MAP-monaiclassify

Lung Nodule Detection module that can be packaged as Monai Application Package. It can be run in isolation and can also be incorporated into Mercure workflow.

This module is available as a Monai Application Package docker image that can be added to an existing Mercure installation using the docker tag: mercureimaging/map-monaiclassify:latest
It will perform lung nodule detection in CT images using the Lung nodule_ct detection MONAI bundle.

The code can be modified to deploy other detection models in the MONAI model zoo.

Installation

Add module to existing mercure installation

Follow instructions on mercure website on how to add a new module. Use the docker tag mercureimaging/map-monaiclassify:latest.

Tips

• Git clone the latest Mercure repo - do a git pull if already installed.
• vagrant --dev up (To get the latest development Mercure with MAP support.)
• Go to configurations->settings and add the following: "support_root_modules": true
• Add module - use docker image "mercureimaging/map-monaiclassify:latest"
• Make sure to select the module type as Monai.
• Make sure the "requires root user" switch is checked.
• Go to module settings and add {"HOLOSCAN_MODEL_PATH":"/opt/holoscan/models/model/lung_model.ts"} in the environment variables section.
• If not able to get the latest version of Mercure with MAP support, set the environment like this: {"HOLOSCAN_MODEL_PATH":"/opt/holoscan/models/model/lung_model.ts", "MONAI_INPUTPATH":"/tmp/data", "MONAI_OUTPUTPATH":"/tmp/output", "HOLOSCAN_INPUT_PATH":"/tmp/data", "HOLOSCAN_OUTPUT_PATH":"/tmp/output"}
• If you have a GPU available, install the NVIDIA container toolkit on your machine before enabling GPU. It can make things way faster.
• Add corresponding rule following the quick start documentation.
• Send files to Mercure

Build module for local testing, modification, and development

1. Clone repo.
2. Install the packages and dependencies using pip install -r lung_app/requirements.txt
3. Run the app locally using: python run lung_app -i <input_path> -o <output_path> -m ./models/model/lung_model.ts
4. You can make changes and package this application as your own using MAP. Packaging documentation can be found here.

Output

Lung nodule count and raw data are written to a specified output directory:

• result.json has the number of nodules count.
• output_raw.json has the bounding boxes for the nodules and scores for the predictions.
• DICOM files updated with bounding boxes around lung nodules.