We built a system to convert annotated nii files into pixel contours, then fill those contours and render them into grids or mp4s.

We wanted to render both the MRI dicoms and make an augmented display of contours around labeled tumors to visually inspect how predictive models are performing.

We used pydom and matplot lib for rendering. We also developed some utility methods for loading nii data, converting nii labels into per-dicom/slice contours, used matplotlib and shapely to augment the dicom display, and used ffmpeg to convert plots images into an mp4 video.

MP4 Contour Rendering (unlisted YouTube video)

It would have been nice to have more attribute data about segmented tumors (texture, density, etc). It also would have been nice to already have nii data in a per-dicom data structure.

Integrate these rendering tools into predictive model pipelines.

No extra tools are needed for rendering, it can function stand-alone.

Computer vision.

Open docker or run our Jupyter notebooks.

*The Docker image contains <R/jupyter> notebooks of all analyses and the dependencies to run them. Be sure to note if you need any special credentials to access data for these analyses, don't package restricted data in your containers!

Instructions for running the following notebooks: be sure to adjust these instructions as necessary! check out https://github.com/Sage-Bionetworks/nf-hackathon-2019 for example containers and instructions

1. docker pull <your dockerhub repo>/<this container> command to pull the image from the DockerHub
2. docker run <your dockerhub repo>/<this container> Run the docker image from the master shell script