One of the issues I have found with the many open source medical imaging projects, is that you can very soon find yourself in dependency hell. It seems just about every project requires a different version of ITK to work properly.
I also recently heard about docker. Docker basically gives you a development environment (or an anything environment) that runs completely within a virtual machine. However, it is efficient and lightweight enough that the performance is not significantly affected. The end result means that you can automatically and repeatably set up a development environment in Linux, with all your required dependencies, but without having to actually install any of them on your own system. Also, I use OSX at home, and Ubuntu at work, so I don't have to worry about getting all the dependencies working on two entirely different systems.
$ docker build -t stir ~/proj/stir-docker/ $ alias drun="docker run --rm -v $(pwd | sed 's/ /\\ /g'):/data -w /data" $ drun stir OSMAPOSL testdata/PETLM_pointy.par
Developing with STIR using Docker
Here are some advantages of using docker.
Isolated library dependencies
You can find and example here, where I have set up a docker for running STIR. STIR requires ITK 3, and although it is possible to have multiple versions installed on your computer, it is much easier to have the two isolated where possible.
Of course, the commands become a little longer to run. Here is the code to run STIR's OSMAPOSL reconstruction algorithm:
docker run --rm -it -v `pwd`:/data -w /data stir:base \ OSMAPOSL osmaposl_eg.par
--rm flag ensures that the container is deleted once the code
has been run. The
-it flags tell docker to run interactively and
through a pseudo tty connection respectively, so that we have the
command output echo back to us. The
-v flag mounts a directory
within the virtual machine. So we are mounting the current working
directory from the host machine to the
/data folder on the virtual
machine, and the
-w flag then tells docker to use the
directory as the virtual machine's working directory. The name of the
container being run is
stir:base: you will have had to built the
Dockerfile in the STIR-docker repository to be able to access the the
image. And finally,
OSMAPOSL osmaposl_eg.par is a command, using the
STIR library, to perform a reconstruction.
Easy installation of tools
Some tools are starting to release themselves in docker form. This is great for tools you might want to install quickly, unsure whether you want to commit to having them permanently installed on your system. Interestingly, this can even be used for GUI tools (at least on OSX and on Ubuntu, I am yet to try on Windows). One example of these is Fiji, and release of ImageJ with a number of plugins. ImageJ is a image viewing tool, written in Java. However, I rarely use Java, and would prefer not to install it.
Fiji is available on docker hub, so it is very easy to use. On Ubuntu, it can be called with:
xhost + docker run --rm -it -v /tmp/.X11-unix:/tmp/.X11-unix:ro \ -e DISPLAY=$DISPLAY -v `pwd`:/fiji/data fiji/fiji xhost -
Be sure to call
xhost - afterwards, and not to use this method on an
untrusted network 1.
Using Docker forces you to have a disposable environment with everything needed to run your program. As such, you end up with explicit documentation of everything you need to run the program. As an added bonus, if you do things write the Dockerfile should be relatively human-readable.