This a BIDS App wrapper for HCP Pipelines v4.3.0. Like every BIDS App it consists of a container that includes all of the dependencies and run script that parses a BIDS dataset. BIDS Apps run on Windows, Linux, Mac as well as HPCs/clusters.
To convert DICOMs from your HCP-Style (CMRR) acquisitions to BIDS try using heudiconv with this heuristic file.
The HCP Pipelines product is a set of tools (primarily, but not exclusively, shell scripts) for processing MRI images for the Human Connectome Project. Among other things, these tools implement the Minimal Preprocessing Pipeline (MPP) described in Glasser et al. 2013.
This BIDS App requires that each subject has at least one T1w and one T2w scan. Lack fMRI or dMRI scans is handled robustly. Note that while anatomicals (T1w, T2w scans) can be processed without a fieldmap, a fieldmap is mandatory for processing fMRI scans. Support for the HCP-Pipelines 'legacy' processing mode will be added in an upcoming release.
Release Notes, Installation, and Usage
Discussion of HCP Pipeline usage and improvements can be posted to the hcp-users discussion list. Sign up for hcp-users at http://humanconnectome.org/contact/#subscribe.
Please open an issue if you encounter errors building this BIDS App or believe you have encountered an error specific to the BIDS App wrapper.
Please cite Glasser et al. 2013 and Smith et al. 2013.
This App has the following command line arguments:
usage: run.py [-h]
[--participant_label PARTICIPANT_LABEL [PARTICIPANT_LABEL ...]]
[--session_label SESSION_LABEL [SESSION_LABEL ...]]
[--n_cpus N_CPUS]
[--stages {PreFreeSurfer,FreeSurfer,PostFreeSurfer,fMRIVolume,fMRISurface} [{PreFreeSurfer,FreeSurfer,PostFreeSurfer,fMRIVolume,fMRISurface} ...]]
[--coreg {MSMSulc,FS}] [--gdcoeffs GDCOEFFS] --license_key
LICENSE_KEY [-v] [--anat_unwarpdir {x,y,z,-x,-y,-z}]
[--skip_bids_validation]
bids_dir output_dir {participant}
HCP Pipelines BIDS App (T1w, T2w, fMRI)
positional arguments:
bids_dir The directory with the input dataset formatted
according to the BIDS standard.
output_dir The directory where the output files should be stored.
If you are running group level analysis this folder
should be prepopulated with the results of
theparticipant level analysis.
{participant} Level of the analysis that will be performed. Multiple
participant level analyses can be run independently
(in parallel) using the same output_dir.
optional arguments:
-h, --help show this help message and exit
--participant_label PARTICIPANT_LABEL [PARTICIPANT_LABEL ...]
The label of the participant that should be analyzed.
The label corresponds to sub-<participant_label> from
the BIDS spec (so it does not include "sub-"). If this
parameter is not provided all subjects should be
analyzed. Multiple participants can be specified with
a space separated list.
--session_label SESSION_LABEL [SESSION_LABEL ...]
The label of the session that should be analyzed. The
label corresponds to ses-<session_label> from the BIDS
spec (so it does not include "ses-"). If this
parameter is not provided, all sessions should be
analyzed. Multiple sessions can be specified with a
space separated list.
--n_cpus N_CPUS Number of CPUs/cores available to use.
--stages {PreFreeSurfer,FreeSurfer,PostFreeSurfer,fMRIVolume,fMRISurface} [{PreFreeSurfer,FreeSurfer,PostFreeSurfer,fMRIVolume,fMRISurface} ...]
Which stages to run. Space separated list.
--coreg {MSMSulc,FS} Coregistration method to use
--gdcoeffs GDCOEFFS Path to gradients coefficients file
--license_key LICENSE_KEY
FreeSurfer license key - letters and numbers after "*"
in the email you received after registration. To
register (for free) visit
https://surfer.nmr.mgh.harvard.edu/registration.html
-v, --version show program's version number and exit
--anat_unwarpdir {x,y,z,x-,y-,z-}
Unwarp direction for 3D volumes
--skip_bids_validation, --skip-bids-validation
assume the input dataset is BIDS compliant and skip
the validation
--processing_mode {hcp,legacy,auto}, --processing-mode {hcp,legacy,auto}
Control HCP-Pipeline modehcp (HCPStyleData): require
T2w and fieldmap modalitieslegacy (LegacyStyleData):
always ignore T2w and fieldmapsauto: use T2w and/or
fieldmaps if available
--doslicetime Apply slice timing correction as part of fMRIVolume.
To run it in participant level mode (for one participant):
docker run -i --rm \
-v /Users/filo/data/ds005:/bids_dataset:ro \
-v /Users/filo/outputs:/outputs \
bids/hcppipelines \
/bids_dataset /outputs participant --participant_label 01 --license_key "XXXXXX"
This BIDS App incorporates several non-free packages required for the HCP Pipeline, including:
If you are considering commercial use of this App please consult the relevant licenses.
- Add DiffusionProcessing stage
- More testing for fMRI with different resolution
- Run fMRI runs in parallel (when n_cpus present)
- Add support for TOPUP and GE fieldmaps for structural scans (please get in touch if you can provide sample data)
- Add support for GE fieldmaps for fMRI scans (please get in touch if you can provide sample data)
- Avoid copying fsaverage folder for every participant
- Add ICA FIX stage
- Add group level analysis
- Add task fMRI model fitting