Great work on MindBoggle!
I may have missed this in the accompanying publications, but how is the jointfusion volume atlas derived (e.g. for OASIS-TRT-20_jointfusion_DKT31_CMA_labels_in_MNI152_v2.nii)? Is it based on one brain, or on an average of multiple brains?
Thank you very much! And thank you for pointing out that the atlas description is missing. I used to have the following description on the mindboggle data web page (http://www.mindboggle.info/data.html) and should bring it back:
OASIS-TRT-20 joint fusion atlas in OASIS-30 and in MNI152 space (2013)
Corresponding label probabilities in OASIS-30 and in MNI152 space
Joint fusion (Hongzhi Wang, 2013; distributed with ANTs) was used to construct probabilistic label volumes incorporating the 20 OASIS-TRT brains, and from these a single maximum probability atlas with the winning labels. Joint fusion was performed on the set of 20 brains after antsRegistration warped them to the brain and cerebellum OASIS-30 Atropos template. Results are also given after affine transformation to the OASIS-30 Atropos template in MNI152 space.
Thank you for your quick response!
Just to make sure I understand it correctly:
You labeled 20 brains, then from these determined label probabilities; the labels in OASIS-TRT-20_jointfusion_DKT31_CMA_labels_in_MNI152_v2.nii then correspond to the winning labels (based on some cutoff - which one?) of the probabilistic labels?
Did I get this right?
Tobias
Correct. The “winning” labels are the maximum probability labels, so if the relative majority (say 17 of 20 votes) for a given voxel is the right amygdala, that is the label assigned to that voxel. See Hongzhi Wang’s 2013 paper for more details about the joint fusion algorithm.
Sorry reviving an old thread here. Did you release the code you used to build OASIS-TRT-20 joint fusion atlas?
I’m interested in doing something similar with different label maps and it would be good to see your code and to check various parameters for ants calls.
Thanks!