Medical Image Analysis using Python and Docker

Rudolph Pienaar, PhD

Staff Scientist, Boston Children's Hospital

rudolph.pienaar@childrens.harvard.edu

quick "bio"

  • MRI analysis
  • Software development
  • Web-based workflows
    • Imaging on the web
    • HPC via web front end
  • encapsulating compute in "docks"

overview

  • introduction to docks
  • "lay of the land"
  • practical examples
  • coding our own example in a dock!

what are docks?

  • simply: docks are a mechanism to encapsulate an application and all its dependencies in a manner that is logically isolated from other contained apps;
  • can be thought of as something between a chroot and a virtual machine

the view from 30,000 feet...

or 10,000m :-)

the lay of the land...

  • what are we going to do?
  • how are we going to do this?
  • why are we going to do this?

what are we going to do?

  • talk about docks;
  • use some sample/example docks;
  • create a python app in a dock;
  • create a web-server dock;
  • seamlessly share data and results between the python dock and the web server dock;

how are we going to do it?

  • fire up some docks ourselves;
  • interface with these docks;
  • follow a tutorial;
  • build our own!

why are we going to do it?

  • demonstrate how powerful and portable docks are;
  • and how fun they can be!

first...

  • some conceptual views on apps...

application / hardware

application / VM hardware

application / VM full

application / streamline

application / containers

application / containers

practicalities...

  • all docks/containers run on the same underlying kernel as the host
  • pros:
    • fast (almost immediate) execution
    • no startup penalty
  • cons:
    • all "apps" must inherently run on the same underlying Linux kernel
    • all "apps" are thus ultimately Linux apps

practicalities...

  • install the docker app on
    • linux (easiest)
    • mac (close second)
    • windows (depends... :-)
  • import some pre-made docker images
http://book.orthanc-server.com/users/docker.html

exercise

  • start a docker running a PACS server
  • start another dock running a stripped down ubuntu
  • send information stored on the host from one dock to another...
# Run the Orthanc PACS container
docker run -p 4242:4242 -p 8042:8042 --rm                      \
    -v ~/research/chris/data/orthanc-db/:/var/lib/orthanc/db/  \
    jodogne/orthanc-plugins

# Run the "FNNDSC" ubuntu container
docker run  -v /Users:/Users -it ubuntu-fnndsc
    
    # On first setup, create group and user:
    addgroup --gid 1102 fnndsc
    adduser --disabled-password --gecos '' --uid 6244 --gid 1102 rudolph 

    # exit
    exit

# "commit" the changes in this dock:
docker ps -a
docker commit 3f06ea36c5ad ubuntu-fnndsc

# Now run the container again, and connect tunnels...
docker run  -v /Users:/Users -it ubuntu-fnndsc

    # connect a forward tunnel from 4242 in the dock to the host 4242
    ssh -g -f -N -X -L 4242:localhost:4242 rudolph.pienaar@172.16.32.31

    # Now, goto some dir containing DCM files on the host FS
    cd someDir

    # and transmit them via the tunnel to the Orthanc dock!
    dicom_dirSend.bash -v 10 -a rudolph -h localhost -p 4242 -E dcm [a-z]* [A-Z]*

# For kicks, attach to the Orthanc dock
docker ps
docker exec -i -t e99a7e5897dd /bin/bash

exercise

  • start a docker running a web server
  • edit a document on the host and display via the docked web server!
# Run the web server container
docker run -d -p 8000:80 nginx 

# Connect to this container and look at the html dir...
docker exec -it <ID> /bin/bash
    
    # The nginx html dir is
    cd /usr/share/nginx/html
    
    # exit the container 
    exit

# and map a local dir "over" this one
docker run -d -p 8000:80 \ 
    -v ~/research/chris/web/:/usr/share/nginx/html/ \
    --name webserver nginx

exercise

  • Now, let's build our python app!
# Run the ubuntu dock with port forwarding for the ipython notebook

docker run  -v /Users:/Users \
    -p 8888:8888 -p 8000:80  \
    -it ubuntu-fnndsc

    # Now run the ipython notebook in the dock
    ipython3 notebook --ip="*"

# and connect to it from the host on localhost:8888


# useful cleanup!
docker rm -v $(docker ps -a -q -f status=exited)

let's program!

finis

Medical image analysis with Python and docks

By Rudolph Pienaar

Medical image analysis with Python and docks

  • 1,320