Skip to content

Latest commit

 

History

History
155 lines (109 loc) · 5.63 KB

readme.md

File metadata and controls

155 lines (109 loc) · 5.63 KB
Raw

fsm2bids

Notes on how to conver the FreeSufer maintenance grant dataset (fsm) to BIDS format

Overview

This repo describes how the FSM dataset was converted to BIDS and contains the heudiconv heuristic file.

Setup

  1. Create a new pyton environment
conda create --name heudiconv python=3.7
conda activate heudiconv
pip install heudiconv
  1. Install dmc2niix, and ensure it's in your PATH

See dmc2niix's installation instructions

  1. Create a DICOM directory

Inside the DICOM directory, create a sub-directory for each subject. The FSM dataset is single session, so we don't have to worry about sub-dividing for sessions.

We'll use the env var DICOM_DIR to refer to this directory.

To obfusacte subject names, we create symlinks in the DICOM_DIR with the obfuscated subject name, that points to the subject directory with the actual DICOMs.

  1. Create an output directory for heudiconv

This directory represents a 'first pass' at creating the BIDS dataset. We'll need to operate over this directory to create the final BIDS dataset and do things like:

  • Fill out the stub json files that heudiconv creates (i.e. dataset_description.json, etc)
  • Deface the data
  • Scrub sensitive information from the json sidecar files

We'll use the env var HEUDICONV_OUT to refer to this directory

  1. Create an env var to point to the heudiconv heuristic file

This is the file fsm2bids-heuristic.py in this repository.

We'll use the env var HEUDICONV_HEURISTIC to refer to this file

  1. Create an output directory for the final BIDS dataset to be published. This will contain a defaced version of the BIDS dataset.

We'll use the env var BIDS_FINAL to refer to this directory

  1. Create a working directory for mideface

Example

This is what I use to get everything setup n my machine:

conda activate heudiconv
export DICOM_DIR=/home/paul/lcn/20220822-fsm-bids/dicom
export HEUDICONV_OUT=/home/paul/lcn/20220822-fsm-bids/heudiconv-out
export BIDS_FINAL=/home/paul/lcn/20220822-fsm-bids/bids-final
export MIDEFACE_WORK=/home/paul/lcn/20220822-fsm-bids/mideface-work
export HEUDICONV_HEURISTIC=/home/paul/lcn/git/fsm2bids/fsm2bids-heuristic.py

Note all of the above directories, with the exception of BIDS_FINAL contains PHI and should not be distributed.

Using heudiconv

Initial scan

To begin, we run heudiconv in 'convertall' mode on a single subject to analyse the dataset and help us build the heuristics file. Here, we are using the subject sub-fsm042

heudiconv \
  -d ${DICOM_DIR}/sub-{subject}/* \
  -s fsm042 \
  -o ${HEUDICONV_OUT} \
  -c none \
  -f convertall \
  --overwrite

This creates the directory .heudiconv under $HEUDICONV_OUT.

Create the $HEUDICONV_HEURISTIC file

The file ${HEUDICONV_OUT}/.heudiconv/fsm042/info/dicominfo.tsv is very helpful for creating the heuristic file.

Other useful resources:

Convert a single subject

Use the heuristic file created above to convert data. We need to delete the ${HEUDICONV_OUT}/.heudiconv directory, otherwise heudiconv will use the cached heurisitc file (in this case, ${HEUDICONV_OUT}/.heudiconv/fsm042/info/heuristic.py and nothing interesting will happen.

So let's go ahead and delete the entire ${HEUDICONV_OUT}/.heudiconv directory:

rm -rf ${HEUDICONV_OUT}/.heudiconv

Then:

heudiconv \
  -d ${DICOM_DIR}/sub-{subject}/* \
  -s fsm042 \
  -o ${HEUDICONV_OUT} \
  -c dcm2niix \
  -f ${HEUDICONV_HEURISTIC} \
  --bids \
  --overwrite

Convince yourself everything converted correctly

See the converted files in ${HEUDICONV_OUT}/sub-fsm042

Repeat for all other subjects

Edit convert-all.bash set the variables HEUDICONV_OUT, DICOM_DIR and HEUDICONV_HEURISTIC accordingly then run.

Create a defaced version of the BIDS dataset.

Edit daface.bash and set:

  • BIDS_DIR_FACE to the value of HEUDICONV_OUT above
  • WORK_DIR to the value of MIDEFACE_WORK above
  • BIDS_DIR_DEFACE to the value of BIDS_FINAL above
  • DERIVATIVES_DIR to ${BIDS_DIR_DEFACE}/derivatives/mideface

Then run the deface.bash script which will iterate through each subject and:

  • Runs mideface on:
    • ${SUB}_acq-gre3d_echo-1_part-mag_MEGRE.nii.gz
    • ${SUB}_acq-gre3d_T2starmap.nii.gz
    • ${SUB}_acq-mp2rage_inv-2_MP2RAGE.nii.gz
    • ${SUB}_acq-mprageVnav_T1wRMS.nii.gz
    • ${SUB}_acq-t2flair_T2w.nii.gz
    • ${SUB}_acq-t2spaceVnav_T2w.nii.gz
  • Applies the resulting mideface facemasks to all other mag contrasts from the same sequence
  • For sequences with phase images:
    • The resulting facemask is noted to produce a mul facemask (face voxels are 0, all other voxels are 1)
    • Multiples the phase image with this mul facemask to deface
  • Copies over .json files from the anat directory
  • Copies over non-defaced directories, including func, fmap and dwi
  • Copes over top-level subject files (${SUB}_scans.tsv)

Finalize the dataset

TODO

  • File/json sidecar cleanup
  • Validate

Todo

  • How to specify a specific session, if multiple sessions of the same sequences exists?
  • Validate
  • Investigate why heudiconv generates 3 ROI files for sub-fsm21em