Data collection and organization

Imaging Protocol

The “Spine Generic” MRI acquisition protocol is available at this link. Each site was instructed to scan six healthy subjects (3 men, 3 women), aged between 20 and 40 y.o. Note: there was some flexibility in terms of number of participants and age range.

If your site is interested in contributing to the publicly-available database, please coordinate with Julien Cohen-Adad.

Multi-center data

In the context of this project, the following dataset have been acquired and are available as open-access:

Data conversion: DICOM to BIDS

To facilitate the collection, sharing and processing of data, we use the BIDS standard. An example of the data structure for one center is shown below:

data-multi-subject
│
├── dataset_description.json
├── participants.json
├── participants.tsv
├── sub-ubc01
├── sub-ubc02
├── sub-ubc03
├── sub-ubc04
├── sub-ubc05
├── sub-ubc06
│   │
│   ├── anat
│   │   ├── sub-ubc06_T1w.json
│   │   ├── sub-ubc06_T1w.nii.gz
│   │   ├── sub-ubc06_T2star.json
│   │   ├── sub-ubc06_T2star.nii.gz
│   │   ├── sub-ubc06_T2w.json
│   │   ├── sub-ubc06_T2w.nii.gz
│   │   ├── sub-ubc06_acq-MToff_MTS.json
│   │   ├── sub-ubc06_acq-MToff_MTS.nii.gz
│   │   ├── sub-ubc06_acq-MTon_MTS.json
│   │   ├── sub-ubc06_acq-MTon_MTS.nii.gz
│   │   ├── sub-ubc06_acq-T1w_MTS.json
│   │   └── sub-ubc06_acq-T1w_MTS.nii.gz
│   │
│   └── dwi
│       ├── sub-ubc06_dwi.bval
│       ├── sub-ubc06_dwi.bvec
│       ├── sub-ubc06_dwi.json
│       ├── sub-ubc06_dwi.nii.gz
│       ├── (sub-ubc06_acq-b0_dwi.json)
│       └── (sub-ubc06_acq-b0_dwi.nii.gz)
│
└── derivatives
    │
    └── labels
        └── sub-ubc06
            │
            ├── anat
            │   ├── sub-ubc06_T1w_RPI_r_seg-manual.nii.gz  <---------- manually-corrected spinal cord segmentation
            │   ├── sub-ubc06_T1w_RPI_r_seg-manual.json  <------------ information about origin of segmentation (see below)
            │   ├── sub-ubc06_T1w_RPI_r_labels-manual.nii.gz  <------- manual vertebral labels
            │   ├── sub-ubc06_T1w_RPI_r_labels-manual.json
            │   ├── sub-ubc06_T2w_RPI_r_seg-manual.nii.gz  <---------- manually-corrected spinal cord segmentation
            │   ├── sub-ubc06_T2w_RPI_r_seg-manual.json
            │   ├── sub-ubc06_acq-T1w_MTS_seg-manual.nii.gz  <-------- manually-corrected spinal cord segmentation
            │   ├── sub-ubc06_acq-T1w_MTS_seg-manual.json
            │   ├── sub-ubc06_T2star_rms_gmseg-manual.nii.gz  <------- manually-corrected gray matter segmentation
            │   └── sub-ubc06_T2star_rms_gmseg-manual.json
            │
            └── dwi
                ├── sub-ubc06_dwi_moco_dwi_mean_seg-manual.nii.gz  <-- manually-corrected spinal cord segmentation
                └── sub-ubc06_dwi_moco_dwi_mean_seg-manual.json

To convert your DICOM data folder to the compatible BIDS structure, you need to install dcm2bids. Once installed, download this config file (click File>Save to save the file), then convert your Dicom folder using the following command (replace xx with your center and subject number):

dcm2bids -d PATH_TO_DICOM -p sub-ID_SITE -c config_spine.txt -o SITE_spineGeneric

For example:

dcm2bids -d /Users/julien/Desktop/DICOM_subj3 -p sub-milan03 -c ~/Desktop/config_spine.txt -o milan_spineGeneric

A log file is generated under tmp_dcm2bids/log/. If you encounter any problem while running the script, please open an issue and upload the log file. We will offer support.

Once you have converted all subjects for the study, create the following files and add them to the data structure:

dataset_description.json (Pick the correct values depending on your system and environment)

{
    "Name": "Spinal Cord MRI Public Database",
    "BIDSVersion": "1.2.0",
    "InstitutionName": "Name of the institution",
    "Manufacturer": "YOUR_VENDOR",
    "ManufacturersModelName": "YOUR_MODEL",
    "ReceiveCoilName": "YOUR_COIL",
    "SoftwareVersion": "YOUR_SOFTWARE",
    "Researcher": "J. Doe, S. Wonder, J. Pass",
    "Notes": "Particular notes you might have. E.g.: We don't have the ZOOMit license, unf-prisma/sub-01 and unf-skyra/sub-03 is the same subject.
}

Example of possible values:

  • Manufacturer: “Siemens”, “GE”, “Philips”
  • ManufacturersModelName: “Prisma”, “Prisma-fit”, “Skyra”, “750w”, “Achieva”
  • ReceiveCoilName: “64ch+spine”, “12ch+4ch neck”, “neurovascular”
  • SoftwareVersion: “VE11C”, “DV26.0”, “R5.3”, …

participants.json (This file is generic, you don’t need to change anything there. Just create a new file with this content)

