NeuroData's MR Graphs package, ndmg (pronounced "nutmeg"), is a turn-key pipeline which uses structural and diffusion MRI data to estimate multi-resolution connectomes reliably and scalably.
- System Requirements
- Installation Guide
- Working with S3 Datasets
- Example Datasets
- Manuscript Reproduction
The ndmg pipeline has been developed as a one-click solution for human connectome estimation by providing robust and reliable estimates of connectivity across a wide range of datasets. The pipelines are explained and derivatives analyzed in our pre-print, available on BiorXiv.
The ndmg pipeline:
- was developed and tested primarily on Mac OSX, Ubuntu (12, 14, 16, 18), and CentOS (5, 6);
- made to work on Python 3.6;
- is wrapped in a Docker container;
- has install instructions via a Dockerfile;
- requires no non-standard hardware to run;
- has key features built upon FSL, AFNI, Dipy, Nibabel, Nilearn, Networkx, Numpy, Scipy, Scikit-Learn, and others;
- takes approximately 1-core, 8-GB of RAM, and 1 hour to run for most datasets.
ndmg relies on FSL, AFNI, Dipy, networkx, and nibabel, numpy scipy, scikit-learn, scikit-image, nilearn. You should install FSL and AFNI through the instructions on their website, then install other Python dependencies as well as the package itself with the following:
git clone https://github.com/neurodata/ndmg.git cd ndmg pip install -r requirements.txt pip install .
For the most up-to-date version of ndmg, you can use the staging branch. This version is updated regularly, but may be less stable:
git clone https://github.com/neurodata/ndmg.git cd ndmg git checkout staging pip install -r requirements.txt pip install .
You can also install ndmg from
pip as shown below. Installation shouldn't take more than a few minutes, but depends on your internet connection. Note that to install the most up-to-date version of the pipeline, we currently recommend installing from github.
Install from pip
pip install ndmg
ndmg is available through Dockerhub, and the most recent docker image can be pulled using:
docker pull neurodata/ndmg_dev:latest
The image can then be used to create a container and run directly with the following command (and any additional options you may require for Docker, such as volume mounting):
docker run -ti --entrypoint /bin/bash neurodata/ndmg_dev:latest
ndmg docker containers can also be made from ndmg's Dockerfile.
git clone https://github.com/neurodata/ndmg.git cd ndmg docker build -t <imagename:uniquelabel> .
Where "uniquelabel" can be whatever you wish to call this Docker image (for example, ndmg:latest). Additional information about building Docker images can be found here. Creating the Docker image should take several minutes if this is the first time you have used this docker file. In order to create a docker container from the docker image and access it, use the following command to both create and enter the container:
docker run -it --entrypoint /bin/bash ndmg:uniquelabel
ndmg has particularly high RAM requirements, and docker containers are given low memory by default. We recommend changing the default preferences to give docker access to at least 6 GB of RAM.
Once you have the pipeline up and running, you can run it with:
ndmg <input_directory> <output_directory>
We recommend specifying an atlas and lowering the default seed density on test runs:
ndmg --seeds 1 --parcellation desikan <input_directory> <output_directory>
You can set a particular scan and session as well (recommended for batch scripts):
ndmg --seeds 1 --parcellation desikan --participant_label <label> --session_label <label> <input_directory> <output_directory>
For more detailed instructions, tutorials on the ndmg pipeline can be found in ndmg/tutorials
The organization of the output files generated by the ndmg pipeline are shown below. If you only care about the connectome edgelists (ndmg's fundamental output), you can find them in
File labels that may appear on output files, these denote additional actions ndmg may have done: RAS = File was originally in RAS orientation, so no reorientation was necessary reor_RAS = File has been reoriented into RAS+ orientation nores = File originally had the desired voxel size specified by the user (default 2mmx2mmx2mm), resulting in no reslicing res = The file has been resliced to the desired voxel size specified by the user /output /anat /preproc Files created during the preprocessing of the anatomical data t1w_aligned_mni.nii.gz = preprocessed t1w_brain anatomical image in mni space t1w_brain.nii.gz = t1w anatomical image with only the brain t1w_seg_mixeltype.nii.gz = mixeltype image of t1w image (denotes where there are more than one tissue type in each voxel) t1w_seg_pve_0.nii.gz = probability map of Cerebrospinal fluid for original t1w image t1w_seg_pve_1.nii.gz = probability map of grey matter for original t1w image t1w_seg_pve_2.nii.gz = probability map of white matter for original t1w image t1w_seg_pveseg.nii.gz = t1w image mapping wm, gm, ventricle, and csf areas t1w_wm_thr.nii.gz = binary white matter mask for resliced t1w image /registered Files created during the registration process t1w_corpuscallosum.nii.gz = atlas corpus callosum mask in t1w space t1w_corpuscallosum_dwi.nii.gz = atlas corpus callosum in dwi space t1w_csf_mask_dwi.nii.gz = t1w csf mask in dwi space t1w_gm_in_dwi.nii.gz = t1w grey matter probability map in dwi space t1w_in_dwi.nii.gz = t1w in dwi space t1w_wm_gm_int_in_dwi.nii.gz = t1w white matter-grey matter interfact in dwi space t1w_wm_gm_int_in_dwi_bin.nii.gz = binary mask of t12_2m_gm_int_in_dwi.nii.gz t1w_wm_in_dwi.nii.gz = atlas white matter probability map in dwi space /dwi /fiber Streamline track file(s) /preproc Files created during the preprocessing of the dwi data #_B0.nii.gz = B0 image (there can be multiple B0 images per dwi file, # is the numerical location of each B0 image) bval.bval = original b-values for dwi image bvec.bvec = original b-vectors for dwi image bvecs_reor.bvecs = bvec_scaled.bvec data reoriented to RAS+ orientation bvec_scaled.bvec = b-vectors normalized to be of unit length, only non-zero b-values are changed eddy_corrected_data.nii.gz = eddy corrected dwi image eddy_corrected_data.ecclog = eddy correction log output eddy_corrected_data_reor_RAS.nii.gz = eddy corrected dwi image reoriented to RAS orientation eddy_corrected_data_reor_RAS_res.nii.gz = eddy corrected image reoriented to RAS orientation and resliced to desired voxel resolution nodif_B0.nii.gz = mean of all B0 images nodif_B0_bet.nii.gz = nodif_B0 image with all non-brain matter removed nodif_B0_bet_mask.nii.gz = mask of nodif_B0_bet.nii.gz brain tensor_fa.nii.gz = tensor image fractional anisotropy map /roi-connectomes Location of connectome(s) created by the pipeline, with a directory given to each atlas you use for your analysis /tensor Contains the rgb tensor file(s) for the dwi data if tractography is being done in MNI space /qa /adjacency /fibers /graphs /graphs_plotting Png file of an adjacency matrix made from the connectome /mri /reg /tensor /tmp /reg_a Intermediate files created during the processing of the anatomical data mni2t1w_warp.nii.gz = nonlinear warp coefficients/fields for mni to t1w space t1w_csf_mask_dwi_bin.nii.gz = binary mask of t1w_csf_mask_dwi.nii.gz t1w_gm_in_dwi_bin.nii.gz = binary mask of t12_gm_in_dwi.nii.gz t1w_vent_csf_in_dwi.nii.gz = t1w ventricle+csf mask in dwi space t1w_vent_mask_dwi.nii.gz = atlas ventricle mask in dwi space t1w_wm_edge.nii.gz = mask of the outer border of the resliced t1w white matter t1w_wm_in_dwi_bin.nii.gz = binary mask of t12_wm_in_dwi.nii.gz vent_mask_mni.nii.gz = altas ventricle mask in mni space using roi_2_mni_mat vent_mask_t1w.nii.gz = atlas ventricle mask in t1w space warp_t12mni.nii.gz = nonlinear warp coefficients/fields for t1w to mni space /reg_m Intermediate files created during the processing of the diffusion data dwi2t1w_bbr_xfm.mat = affine transform matrix of t1w_wm_edge.nii.gz to t1w space dwi2t1w_xfm.mat = inverse transform matrix of t1w2dwi_xfm.mat roi_2_mni.mat = affine transform matrix of selected atlas to mni space t1w2dwi_bbr_xfm.mat = inverse transform matrix of dwi2t1w_bbr_xfm.mat t1w2dwi_xfm.mat = affine transform matrix of t1w_brain.nii.gz to nodif_B0.nii.gz space t1wtissue2dwi_xfm.mat = affine transform matrix of t1w_brain.nii.gz to nodif_B0.nii.gz, using t1w2dwi_bbr_xfm.mat or t1w2dwi_xfm.mat as a starting point xfm_mni2t1w_init.mat = inverse transform matrix of xfm_t1w2mni_init.mat xfm_t1w2mni_init.mat = affine transform matrix of preprocessed t1w_brain to mni space
Other files may end up in the output folders, depending on what settings or atlases you choose to use. Using MNI space for tractography or setting
True will result in fewer files.
The ndmg pipeline can be used to generate connectomes as a command-line utility on BIDS datasets with the following:
ndmg /input/bids/dataset /output/directory
Note that more options are available which can be helpful if running on the Amazon cloud, which can be found and documented by running
If running with the Docker container shown above, the
entrypoint is already set to
ndmg, so the pipeline can be run directly from the host-system command line as follows:
docker run -ti -v /path/to/local/data:/data neurodata/ndmg_dev /data/ /data/outputs
This will run ndmg on the local data and save the output files to the directory /path/to/local/data/outputs. Note that if you have created the docker image from github, replace
Also note that currently, running
ndmg on a single bids-formatted dataset directory only runs a single scan. To run the entire dataset, we recommend parallelizing on a high-performance cluster or using
ndmg's s3 integration.
Working with S3 Datasets
ndmg has the ability to work on datasets stored on Amazon's Simple Storage Service, assuming they are in BIDS format. Doing so requires you to set your AWS credentials and read the related s3 bucket documentation. You can find a guide here.
Derivatives have been produced on a variety of datasets, all of which are made available on our website. Each of these datsets is available for access and download from their respective sources. Alternatively, example datasets on the BIDS website which contain diffusion data can be used and have been tested;
ds114, for example.
For some downsampled test data, see neuroparc
This project is covered under the Apache 2.0 License.
If you're having trouble, notice a bug, or want to contribute (such as a fix to the bug you may have just found) feel free to open a git issue or pull request. Enjoy!