the SVGs to see what overlaps. Note that you are likely to find 1 to N relationships and also structure from one only partially overlapping the structure of the other.

  • “Human, 34 years, Cortex - Gyral” 138322605
  • "“Human, 34 years, Cortex - Mod. Brodmann” 265297126
  • "Human Brain Atlas Guide " 265297125
1 Like

Thank you so much for sharing this.
The python script is beautiful!

Do you think that a Nifti version (similar to the Gyral) but for the Brodmann segmentation is possible?
Even at a lower resolution?
Honestly, it would be amazing for the scientific community!



Does anyone know where to get the anatomical names for each region. When I download the STL files, they have things like:

I am looking for a list that maps the names in the download listed above with the proper anatomical names for the brain region.



Take a look at the voxel_count example that shows you how to connect to the API to get structure acronym and name. Or skip to just looking at the voxel_count.csv file for spreadsheet version.

1 Like

@lydian thank you. One more follow up question. I think your response was pertaining to the nifti data? I was actually talking about the ITK Snap data with the surfaces already segmented out and export. During the export step 5) here (Human Brain Atlas Mesh Files - #2 by nileg) each region is exported as a different file which can be loaded up into blender. When this happens the shapes lose their names. I can manually look at itksnap and the blender file and determine the region, I was just checking if this list already existed. Thanks,


I see - you will need to join a couple of data files together.

This file label description file has some of the information you need.

The first column is the label id, the should correspond to the number in your filename (+ leading zeros). The last column is the “structure_acronym - database_id”. You can use the acronym or datasbase id to index into the voxel_count.csv file to get fullname.

1 Like