The goal of this challenge is multimodal image registration between pre-operative MR and intra-operative Ultrasound for the prostate gland. Details of the challenge are available here.
In this repository we build two simple baselines (i.e., variants of localnet and voxelmorph, with simplified backbone networks) for use with our dataset. The usage instructions are outlined below. Note that these are small, simplified networks, using resampled smaller images for training, for demonstration purposes on a wide variety of hardware. The model checkpoints after training with the dataset may be found here for localnet and here for voxelmorph.
git clone https://github.com/muregpro/Baseline-Networks.git
The dataset may be downloaded from this link. For training, two directories are requried: nifti_data/train and nifti_data/val and the user may split data into these, before training. The directory nifti_data must then be placed into the Baseline-Networks directory, so that they may be added into the docker container (see below).
Note: sudo or docker group permissions may be needed to run the following commands.
- navigate to the root directory
cd Baseline-Networks
- build mureg docker image from Dockerfile
docker build -t mureg .
- create mureg docker container from mureg docker image
docker container create -it --name mureg mureg
- start the mureg docker container
docker container start mureg
docker exec mureg <command>
Examples:
docker exec mureg ls
docker exec mureg python3 train_localnet.py
docker exec mureg python3 train_voxelmorph.py
docker exec mureg python3 test_localnet.py
docker exec mureg python3 test_voxelmorph.py
docker container stop mureg
docker container rm -f mureg