Skip to content

versaFlow is an automated multi-resolution diffusion MRI processing pipeline implemented in Nextflow.

License

Notifications You must be signed in to change notification settings

AlexVCaron/versaFlow

Repository files navigation

Release build Docker containers DOI

versaFlow

versaflow is a diffusion MRI processing pipeline configured to handle high angular and spatial resolution data. By default it is configured to process Maccaca Mulatta brain images; profiles to handle other types of primates, as well as the human brain are being developped. The pipeline's pre-processing workflow contains a collection of steps focussed at increasing SNR and CNR of noisy diffusion data and improving the alignment of longitudinal acquisitions, using state-of-the-art algorithms such as MP-PCA denoising, Eddy and Topup, and libraries like Scipy, Dipy, Mrtrix, FSL and ANTs.

If you use this tool for your research, please cite the following

Valcourt Caron A., Shmuel A., Hao Z., Descoteaux M.,
“versaFlow: a versatile pipeline for resolution adapted diffusion MRI processing 
and its application to studying the variability of the PRIME-DE database”,
Frontiers in Neuroinformatics, 10.3389/fninf.2023.1191200, doi.org/10.3389/fninf.2023.1191200.

Requirements

To run the pipeline, the following tools must be installed :

Dependencies

All dependencies required to run the pipeline have been packaged in apptainer and docker images. The latest versions come pre-packaged with the CUDA runtime and require a Nvidia GPU to execute. For usage on a machine without a Nvidia GPU, use the images tagged nogpu.

  • Docker : Docker Hub
    • docker pull avcaron/versa:latest
  • Apptainer : apptainer images must be built locally, using the following command :
    • apptainer build <image.sif> docker://avcaron/versa:latest .

Data input format

** All images must be in the compressed Nifti format (.nii.gz) **

A digestible subject must be composed of a minimum of the following images :

  • A DWI image and its b-values (.bval) and b-vectors (.bvec) files, in the FSL format
  • A T1 anatomical image
  • A T1 or DWI mask (or both)

In addition to those, a subject can contain :

  • A reverse phase acquired b0 or DWI image

A single subject is identified by a unique key, prefixing the name of all images associated to it. This key must be composed of the directory names leading to the files, separated by a underscore symbol "_". Thus, all images must abide to the following naming convention :

  • <subject folder>
    • ...
      • <key>_dwi.nii.gz
      • <key>_dwi.bval
      • <key>_dwi.bvec
      • <key>_dwi.json (see here)
      • <key>_t1.nii.gz
      • <key>_t1_mask.nii.gz (optional, if DWI mask is supplied)
      • <key>_dwi_mask.nii.gz (optional, if T1 mask is supplied)
      • <key>_rev.nii.gz (optional)
      • <key>_rev.bval (optional, unrequired if reverse acquisition is only a b0)
      • <key>_rev.bvec (optional, unrequired if reverse acquisition is only a b0)
      • <key>_rev.json (see here)
      • <key>_wm_pvf.nii.gz (optional, will also be used to generate tissues mask)
      • <key>_gm_pvf.nii.gz (optional, will also be used to generate tissues mask)
      • <key>_csf_pvf.nii.gz (optional, will also be used to generate tissues mask)

Examples of name formatting :

  • For a subject in a directory named sub1, all files associated to it must have the key prefix sub1. A diffusion image should thus be named sub1_dwi.nii.gz.
  • For a session in a subdirectory ses1 of the main directory of a subject sub1, the key would be sub1_ses1. A diffusion image would then be named sub1_ses1_dwi.nii.gz.

Content of diffusion acquisition specification file (.json) :

The json file included with a dwi or rev image describes the parameters related to their specific acquisition. They are listed as a json object, in the fashion displayed below :

{
  "direction" : <phase encoding direction>,
  "slice_direction" : <slicing direction>,
  "readout" : <readout>,
  "interleaved" : <false|true>,
  "multiband_factor" : <multiband factor>
}

Optionally, one can replace the "interleaved" and "multiband_factor" fields by the field "slice_indexes", a list of list of indexes defining the sequence of acquisition of the image's slices. Two examples :

  • An image with 10 slices, acquired interleaved, multiband factor of 1 :
    [[0], [2], [4], [6], [8], [1], [3], [5], [7], [9]]
    
  • An image with interleaved slices, multiband factor of 2, 10 slices :
    [[0, 5], [2, 7], [4, 9], [1, 6], [3, 8]]
    

It is also possible instead of providing .json files to specify those parameters using default values passed through command line when calling the pipeline. For more information, refer to this section below.

The best way to extract the information to fill the .json configuration is to use dcm2niix with the output option --json when unpacking the base DICOM images. It will produce a json file of metadata.

As of now, this json is not directly consumable by the pipeline and must be parsed and transformed by the user to the structure presented above. This will be automatically handled in a future version of the pipeline.

Running the pipeline

To run the pipeline, we recommend creating a directory per run that contains log and cache of execution from Nextflow. In that directory, then run the following command

nextflow run \
    -resume \
    -w cache \
    -dsl2 \
    <root of versaFlow>/main.nf \
    --data_root <data root>

Additional parameters to this command can be supplied, such as :

  • Running with docker image : -with-docker <docker image>
  • Running with apptainer image : -with-apptainer <apptainer image>
  • Changing output directory : --output_root <output folder>
  • Display help and usage : --help

Alternative specification of the diffusion acquisition parameters

By construction, the .json files provided with the dwi volumes are required. However, it is possible to define default values for all the fields in the .json specification presented above, either via command line or through the Nextflow.config file, in which case the .json files will be optional. The command line parameters are :

--default_readout <readout value>
--default_multiband_factor <multiband factor>
--default_slicing_direction <slicing direction>
--default_phase_direction <phase direction>
--default_is_interleaved

Nextflow.config file

If a nextflow.config file is present in the execution directory, it will be used as configuration for the run. A user can then modify the effective configuration for the run and turn on and off the execution of different processes.

Some configuration capabilities have been extracted to separate configuration files. We refer the user to the header of the nextflow.config file for more information. Those configuration files can be found in the .config directory.


About

versaFlow is an automated multi-resolution diffusion MRI processing pipeline implemented in Nextflow.

Resources

License

Stars

Watchers

Forks

Packages

No packages published