Skip to content

Tool to convert raw MRI datasets to BIDS-compatible datasets

License

Notifications You must be signed in to change notification settings

spinoza-centre/bidsify

 
 

Repository files navigation

bidsify - converts your (raw) data to the BIDS-format

https://travis-ci.org/spinoza-rec/bidsify.svg?branch=master https://ci.appveyor.com/api/projects/status/d9a7bjjqg204kofm?svg=true https://coveralls.io/repos/github/spinoza-rec/bidsify/badge.svg?branch=master

## This project has been archived as I finished my PhD! It won't be maintained anymore

This package offers a tool to convert your raw (f)MRI data to the "Brain Imaging Data Structuce" (BIDS) format. Using only a simple (json or yaml) config-file, it renames, reformats, and restructures your files such that it fits the BIDS naming scheme and conforms to file-formats specified by BIDS. After using bidsify, you can run your data through BIDS-compatible analysis/preprocessing pipelines such as fmriprep and mriqc package.

Currently, we use bidsify at the Spinoza Centre for Neuroimaging (location REC) to convert data to BIDS after each scan-session. We automated this process, including automatic preprocessing and quality control, using another package, nitools (which essentially "glues together" bidsify, fmriprep, and mriqc).

This package was originally developed to handle MRI-data from Philips scanners, which are traditionally exported in the "PAR/REC" format. Currently, bidsify also supports Philips (enhanced) DICOM (DICOM/DICOMDIR format) and Siemens DICOM (.dcm extension), but the latter has not been fully tested yet!

bidsify is still very much in development, so there are probably still some bugs for data that differs from our standard format (at the Spinoza Centre in Amsterdam) and the API might change in the future. If you encounter any issues, please submit an issue or (better yet), submit a pull-request with your proposed solution!

Installing bidsify & dependencies

This package can be installed using pip:

$ pip install bidsify

To get the "bleeding edge" version, you can install the master branch from github:

$ pip install git+https://github.com/spinoza-rec/bidsify.git@master

In terms of dependencies: bidsify uses dcm2niix under the hood to convert PAR/REC and DICOM files to nifti. Make sure you're using release v1.0.20181125 or newer.

Apart from dcm2niix, bidsify depends on the following Python packages:

  • nibabel
  • scipy
  • numpy
  • joblib (for parallelization)
  • pandas

Moreover, if you want to use the defacing option (i.e., removing facial features from anatomical images), make sure you have FSL installed, as well as the pydeface Python package. Also, to enable validating the BIDS-conversion process,(i.e., running bidsify with the -v flag), make sure to install bids-validator.

Lastly, if you want to use the Docker interface (i.e., running bidsify with the -D flag), which obviates the need for installing dcm2niix/FSL/bids-validator, make sure to install Docker and make sure your user account has permission to run Docker (see below).

Using Docker

The current version (master branch) allows you to run bidsify from docker, so you don't have to install all the (large) dependencies (FSL, pydeface, dcm2niix, bids-validator, etc.). To do so, you need to do the following.

  1. Install Docker (if you haven't already) and make sure you have permission to run Docker;
  2. Pull the Docker image: docker pull lukassnoek/bidsify:0.x.x (fill in the latest version at the x.x);
  3. Run bidsify with the -D flag (e.g., bidsify -c /home/user/config.yml -d /home/user/data -D)

Now you can use bidsify even without having FSL, dcm2niix, and other dependencies installed! (You do need to install bidsify itself though.)

How does it work?

After installing, the bidsify command can be called as follows:

$ bidsify [-c config_file] [-d path_to_data_directory] [-o output_directory] [-v] [-D]

The -c flag defaults to config.yml in the current working directory.

The -d flag defaults to the current working directory.

The -o flag defaults to the parent-directory of the data-directory.

The -v flag calls bids-validator after BIDS-conversion (optional).

The -D flag runs bidsify from Docker (recommended; see "Docker" section above).

For example, if you would call the following command ...

$ bidsify -c /home/user/data/config.yml -d /home/user/data

... your bidsified data will be in the following location:

/home/user
        ├── data
        |   ├── config.yml
        |   ├── s01
        |   └── s02
        |
        └── bids
            ├── dataset_description.json
            ├── participants.tsv
            ├── sub-01
            └── sub-02

Features

This package aims to take in any MRI-dataset and convert it to BIDS using information from the config-file provided by the user. Obviously, bidsify cannot handle all types of scans/data, but it can process most of the default scans/files we use at our MRI centre (Spinoza Centre), including

  • Standard (gradient-echo) EPI scans, both multiband and sequential
  • Standard (spin-echo) DWI scans
  • "Pepolar" (gradient-echo) EPI scans (also called "topup")
  • B0-based fieldmap scans (1 phase-difference + 1 magnitude image)
  • T1-weighted and T2-weighted scans

bidsify can handle both PAR/REC and DICOM files. Moreover, in the future we want to enable processing of:

  • Philips physiology-files ("SCANPHYSLOG" files; WIP, not functional yet)

In terms of "structure", this package allows the following "types" of datasets:

  • Multi-subject, multi-session datasets

The config file

bidsify only needs a config-file in either the json or YAML format. This file should contain information that can be used to rename and convert the raw files.

