ReproIn

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:

• HeuDiConv: a flexible DICOM converter for organizing brain imaging data into structured directory layouts. ReproIn heuristic was developed and now is shipped within HeuDiConv, so it could be used independently of the ReproIn setup on any HeuDiConv installation (specify -f reproin to heudiconv call).
• DataLad: a modular version control platform and distribution for both code and data. DataLad support was contributed to HeuDiConv, and could be enabled by adding --datalad option to the heudiconv call.

Specification

The header of the heuristic file describes details of the specification on how to organize and name study sequences at MR console.

Overall workflow

Schematic description of the overall setup:

Note: for your own setup, dcm2niix author recommends to avoid dcm4che and choose another PACS.

Tutorial/HOWTO

Data collection

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

Conversion

1. Install HeuDiConv and DataLad: e.g. apt-get update; apt-get install heudiconv datalad in 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: you can either
     • convert from the docker image: singularity pull docker://repronim/reproin
     • download the most recent version from http://datasets.datalad.org/?dir=/repronim/containers/images/repronim which is a DataLad dataset which you can install via datalad install ///repronim/containers (see/subscribe #64 for HOWTO setup YODA style dataset)

2. 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

3. 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 INPUT DICOMs and place then within the hierarchy of DataLad datasets rooted at OUTPUT. So we will start with a single accession of phantom-1/

    heudiconv -f reproin --bids --datalad -o OUTPUT --files bids_test4-20161014/phantom-1
   

   and inspect the result under OUTPUT, probably best with datalad ls command:

    ... WiP ...
   

HeuDiConv options to overload autodetected variables:

• --subject
• --session
• --locator

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.

Containers/Images etc

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.

Getting started from scratch

Setup environment

reproin script relies on having datalad, datalad-containers, and singularity available. The simplest way to get them all is to install a conda distribution, e.g. miniforge (link for amd64), and setup the environment with all components installed:

mamba create -n reproin -y datalad datalad-container singularity

Note that in future sessions you will need to activate this environment:

mamba activate reproin

Then make sure you have your git configured. If git config --list does not include these entries, add (adjust to fit your persona)

git config --global user.name  "My Name"
git config --global user.email  "MyName@example.com"

and install the ReproNim/containers

datalad clone https://github.com/ReproNim/containers repronim-containers
cd repronim-containers

which would clone the dataset from GitHub and auto-enable datasets.datalad.org remote to actually get annexed content of the images. Now fetch the image for the most recent version of reproin from under images/repronim, e.g.

datalad get images/repronim/repronim-reproin--0.13.1.sing
cd ..

"Install" reproin script

The singularity image we fetched already comes with reproin installed inside, but to "drive" conversion we need to have reproin available in the base environment. Because we do not have it (yet) packaged for conda distribution, we will just clone this repository and gain access to the script:

git clone https://github.com/ReproNim/reproin

To avoid typing the full path to the reproin script, can do

export "PATH=$PWD/reproin/bin/:$PATH"

to place it in the PATH.

NB. It is important ATM to not just cp that reproin script elsewhere because it relies on being able to find other resources made available in that repository (e.g., cfg_reproin_bids.py).

"Configure" the reproin setup

Currently reproin script hardcodes the path to DICOMS to reside under /inbox/DICOM and extracted lists and converted data to reside under /inbox/BIDS. It is possible to overload location for BIDS via BIDS_DIR env variable, so we can do e.g.

export BIDS_DIR=$HOME/BIDS-demo

and then let's create the top-level datalad dataset to contain all converted data, configuring to store text files in git rather than git-annex,

datalad create -c text2git "$BIDS_DIR"

Collect DICOMs listing

ATM reproin container has an older version of the script, so to use newer version we would just bind mount our cloned script inside,

singularity run -e -c \
   --env BIDS_DIR=$BIDS_DIR \
   -B $HOME/reproin/bin/reproin:/usr/local/bin/reproin \
   -B /inbox/DICOM:/inbox/DICOM:ro \
   -B $BIDS_DIR:$BIDS_DIR \
   ~/repronim-containers/images/repronim/repronim-reproin--0.13.1.sing lists-update-study-shows

which should output summary over the studies it found under /inbox/DICOM, e.g.

dbic/QA: new=16 no studydir yet
PI/Researcher/1110_SuperCool: new=12 no studydir yet

Create target dataset

Now we can create "studydir" for the study of interest, e.g.

reproin study-create dbic/QA

which would

• create target BIDS dataset within the hierarchy

• install repronim/containers borrowing the image from the ~/repronim-containers

• rerun study-show to output summary over the current state like

  todo=4 done=0 /afs/.dbic.dartmouth.edu/usr/haxby/yoh/BIDS-demo/dbic/QA/.git/study-show.sh 2024-11-11

Convert the dataset

Go to the folder of the dataset, e.g.

cd "$BIDS_DIR/dbic/QA"

to see that reproin pre-setup everything needed to run conversion (cat .datalad/config). And now you should be able to run conversion for your study via "datalad-container" extension:

datalad containers-run -n repronim-reproin study-convert dbic/QA

Gotchas

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.

• /inbox/DICOM/{YEAR}/{MONTH}/{DAY}/A00{ACCESSION}
• /inbox/BIDS/{PI}/{RESEARCHER}/{ID}_{name}/

CRON job

# 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

NB: that 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.

reproin study-create

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

reproin study-convert

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.

XNAT

Anonymization or other scripts might obfuscate "Study Description" thus ruining "locator" assignment. See issue #57 for more information.

TODOs/WiP/Related

• add a pre-configured DICOM receiver for fully turnkey deployments
• heudiconv-monitor to fully automate conversion of the incoming data
• BIDS dataset manipulation helper