ENH: Docker-compose

[mgxd]

Addresses #24

This PR adds docker-compose file which is made up of:

• a DICOM receiver
• an automated heudiconv trigger on receiver output directory

[yarikoptic]

Could someone remind me pragmatic benefits from compose, compared to a single docker image shipping all needed components?

[satra]

most people already have dicom receivers. but our collabs at northeastern wanted an easy to use setup. docker-compose allows us to use different docker images without worrying about interactions within the images. for example i have used the dcmtk image in the past to do a bunch of things without having to create my own image.

i don't think i have any fundamental objection to things being in the same container. just wanted to keep the heudiconv bit separate from the dicom-receiver and bids processing bits.

[mirestrepo]

Hi All,

At Brown University we are working on similar efforts. I'm wondering where the docker file for mgxd/dcmrecv come from. We are thinking to using Orthanc but I'm wondering if dcmrecv would be a better (slimmer) alternative for the receiver part.

Thanks!

[mgxd]

@mirestrepo here is the Dockerfile for the image

[yarikoptic]

Please accept my apologies for forgetting about this PR. It would be great to revive/finalize and use it. I wonder if this (or derived) proposed setup was used anywhere by now?

Meanwhile I was prototyping a reproin helper to reflect my experiences of working with reproin'ed acquisitions in the past year. That helper provides interfaces for

• producing lists of acquisitions from year/month/day/accession directories (reproin lists-update using `heudiconv -f reproin --command list) -- something to run on cron
• be able to show (quickly) for a study which accessions were converted or not yet (reproin study-show)
• convert not yet converted accessions (reproin study-convert)
  • refuses to convert a subject/session if that one was found already present
• mark some accessions to be skipped (reproin study-accession-skip) (e.g. for pilot runs etc)
• "integrate" with bids-validator and populate output of the validation within the converted datast (all the reproin validator* commands)
• it is using heudiconv with --datalad but relies on reproin study-create to create an initial dataset

Ideally I would like to expand that reproin helper to support using bundled (if present) within dataset (singularity) container so that the environment used to convert is present within dataset so all subjects would be converted (or reconverted) consistently. (also reproin would error if found dcm2niix version is different from the one in the dataset metadata)

So I am thinking, if this setup was not yet used, we could RF it to accommodate reproin helper setup.

[mgxd]

I believe this was prepared for use in northeastern's imaging center, but @satra would know if was actually implemented.

If not, I'm fine with this being a starter point for the reproin implementation

[satra]

i don't think they ended up doing this, but still worth a setup i think both for testing, but also for people to potentially use. not a super high priority given that most places already have some kind of receiver in place.

[yarikoptic]

I will convert to draft to not be as vivid for now