TRABIT team for BraTS 2021 and FeTA 2021 using MONAI

Our code reproduces the segmentation performance of nnU-Net using MONAI.

Installation with Docker (recommended)

Create the docker image brats21/trabit by running

sh build docker.sh

Your nvidia driver needs to be version 470.42 or higher.

You can then create a docker container named trabit using a command like

nvidia-docker run --ipc=host -it -v ~/workspace:/workspace -v ~/data:/data --name trabit brats21/trabit:latest

Feel free to change which folders you mount with -v .

Installation (without Docker)

Install MONAI and all its dependencies with

python -m pip install 'monai[all]'==0.5.2

(Optional) To use the Generalized Wasserstein Dice loss you have to run in a terminal

python -m pip install git+https://github.com/LucasFidon/GeneralizedWassersteinDiceLoss.git

(Optional) To use the AdamP or SGDP optimizers first install the adamp library using

python -m pip install adamp

(Optional) to use partially supervised learning (as we did for the FeTA 2021 challenge) you have to install the label-set loss functions package

python -m pip install git+https://github.com/LucasFidon/label-set-loss-functions.git

Instruction to use the code for BraTS (What you have to change)

You need to change the paths in /dataset_config/brats.yml

Instruction for distributed training e.g. using GPU 0 3 5

python3 -m torch.distributed.launch --nproc_per_node=3 run_train.py --cuda_visible_device 0 3 5 --config <config file> --data_config <data config file>

Instruction to extend the code to a new segmentation application (What you have to change)

For training:

• You need to create a dataset config file for your application in /dataset_config
• You need to define a data augmentation/preprocessing pipeline for your dataset in /data
• You need to define a dataloader pipeline for your dataset in /data
• You need to update data/factory to use your data augmentation pipeline and dataloader for your dataset

BraTS inference

The weights of our model for the BraTS 2021 challenge can be downloaded at https://drive.google.com/drive/folders/1b9rZDv9HHCxiEtiHz4h4I1vvjzScivpg?usp=sharing

Please put the weights in the folder \trained_weights and make sure the subfolders are named model_i for i integer from 1 to 7

How to cite

If you find this repository useful for your research, please cite our work

• L. Fidon, S. Shit, I. Ezhov, J. C. Paetzold, S. Ourselin, T. Vercauteren. Generalized Wasserstein Dice Loss, Test-time Augmentation, and Transformers for the BraTS 2021 challenge. BraTS 2021.
• L. Fidon, M. Aertsen, S. Shit, P. Demaerel, S. Ourselin, J. Deprest, T. Vercauteren. Partial supervision for the FeTA challenge 2021. FeTA Challenge 2021.

BibTeX:

@article{fidon2021generalized,
  title={Generalized Wasserstein Dice Loss, Test-time Augmentation, and Transformers for the BraTS 2021 challenge},
  author={Fidon, Lucas and Shit, Suprosanna and Ezhov, Ivan and Paetzold, Johannes C and Ourselin, S{\'e}bastien and Vercauteren, Tom},
  journal={arXiv preprint arXiv:2112.13054},
  year={2021}
}
@article{fidon2021partial,
  title={Partial supervision for the FeTA challenge 2021},
  author={Fidon, Lucas and Aertsen, Michael and Shit, Suprosanna and Demaerel, Philippe and Ourselin, S{\'e}bastien and Deprest, Jan and Vercauteren, Tom},
  journal={arXiv preprint arXiv:2111.02408},
  year={2021}
}