LONI Probabilistic Brain Atlas (LPBA40)


The LPBA40 data are available for download on the LONI Atlas Site.


Shattuck DW, Mirza M, Adisetiyo V, Hojatkashani C, Salamon G, Narr KL, Poldrack RA, Bilder RM, Toga AW (2008) Construction of a 3D Probabilistic Atlas of Human Brain Structures NeuroImage 39(3):1064-1080 PMCID: PMC2757616


The LONI Probabilistic Brain Atlas (LPBA40) is a series of maps of brain anatomic regions. These maps were produced from a set of whole-head MRI of 40 human volunteers. Each MRI was manually delineated to identify a set of 56 structures in the brain, most of which are within the cortex. These delineations were then transformed into a common atlas space to produce a set of coregistered anatomical labels. The original MRI data were also transformed into the atlas space. The 3D volumes contained within this data set represent:

  • Intensity averages of the co-registered whole-head and skull-stripped MRI volumes.

  • Probabilistic tissue maps based on automated classification of the native-space MRI into white matter, grey matter, and cerebrospinal fluid.

  • Probabilistic maps for each structure.

  • Probabilistic maps for each structure confined to grey matter voxels.

  • Maximum likelihood maps that identify the most likely structure at each voxel in the atlas space.

A brief video describing the LPBA40.



MRI data from 40 subjects were preprocessed according to existing LONI protocols to produce skull-stripped brain volumes. These volumes were aligned to the ICBM305 atlas using rigid-body transformation to correct for head tilt and reduce bias in the delineation process. This produced a transform from native space to delineation space (native.to.delineation) and an associated inverse transform (delineation.to.native). In each of the 40 subjects, 56 structures were delineated manually according to protocols developed for this project (see LPBA40 Protocols) using BrainSuite.

Image processing

Brain masks were constructed from the manual delineations and projected back to the native delineation space. The MRI in native space were masked to produce brain-only MRI volumes. These volumes were then corrected for non-uniformity using BrainSuite’s Bias Field Corrector. The bias-corrected brain data were processed with PVC (tissue classification) software to produce maps of grey matter, white matter, and cerebrospinal fluid for each subject volume.

Each corrected brain volume was aligned from its native space to the ICBM-452 template using AIR alignlinear. This produce a native space to ICBM452 space transform (native.to.icbm452). Volumes were cropped prior to registration (pad: 8); the cropped volumes were aligned to the ICBM-452 atlas (settings: -m 12 –t1 1 –t2 1). The corrected brain volumes were resliced into the ICBM452 space using chirp-Z interpolation (AIR).

Atlas construction

Delineation space to ICBM452 space transforms (delineation.to.icbm452) were computed using combineair to composite (delineation.to.native) and (native.to.icbm452). The delineated labels were resampled into the ICBM452 space using AIR's reslice with nearest-neighbor interpolation. The whole-head MRI and skull-stripped/bias-corrected MRI were resampled into ICBM452 space using AIR's reslice with chirp-Z interpolation to reduce resampling artifacts. The tissue class labels from PVC were transformed using reslice with nearest-neighbor interpolation.

We constructed a series of maps from the transformed data (each of which is available from this site):

  • Intensity atlas (LPBA40intensityavg) - a voxel-wise average of the skull-stripped MRI volumes in ICBM452 space

  • Whole-head intensity atlas (LPBA40wholeheadavg) - a voxel-wise average of the whole-head MRI volumes in ICBM452 space

  • Tissue class probabilistic atlas (LPBA40tissue) - three volumes, where each voxel contains a count of the number of subject volumes that had that voxel labeled as a the associated tissue type (GM, WM, or CSF)

  • Structure probability maps (LPBA40structure) - 56 volumes, each of which contains the voxel-wise count of the number of subjects that had that voxel labeled as the given structure.

  • Grey matter masked structure probability maps (LPBA40GMstructure) - 56 volumes, each of which contains the voxel-wise count of the number of subjects that had that voxel labeled as the given structure and also as GM.

  • Maximum likelihood atlas(LPBA40) - for each voxel, we computed the most likely tissue type based on the 40 delineated subject volumes; in the case of a tie, the tissue type was chosen at random from the equilikely labels. This data set contains both the maximum likelilood label volume and a volume composed of the maximum count at each voxel.

  • Maximum likelihood grey matter atlas(LPBA40) - This is similar to LPBA40, above, but the tissue labels were screened by the grey matter voxels before performing the maximum likelihood computation.

  • The dataset also includes LPBA40_labelid.txt, which provides the name of each structure and the integer ID number used in the label volumes.