Setting up MONAILabel with Orthanc and OHIF

Principles

• uses a DICOM image storage (Orthanc) for MONAILabel tasks

  • advantage over file based storage: no conversion needed (i.e. to Niftis), original DICOM files can be used directly without any manipulation/wrangling

• resulting segmentation files are DICOM-SEGs stored in Orthanc

• uses three docker images:

  • NGINX (serves as reverse proxy)
  • Orthanc (holds image and segmentation DICOM data)
  • MONAILabel (used for active learning, model training, pre-segmentations, includes OHIF)

Installation

Prerequisites

• ensure correctly configured corporate proxy server
• make sure docker and docker-compose are installed and configured correctly
• CUDA-enabled environment strongly recommended

Bring up all relevant containers

• clone this repo:

git clone https://github.com/jlvahldiek/tutorial-monailabel-orthanc.git

• adjust parameters in ./.env first:
  • $SERVICE_HOST: machine's hostname, i.e. hostname.domain.com or just ip address
  • $LOCAL_WORKSPACE: local path to a directory that will contain:
    • db-index: Orthanc stores its database index in here (cannot be stored on network mount)
    • anatomy-model: MONAILabel stores its model files in here
  • $IMAGE_MOUNT: path to a directory where Orthanc stores its images
    • i.e. /mnt/data/ORTHANC/anatomy-db
• bring up Orthanc, OHIF and MONAILabel:

docker-compose up

• if no GPU available use

docker-compose -f docker-compose.cpu.yml up

• Orthanc will be availabel via Web http://$SERVICE_HOST:8042 or via DICOM QR $SERVICE_HOST:4242
  • populate Orthanc with images via Web app or using batch upload
• MONAILabel will be available via http://$SERVICE_HOST:8000
• all images on Orthanc can be easily visualized using OHIF http://$SERVICE_HOST:8000/ohif
• MONAILabel plugin in 3D Slicer should point to http://$SERVICE_HOST:8000
• Also consider to make use of 3D Slicer's Plugin [DICOMWebBrowser] (https://github.com/lassoan/SlicerDICOMwebBrowser), very useful

Usage

First of all populate Orthanc with DICOM images. Make sure that you import DICOM images that are captured by $SEARCH_FILTER of .env file - otherwise MONAILabel will not recognize them.

As an alternative specify paths to an existing Orthanc database: $IMAGE_MOUNT should point to Orthanc's data folder and $LOCAL_WORKSPACE/db-index should point to Orthanc's index folder.

Use 3D Slicer's MONAILabel plugin to segment images and to train new models.