{
    "participant_id": {
        "LongName": "Participant ID",
        "Description": "Unique ID"
    },
    "sex": {
        "LongName": "Participant gender",
        "Description": "M or F"
    },
    "age": {
        "LongName": "Participant age",
        "Description": "yy"
    },
    "date_of_scan": {
        "LongName": "Date of scan",
        "Description": "yyyy-mm-dd"
    }
}

participants.tsv (Tab-separated values)

participant_id  sex age date_of_scan    institution_id  institution manufacturer    manufacturers_model_name    receive_coil_name   software_versions   researcher
sub-unf01   F   24  2018-12-07  unf Neuroimaging Functional Unit (UNF), CRIUGM, Polytechnique Montreal  Siemens Prisma-fit  HeadNeck_64 syngo_MR_E11    J. Cohen-Adad, A. Foias
sub-unf02   M   29  2018-12-07  unf Neuroimaging Functional Unit (UNF), CRIUGM, Polytechnique Montreal  Siemens Prisma-fit  HeadNeck_64 syngo_MR_E11    J. Cohen-Adad, A. Foias
sub-unf03   M   22  2018-12-07  unf Neuroimaging Functional Unit (UNF), CRIUGM, Polytechnique Montreal  Siemens Prisma-fit  HeadNeck_64 syngo_MR_E11    J. Cohen-Adad, A. Foias
sub-unf04   M   31  2018-12-07  unf Neuroimaging Functional Unit (UNF), CRIUGM, Polytechnique Montreal  Siemens Prisma-fit  HeadNeck_64 syngo_MR_E11    J. Cohen-Adad, A. Foias
sub-unf05   F   23  2019-01-11  unf Neuroimaging Functional Unit (UNF), CRIUGM, Polytechnique Montreal  Siemens Prisma-fit  HeadNeck_64 syngo_MR_E11    J. Cohen-Adad, A. Foias
sub-unf06   F   27  2019-01-11  unf Neuroimaging Functional Unit (UNF), CRIUGM, Polytechnique Montreal  Siemens Prisma-fit  HeadNeck_64 syngo_MR_E11    J. Cohen-Adad, A. Foias

Once you’ve created the BIDS dataset, remove any temp folders (e.g., tmp_dcm2bids/) and zip the entire folder. It is now ready for sharing! You could send it to Julien Cohen-Adad via any cloud-based method (Gdrive, Dropbox, etc.).

Checking acquisition parameters

To ensure the acquisition protocol was properly followed by each site, we implemented a parameter validator that verifies if the pulse sequence parameters match the required ones from the generic protocol (within a tolerance range). Basic parameters are checked, including: repetition time, echo time, flip angle. These parameters are read from the json sidecar file (generated by the DICOM to BIDS conversion). Note that BIDS file naming convention is also checked by the validator. If a parameter does not match, a warning message is triggered.

This validator is exposed in this command line interface (CLI) function: sg_params_checker. This function is run during continuous integration (CI), for each dataset, ensuring valid dataset throughout the life cycle of the project.

The json file containing the recommended acquisition parameters is located under /spinegeneric/cli/specs.json.

Example usage and expected output:

sg_params_checker -path-in ~/data-single-subject/
WARNING: Incorrect FlipAngle: sub-douglas_T2w.nii.gz; FA=120 instead of 180
WARNING: Incorrect RepetitionTime: sub-mgh_T2w.nii.gz; TR=2 instead of 1.5
WARNING: Incorrect FlipAngle: sub-tokyoSigna1_T2star.nii.gz; FA=20 instead of 30
WARNING: Incorrect FlipAngle: sub-tokyoSigna2_T2star.nii.gz; FA=20 instead of 30
WARNING:sub-ucl_T2star.nii.gz Missing Manufacturer in json sidecar; Cannot check parameters.

Ethics and anonymization

Each subject consented to be scanned and to have their anonymized data put in a publicly-available repository. To prove it, an email from each participant should be sent to the manager of the database (Julien Cohen-Adad). The email should state the following: “I am the subject who volunteered and I give you permission to release the scan freely to the public domain.”

Anatomical scans where facial features are visible (T1w) could be “defaced” before being collected (at the discretion of the subject).

This can be done automatically using R or manually, in case the automatic defacing fails.

Automatic defacing with R

  1. Install R, then open R (type “r” in the Terminal) and install the following dependencies:
install.packages("sessioninfo")
install.packages("remotes")
remotes::install_github("muschellij2/oro.nifti")  # answer "Yes" to "install from source?"
install.packages("fslr")
install.packages("argparser")
install.packages("devtools")
remotes::install_github("muschellij2/extrantsr")  # choose "1" when prompted
  1. Download this repository and install Python’s dependencies as instructed in Getting started.
  2. Run:
sg_deface_using_r -i PATH_TO_BIDS_DATASET -o PATH_TO_DEFACED_BIDS_DATASET -f
sg_deface_using_r -i PATH_TO_BIDS_DATASET -o PATH_TO_DEFACED_BIDS_DATASET
  1. To launch the QC report of the defacing across multiple subjects, run:
sg_qc_bids_deface

Manual Defacing

Automatic defacing might fail in some subjects, so this section explains how to deface manually. This procedure takes less than a minute per subject. Here we use FSLeyes but you can use any other NIfTI image editor.

Open FSLeyes and load the T1w scan. Go to Tools > Edit mode, Select the pencil with size 100, deface, then save.

Below is an example of a defaced subject:

example\_defacing

Example of manual defacing.

Example of datasets

T1w - sub-vuiisAchieva02

T2w - sub-milan01

T2star - sub-brnoCeitec01

MTon_MTS - sub-barcelona04

MToff_MTS - sub-barcelona04

T1w_MTS - sub-barcelona04

DWI - sub-barcelona04