Derek Merck [email protected] Rhode Island Hospital
Contributions from Vanessa Sochat <[email protected]) Stanford University School of Medicine
Configures and spins up a Docker-based open-source1 medical imaging informatics platform. Originally developed to support the RIH Clinical Imaging Research Repository (CIRR).
- Clinical/PHI Facing Repository - Orthanc 1.0 on 4280 (HTTP/REST), 4242 (DICOM)
- Clinical/PHI Facing Receiver - Orthanc on 4380 (HTTP/REST), 4342 (DICOM)
- Research/Anonymized Facing Repository - XNAT 1.6.5 on 8080 (HTTP/REST), 8042 (DICOM)
- Database - Postgresql 9.5 on 3432 (SQL)
- Data Orchestration - In progress [](Tithonus on 6080 (HTTP/REST))
- Log Monitoring - Splunk Lite on 1580 (HTTP/REST), 1514 (syslog)
- Docker, docker-compose for service virtualization
- Python, pyyaml, jinja2 for
bootstrap.py
Warning: once data has been ingested, do not use docker-compose down, or you will drop the data volume!
Furthermore, do not use docker-compose up with the xnat service; use docker-compose up --no-recreate or it will fail because it can't rebuild itself if the database already exists.
bootstrap.py will read a file called docker-compose.shadow.yml and use any override variables or config information provided there. All generated configuration files are similarly tagged as "shadow" and should not be indexed by git because they will contain plain-text account credentials. Depending on which variables are used, docker-compose.shadow.yml may not be necessary to include when creating the containers themselves.
I (@vsoch) am using the simple Orthanc server (to start) located in orthanc-docker. I've modified the code below to work with Python 3.5, however I'm starting with the simpler of these two versions. Complete instructions can be found in the README
This is an isolated Orthanc using a Postgres backend that can be created directly using docker-compose from the orthancp-docker directory. This version has a much simpler configuration process, and by default it will create services on a separate network.
Orthanc and XNAT services rely on Postgres. The Postgres service has to be available before Orthanc or XNAT can be configured.
$ docker-compose up -d postgres
First you want to set up orthanc.json from the template with the configuration from the running database. You can do this as follows:
$ python bootstrap.py orthanc
$ docker-compose up -d orthancOrthanc with receiver proxy:
$ python bootstrap.py orthanc orthanc-receiver
$ docker-compose up -d orthanc orthanc-recieverThe additional DICOM receiver can be used as a proxy to accept DICOM transfers and queue them for the main clinical-facing repository. The main repo slows down considerably as the DB grows large, particularly if compression is on.2
$ python bootstrap.py xnat # Initializes config from template, creates image, drops db if it exists
$ docker-compose up xnat$ python bootstrap.py orthanc xnat
$ docker-compose up orthanc xnat splunk
$ python bootstrap.py orthanc orthanc-receiver xnat
$ docker-compose upTo inspect the data or logs, mount the data volumes on another container.
$ docker-compose up -f docker-compose.admin.yml admin
Or manually for a single service:
$ docker-compose run -it --volumes-from orthanc ubuntu /bin/bashTo perform a data backup, run admin or otherwise mount the volumes and use tar.
$ docker-compose run -f docker-compose.admin.yml admin tar zcvf /backup/postgres.tar.gz /var/lib/postgresql/data
$ docker-compose run -f docker-compose.admin.yml admin tar zcvf /backup/xnat.tar.gz /var/lib/xnat/dataUses Docker images from:
Run this command to configure your shell to use docker-compose on OSX or Windows:
$ eval "$(docker-machine env default)"
MIT
1: Splunk is not open source, but Splunk Lite will work for this volume of logs and it is free. Replace it with you open-source syslog server of choice if necessary.
2: On a reasonable machine, we measured about 20 images/second in an empty, uncompressed repo, but only about 1.5 scans/sec in a repo w 100k instances and compression on.