This project is a part of the ReproNim Center suite of tools and frameworks. Its goal is to provide a turnkey flexible setup for automatic generation of shareable, version-controlled BIDS datasets from MR scanners. To not reinvent the wheel, all actual software development is largely done through contribution to existing software projects:
a flexible DICOM converter for organizing brain imaging data into structured
ReproIn heuristic was developed and now is shipped within HeuDiConv,
so it could be used independently of the ReproIn setup on any HeuDiConv
-f reprointo heudiconv call).
a modular version control platform and distribution for both code and
data. DataLad support was contributed to HeuDiConv, and could be
enabled by adding
--dataladoption to the
The header of the heuristic file describes details of the specification on how to organize and name study sequences at MR console.
Schematic description of the overall setup:
Note: for your own setup, dcm2niix author recommends to avoid dcm4che and choose another PACS.
Making your sequence compatible with ReproIn heuristic
- Walkthrough #1: guides you through ReproIn approach to organizing exam cards and managing canceled runs/sessions on Siemens scanner(s)
Renaming sequences to conform the specification needed by ReproIn
TODO: Describe how sequences could be renamed per study by creating a derived heuristic
Install HeuDiConv and DataLad: e.g.
apt-get update; apt-get install heudiconv dataladin any NeuroDebian environment. If you do not have one, you could get either of
- NeuroDebian Virtual Machine
- ReproIn Docker image:
docker run -it --rm -v $PWD:$PWD repronim/reproin
- ReproIn Singularity image:
singularity pull shub://ReproNim/reproin
Collect a subject/session (or multiple of them) while placing and naming sequences in the scanner following the specification. But for now we will assume that you have no such dataset yet, and want to try on phantom data:
datalad install -J3 -r -g ///dicoms/dartmouth-phantoms/bids_test4-20161014
to get all subdatasets recursively, while getting the data as well in parallel 3 streams. This dataset is a sample of multi-session acquisition with anatomicals and functional sequences on a friendly phantom impersonating two different subjects (note: fieldmaps were deficient, without magnitude images). You could also try other datasets such as ///dbic/QA
We are ready to convert all the data at once (heudiconv will sort into accessions) or one accession at a time. The recommended invocation for the heudiconv is
heudiconv -f reproin --bids --datalad -o OUTPUT --files INPUT
to convert all found in
INPUTDICOMs and place then within the hierarchy of DataLad datasets rooted at
OUTPUT. So we will start with a single accession of
heudiconv -f reproin --bids --datalad -o OUTPUT --files bids_test4-20161014/phantom-1
and inspect the result under OUTPUT, probably best with
... WiP ...
HeuDiConv options to overload autodetected variables:
Sample converted datasets
You could find sample datasets with original DICOMs
- ///dbic/QA is a publicly
available DataLad dataset with historical data on QA scans from DBIC.
You could use DICOM tarballs under
sourcedata/for your sample conversions. TODO: add information from which date it is with scout DICOMs having session identifier
- ///dicoms/dartmouth-phantoms provides a collection of datasets acquired at DBIC to establish ReproIn specification. Some earlier accessions might not be following the specification. bids_test4-20161014 provides a basic example of multi-subject and multi-session acquisition.
This repository provides a Singularity environment definition file used to generate a complete environment needed to run a conversion. But also, since all work is integrated within the tools, any environment providing them would suffice, such as NeuroDebian docker or Singularity images, virtual appliances, and other Debian-based systems with NeuroDebian repositories configured, which would provide all necessary for ReproIn setup components.
Complete setup at DBIC
It relies on the hardcoded ATM in
reproin locations and organization
of DICOMs and location of where to keep converted BIDS datasets.
# m h dom mon dow command 55 */12 * * * $HOME/reproin-env-0.9.0 -c '~/proj/reproin/bin/reproin lists-update-study-shows' && curl -fsS -m 10 --retry 5 -o /dev/null https://hc-ping.com/61dfdedd-SENSORED
curl at the end is to make use of https://healthchecks.io
to ensure that we do have CRON job ran as we expected.
ATM we reuse a singularity environment based on reproin 0.9.0 produced from this repo and shipped within ReproNim/containers. For the completeness sake
(reproin-3.8) [bids@rolando lists] > cat $HOME/reproin-env-0.9.0 #!/bin/sh env -i /usr/local/bin/singularity exec -B /inbox -B /afs -H $HOME/singularity_home $(dirname $0)/reproin_0.9.0.simg /bin/bash "$@"
which produces emails with content like
Wager/Wager/1102_MedMap: new=92 todo=5 done=102 /inbox/BIDS/Wager/Wager/1102_MedMap/.git/study-show.sh 2023-03-30 PI/Researcher/ID_name: new=32 no studydir yet Haxby/Jane/1073_MonkeyKingdom: new=4 todo=39 done=8 fixups=6 /inbox/BIDS/Haxby/Jane/1073_MonkeyKingdom/.git/study-show.sh 2023-03-30
where as you can see it updates on the status for each study which was scanned for from the
beginning of the current month. And it ends with the pointer to
study-show.sh script which
would provide details on already converted or heudiconv line invocations for what yet to do.
For the "no studydir yet" we need first to generate study dataset (and
possibly all leading
PI/Researcher super-datasets via
reproin study-create PI/Researcher/ID_name
Unless there are some warnings/conflicts (subject/session already converted, etc) are found,
reproin study-convert PI/Researcher/ID_name
could be used to convert all new subject/sessions for that study.
Anonymization or other scripts might obfuscate "Study Description" thus ruining "locator" assignment. See issue #57 for more information.
- add a pre-configured DICOM receiver for fully turnkey deployments
- heudiconv-monitor to fully automate conversion of the incoming data
- BIDS dataset manipulation helper