README.md

ivdm3seg

Results of our method on the non-disclosed IVDM3Seg challenge data as generated by the challenge organizers using our prepared Docker containers as described in the paper.

Docker containers and results

• binary w/ latent variable
  • Container name: ivdm3seg/binary
  • Container image
  • Results
• ternary
  • Container name: ivdm3seg/ternary
  • Container image
  • Results
• binary w/ latent variable (larger localizer patch size)
  • Container name: ivdm3seg/binary_large
  • Container image
  • Results

How to run a Docker container

1. Download the container un-7zip it, such that you are left with an IMG.tar file
2. Load the image into Docker, i.e., docker load --input IMG.tar
3. Finnally, you can use the instructions as specified by the challenge to obtain segmentation results for an image:

   CID=`docker run -dit --runtime=nvidia --rm -v /input -v /output CONTAINER_NAME`
   
   docker cp TESTIMAGE_fat.nii.gz $CID:/input/fat.nii.gz
   docker cp TESTIMAGE_wat.nii.gz $CID:/input/wat.nii.gz
   docker cp TESTIMAGE_inn.nii.gz $CID:/input/inn.nii.gz
   docker cp TESTIMAGE_opp.nii.gz $CID:/input/opp.nii.gz
   
   docker exec $CID /challenge/run
   
   docker cp $CID:/output/result.nii.gz TESTIMAGE_segmentation.nii.gz
   
   docker stop $CID

Note: You need the NVIDIA runtime to run the Docker container.