alvarodeleglise / easy_lausanne

Easy creation of Lausanne2008 atlases

Easy Lausanne

This is an extremely stripped-down version of the connectome mapper, all it does is create the Lausanne2008 parcellations from an existing FreeSurfer directory and align them to a target volume (BOLD or B0) using bbregister.

update 5/8/20 This software was created with an old version of python and freesurfer. There will likely be problems with newer versions of either, and PRs would be welcome! If you make one, please add yourself to the Authors section below. update 7/10/15 you can use the --run_500 option to generate scale500 along with the other scales

Prerequisites:

• FreeSurfer
• FSL
• numpy/scipy/nibabel/networkx

All the hard work was done by the LTS5 folks and they should be credited if you use these atlases in your work.

A. Daducci, S. Gerhard, A. Griffa, A. Lemkaddem, L. Cammoun, X. Gigandet, R. Meuli, P. Hagmann and J.-P. Thiran The Connectome Mapper: An Open-Source Processing Pipeline to Map Connectomes with MRI. Plos One 7(12):e48121 (2012)

Installation

$ git clone https://github.com/mattcieslak/easy_lausanne.git
$ cd easy_lausanne
$ python setup.py install

This installs the easy_lausanne and atlas_dilate executables on your path.

Example use

Assuming FreeSurfer is installed, you've run the Freesurfer setup script, and recon-all has completely finished for "SUBJECT", you use easy_lausanne to create the Lausanne2008 atlases aligned to VOLUME.nii.gz for "SUBJECT"

$ easy_lausanne \
     --subject_id SUBJECT \
     --target_volume /path/to/VOLUME.nii.gz \
     --target_type diffusion OR bold \
     --output_dir /where/you/want/results

This python package also includes a utility for dilating these cortical regions. You can use atlas_dilate like so.

$ atlas_dilate input_atlas.nii.gz thickened_by_1_voxel.nii.gz

Here is the original ROIv_scale33 from a subject

And here is the dilated version

atlas_dilate works by finding all non-labeled voxels that neighbor at least one labeled voxel. It extracts the labels from all neighboring voxels and fills the originally-zero voxel with the statistical mode of its neighbors. No labeled voxel is ever overwritten with a new value.

Notes

If registering to a BOLD image, the --target_type argument should be "bold". If it's a B0 volume from a DWI, then use "diffusion". If there is poor alignment between the FreeSurfer surfaces and the B0 volume, you can export the native space GFA/QA/FA image and specify --target_type anisotropy.

Proof of usability

I ran a DSI dataset through the connectomemapper and resampled the Lausanne2008 labels into native B0 space. easy_lausanne was then used on a clean copy of this data to generate the atlases directly in B0 space. Here are plots of the overlap in voxels and the relative sizes of each region for the cmp-generated and easy_lausanne-generated atlases.

Voxelwise labeling analysis (blue)

Here each non-zero voxel was compared between the two atlases. The label value in each voxel is plotted.

Region size comparison (red)

The proportion of nonzero voxels for each region label is plotted for both versions of the Lausanne2008 atlas.

Plots

Credits

This source code was sponsored by the U.S. Army Research Office and the content of the information does not necessarily reflect the position or the policy of the government, and no official endorsement should be inferred.

Authors

• Matt Cieslak
• Scott Grafton