MS-TCNet: An effective Transformer–CNN combined network using multi-scale feature learning for 3D medical image segmentation

paper

Installation

git clone https://github.com/AustinYuAo/MS-TCNet.git

cd MS-TCNet

conda env create -f environment.yml

source activate MS-TCNet

Dataset

1.Download

Synapse dataset download

ACDC dataset download

MSD BraTS dataset download

2. your dataset folders should be organized as follows:

├── dataset/  
    ├── Synapse/  
        ├── imagesTr/  
        ├── imagesTs/  
        ├── labelsTr/  
        ├── labelsTs/  
        ├── dataset.json  
    ├── ACDC/  
        ├── imagesTr/  
        ├── imagesTs/  
        ├── labelsTr/  
        ├── labelsTs/  
        ├── dataset.json  
    ├── MSD BraTS/    
        ├── imagesTr/   
        ├── imagesTs/  
        ├── labelsTr/  
        ├── labelsTs/  
        ├── dataset.json  

You can refer to the corresponding JSON files for the data partitioning of each dataset. We have stored these files in the data json folder. You can also copy these files to the corresponding dataset folder for training.

Trainning

Synapse dataset

python main.py --max_epochs=8000 --batch_size=2 --logdir=mstcnet_synapse --save_checkpoint --data_dir=your_dataset_dir --json_list=dataset_Synapse.json --model_name=mstcnet_synapse --workers=6

ACDC dataset

python main.py --max_epochs=8000 --batch_size=8 --sw_batch_size=4 --in_channels=1 --out_channels=4 --space_x=1.25 --space_y=1.25 --space_z=10 --roi_x=128 --roi_y=128 --roi_z=6 --logdir=mstcnet_acdc --save_checkpoint --data_dir=your_dataset_dir --json_list=dataset_ACDC.json --model_name=mstcnet_acdc --workers=6

MSD BraTS dataset

python main.py --max_epochs=800 --batch_size=3 --sw_batch_size=2 --in_channels=4 --out_channels=3 --a_min=0 --a_max=300 --b_min=0 --b_max=1.0 --space_x=1 --space_y=1 --space_z=1 --roi_x=128 --roi_y=128 --roi_z=128 --logdir=mstcnet_brats --save_checkpoint --data_dir=your_dataset_dir --json_list=dataset_BraTS.json --model_name=mstcnet_brats --val_every=50 --workers=6

Test

Synapse dataset

python test_synapse.py --data_dir=your_dataset_dir --json_list=dataset_Synapse.json --pretrained_dir=your_pretrained_dir --pretrained_model_name=model.pth --saved_checkpoint=ckpt

ACDC dataset

python test_acdc.py --data_dir=your_dataset_dir --json_list=dataset_ACDC.json --pretrained_dir=your_pretrained_dir --pretrained_model_name=model.pth --saved_checkpoint=ckpt

MSD BraTS dataset

python test_brats.py --data_dir=your_dataset_dir --json_list=dataset_BraTS.json --pretrained_dir=your_pretrained_dir --pretrained_model_name=model.pth --saved_checkpoint=ckpt

References

@article{ao2024ms,
  title={MS-TCNet: An effective Transformer--CNN combined network using multi-scale feature learning for 3D medical image segmentation},
  author={Ao, Yu and Shi, Weili and Ji, Bai and Miao, Yu and He, Wei and Jiang, Zhengang},
  journal={Computers in Biology and Medicine},
  volume={170},
  pages={108057},
  year={2024},
  publisher={Elsevier}
}

Acknowledgement

The code is implemented based on UNETR. We would like to express our sincere thanks to the contributors.