This repo is the official implementation for TransBTS: Multimodal Brain Tumor Segmentation Using Transformer. The multimodal brain tumor datasets (BraTS 2019 & BraTS 2020) could be acquired from here.
Architecture of 3D TransBTS.
- python 3.7
- pytorch 1.6.0
- torchvision 0.7.0
- pickle
- nibabel
After downloading the dataset from here, data preprocessing is needed which is to convert the .nii files as .pkl files and realize date normalization.
python3 preprocess.py
Run the training script on BraTS dataset. Distributed training is available for training the proposed TransBTS, where --nproc_per_node decides the numer of gpus and --master_port implys the port number.
python3 -m torch.distributed.launch --nproc_per_node=4 --master_port 20003 train.py
If you want to test the model which has been trained on the BraTS dataset, run the testing script as following.
python3 test.py
After the testing process stops, you can upload the submission file to here for the final Dice_scores.
Quantitive comparison of performance on BraTS2019 validation set as well as BraTS2020 validation set between our proposed TransBTS with other SOTA methods.
Here are some samples from BraTS 2019 dataset for visual comparison between our proposed TransBTS with other SOTA methods.
If you use our code or model in your work or find it is helpful, please cite the paper:
@inproceedings{wang2021transbts,
title={TransBTS: Multimodal Brain Tumor Segmentation Using Transformer},
author={Wang, Wenxuan and Chen, Chen and Ding, Meng and Li, Jiangyun and Yu, Hong and Zha, Sen},
booktitle={International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)},
year={2021}
}