Eve Atlas used for segmentations
This repository contains the "Eve Atlas" from the Johns Hopkins University School of Medicine, Department of Radiology, Center for Brain Imaging Science. It is based on a single-subject data as described in Oishi et al, 2009. There are co-registered T1 (MPRAGE), T2, with associated atlases and look up tables (LUT files). The original format is nrrd
, which was converted to NIfTI
using Slicer
. All original images have 181x217x181 / 1x1x1 mm dimensions.
The 2 files sct1_unsmooth.nii.gz
and betsct1_unsmooth.nii.gz
correspond to the T1 and skull-stripped (with BET) T1 image from the Clinical toolbox for SPM from Christopher Rorden as described in doi: 10.1016/j.neuroimage.2012.03.020. This is referred to as the Rorden template.
The Eve T1-weighted image was skull-stripped using BET. This image was transformed into the space of Rorden template and MNI 152 (6th generation) Template included with FSL.
Two methods were used for transforming the data: Symmetric Normalization (SyN) doi: 10.1016/j.media.2007.06.004 using the Advanced Normalization Tools (ANTs) and the associated R package ANTsR and Deformable Registration via Attribute Matching and Mutual-Saliency Weighting (DRAMMS) doi: 10.1016/j.media.2010.07.002.
The estimated transformation was used to warp the atlases to the associated space and a nearest-neighbor interpolation was used to create labels.
The Eve atlas has been used previously to describe anatomical locations of the brain. One problem is that the Eve atlas and the MNI 152 from FSL (and other MNI data) are not necessarily located in the same space. See Dr. Mori's discussion of this issue with the dimensions:
Hello,
The ICBM-152 Atlas that comes from the UCLA web site has dimensions of 182 x 218 x 182 . Within DTI Studio it is 181 x 217 x 181. There seems to be a single slice difference between both templates? Where would the single slice need to be added for both templates to match?
Regards,
Arun
and the response
Hi Arun,
It is likely that the difference is just 0 padding at the 182th, 218th, and 182th line. You can convert our 181x217x181 atlas to 182x218x182 using the "Crop" function in Landmarker. Another possibility is that it was interpolated, which can also be done by Landmarker. You may want to make sure by subtracting the images (you can do it in RoiEditor).
We also got ICBM-152 from UCLA and don't know why there are two versions with different dimensions.
Susumu
Now as you can see in
that they agree relatively well if you drop the x=182
, y=218
, and z=182
slices of the MNI template (or similarly zero pad the Eve template/atlas). For those who cannot manipulate images easily (such as dropping slices), this is a problem.
The possible solutions are as follows:
JHU_MNI_SS_T1_brain_182x218x182.nii.gz
, which has zeros padded to the last dimension of the imageMNI152_T1_1mm_brain_181x217x181.nii.gz
, which has the last slice of each dimension removedJHU_MNI_SS_T1_brain_to_MNI_brain_DRAMMS.nii.gz
, which has been non-linearly transformed to MNI using DRAMMS, and thus has the dimension of the MNI template.JHU_MNI_SS_T1_brain_to_MNI_brain.nii.gz
, which has been non-linearly transformed to MNI using SyN, and thus has the dimension of the MNI template.Unlike the MNI template, the Clinical Toolbox dimensions is 181x217x181, using the bounding box: [-90, -126, -72; 90, 90, 108]
. The problem, as demonstrated by
is that the images do not overlap very well. This is because the Rorden T1 template was not intended to match exactly the MNI template, as its purpose is to supercede the MNI template for aging subjects. Therefore, if using the Rorden template in combination with the Eve atlas, there may be some mismatch of cortical structures.
JHU_MNI_SS_T1_brain_to_Rorden_brain_DRAMMS.nii.gz
, which has been non-linearly transformed to Rorden T1 space using DRAMMS, and thus has the dimension of the Rorden template.JHU_MNI_SS_T1_brain_to_Rorden_brain.nii.gz
, which has been non-linearly transformed to Rorden T1 space using SyN, and thus has the dimension of the Rorden template.