The config file contains a couple of sections, which are explained below (we'll use the YAML format).

"options"

The first (top-level) section (or "attribute" in JSON/YAML-lingo) in the file is the "options" section. An example of this section could be:

options:
  mri_ext: PAR  # alternatives: DICOM, dcm, nifti
  debug: False
  n_cores: -1
  subject_stem: sub
  deface: True
  spinoza_data: True
  out_dir: bids

No options need to be set explicitly as they all have sensible defaults. The attribute-value pairs mean the following:

  • mri_type: filetype of MRI-scans (PAR, dcm, DICOM, nifti; default: PAR)
  • n_cores: how many CPUs to use during conversion (default: -1, all CPUs)
  • debug: whether to print extra output for debugging (default: False)
  • subject_stem: prefix for subject-directories, e.g. "subject" in "subject-001" (default: sub)
  • deface: whether to deface the data (default: True, takes substantially longer though)
  • spinoza_data: whether data is from the Spinoza centre (default: False)
  • out_dir: name of directory to save results to (default: bids), relative to project-root.

Note that with respect to DICOM files, the mri_type can be set to DICOM (referring to Philips [enhanced] DICOM files) or dcm (referring to Siemens DICOM files with the extension .dcm).

"mappings"

The BIDS-format specifies the naming and format of several types of MRI(-related) filetypes. These filetypes have specific suffixes, which are appended to the filenames in the renaming process handled by bidsify. The "mappings" section in the config is meant to tell bidsify what filetype can be identified by which "key". Thus, the mappings section consists of "filetype": "identifier" pairs. Basically, if BIDS requires a specific suffix for a filetype, you need to specify that here. For example, a standard dataset with several BOLD-fMRI files, a T1, and physiological recordings could have a mappings section like this:

options:
  # ............. #

mappings:
  bold: _func
  T1w: 3DT1
  dwi: DWI
  physio: ppuresp
  events: log
  phasediff: _ph
  magnitude: _mag
  epi: topup
  T2w: T2w

Note that every file should belong to one, and only one, file-type! In other words, bidsify should be able to figure out what kind of file it's dealing with from the filename. For example, if you have a file named my_mri_file.PAR and you have configured the mappings as in the example above, bidsify won't be able to figure out what file-type it's dealing with (a bold file? A T1w file?), because the filename does not contain any of the mappings (e.g., _func, 3DT1, or DWI).

Moreover, the filename should not contain more than one file-type identifier! Suppose you have a file named workingmemory_func_ppuresp.nii.gz; with the above mappings, bidsify would conclude that it's either a bold file (because the name contains _func) OR a physio file (because the name contains ppuresp). As such, bidsify is going to skip converting/renaming this file and move it to the unallocated directory. In summary: files should contain one, and only one, identifier (such as _func) mapping to a particular file-type (e.g., bold).

Also, check the BIDS-specification for all filetypes supported by the format.

"metadata"

At the same (hierarchical) level as the "mappings" and "options" sections, a section with the name "metadata" can be optionally specified. This attribute may contain an arbitrary amount of attribute-value pairs which will be appended to each JSON-metadata file during the conversion. These are thus "dataset-general" metadata parameters. For example, you could specify the data of conversion here, if you'd like:

options:
  # some options

mappings:
  # some mappings

metadata:
  MagneticFieldStrength: 3
  ParallelAcquisitionTechnique: SENSE
  InstitutionName: Spinoza Centre for Neuroimaging, location REC

The func, anat, dwi, and fmap sections

After the options, mappings, and (optionally) the metadata sections, the specifications for the four general "BIDS-datatypes" - func, anat, dwi, and fmap - are listed in separate sections.

Each section, like func, can contain multiple sub-sections referring to different scans for that datatype. For example, you could have two different functional runs with each a different task ("workingmemory" and "nback"). In that case, the "func" section could look like:

options:
  # some options

mappings:
  # some mappings

func:

  wm-task:
    id: wmtask
    task: workingmemory

  nback-task:
    id: nbacktask
    task: nback

The exact naming of the "attributes" (here: wm-task and nback-task) of the sub-sections do not matter, but the subsequent key-value pairs do matter. You always need to set the id key, which is used to identify the files that belong to this particular task. Any key-value pair besides the id key-value pair are append to the renamed filename along the BIDS-format.

For example, suppose you have a raw file sub-001_wmtask.PAR. With the above config-file, this file will be renamed into sub-001_task-workingmemory_bold.nii.gz.

As discussed, any key-value pair besides id will be appended (in the format "key-value") to the filename during the renaming-process. Imagine, for example, that you have only one task - "nback" - but you acquired four runs of it per subject, of which the first two were acquired with a sequential acquisition protocol, but the last two with a multiband protocol (e.g. if you'd want to do some methodological comparison).

The config-file should, in that case, look like:

options:
  # some options

mappings:
  # some mappings

func:

  nback-task1:
    id: nback1
    task: nback
    run: 1
    acq: sequential

  nback-task2:
    id: nback1
    task: nback
    run: 2
    acq: sequential

  nback-task3:
    id: nback3
    task: nback
    run: 3
    acq: multiband

  nback-task4:
    id: nback4
    task: nback
    run: 4
    acq: multiband

bidsify will then create four files (assuming that they can be "found" using their corresponding ``id``s):

  • sub-001_task-nback_run-1_acq-sequential_bold.nii.gz
  • sub-001_task-nback_run-2_acq-sequential_bold.nii.gz
  • sub-001_task-nback_run-3_acq-multiband_bold.nii.gz
  • sub-001_task-nback_run-4_acq-multiband_bold.nii.gz

The same logic can be applied to the "dwi", "anat", and "fmap" sections. For example, if you would have two T1-weighted structural scans, the "anat" section could look like:

options:
  # some options

mappings:
  # some mappings

anat:

  firstT1:
    id: 3DT1_1
    run: 1

    secondT1:
      id: 3DT1_2
      run: 2

Importantly, any UNIX-style wildcard (e.g. *, ?, and [a,A,1-9]) can be used in the id values in these sections!

Lastly, apart from the different elements (such as nback-task1 in the previous example), each datatype-section (func, anat, fmap, and dwi) also may include a metadata section, similar to the "toplevel" metadata section. This field may include key-value pairs that will be appended to each JSON-file within that datatype. This is especially nice if you'd want to add metadata that is needed for specific preprocessing/analysis pipelines that are based on the BIDS-format. For example, the fmriprep package provides preprocessing pipelines for BIDS-datasets, but sometimes need specific metadata. For example, for each BOLD-fMRI file, it needs a field EffectiveEchoSpacing in the corresponding JSON-file, and for B0-files (one phasediff, one magnitude image) it needs the fields EchoTime1 and EchoTime2. To include those metadata fields in the corresponding JSON-files, just include a metadata field under the appropriate datatype section. For example, to do so for the previous examples:

func:

  metadata:
    EffectiveEchoSpacing: 0.00365
    PhaseEncodingDirection: "j"

  nback:
    id: nback
    task: nback

fmap:

  metadata:
    EchoTime1: 0.003
    EchoTime2: 0.008

  B0:
    id: B0

How to use bidsify

After installing this package, the bidsify command should be available. This command assumes a specific organization of your directory with raw data. Below, I outlined the assumed structure for a simple dataset with one BOLD run and one T1-weighted scan across two sessions:

/home/user/data/
            ├── config.yml
            ├── sub-01
            │   ├── ses-1
            │   │   ├── boldrun1.PAR
            │   │   ├── boldrun1.REC
            │   │   ├── T1.PAR
            │   │   └── T1.REC
            │   └── ses-2
            │       ├── boldrun1.PAR
            │       ├── boldrun1.REC
            │       ├── T1.PAR
            │       └── T1.REC
            └── sub-02
                ├── ses-1
                │   ├── boldrun1.PAR
                │   ├── boldrun1.REC
                │   ├── T1.PAR
                │   └── T1.REC
                └── ses-2
                    ├── boldrun1.PAR
                    ├── boldrun1.REC
                    ├── T1.PAR
                    └── T1.REC

(If you have DICOM-files with the .dcm extension, just replace the PAR/REC files with a single dcm file.)

So all raw files should be in a single directory, which can be the subject-directory or, optionally, a session-directory. Note: the session directory must be named "ses-<something>".

About

Tool to convert raw MRI datasets to BIDS-compatible datasets

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 83.3%
  • Dockerfile 13.5%
  • Shell 3.2%