Home

Awesome

UU-Mamba

Ting Yu Tsai, Li Lin, Shu Hu, Ming-Ching Chang, Hongtu Zhu, Xin Wang

This repository is the official implementation of our paper UU-Mamba: Uncertainty-aware U-Mamba for Cardiac Image Segmentation.

We introduce a new model UU-Mamba for segmenting cardiac MRI images. The model combines the U-Mamba model, an uncertainty-aware loss function, and the SAM optimizer.

Installation

Requirements: Ubuntu: 20.04, CUDA 11.8

  1. Install Pytorch 2.0.1:
pip install torch==2.0.1 torchvision==0.15.2 --index-url https://download.pytorch.org/whl/cu118
  1. Install Mamba:
pip install causal-conv1d>=1.2.0 and pip install mamba-ssm --no-cache-dir
  1. Download code:
git clone https://github.com/tiffany9056/UU-Mamba.git
  1. cd U-Mamba/umamba and run pip install -e .
cd UU-Mamba/uumamba
pip install -e

https://github.com/tiffany9056/UU-Mamba/assets/39300518/5fcceb3a-0820-47a4-aeac-a83368102fd1

Model Training

Download dataset here and put them into the data folder. Since U-Mamba is build on the nnU-Net framework, they have updated some code in the new version, you need to update the new version using this website.

Set environment variables

The default data directory for UU-Mamba is preset to UU-Mamba/data. Users with existing nnU-Net setups who wish to use alternative directories for nnUNet_raw, nnUNet_preprocessed, and nnUNet_results can easily adjust these paths in UU-Mamba/umamba/nnunetv2/path.py to update your specific nnUNet data directory locations, as demonstrated below:

export nnUNet_raw="/UU-Mamba/data/nnUNet_raw"
export nnUNet_preprocessed="/UU-Mamba/data/nnUNet_preprocessed"
export nnUNet_results="/UU-Mamba/data/nnUNet_results"

or

# An example to set other data path,
base = '/UU-Mamba/data' # or you can set your own path, e.g., base = '/home/user_name/Documents/UU-Mamba/data'
nnUNet_raw = join(base, 'nnUNet_raw') # or change to os.environ.get('nnUNet_raw')
nnUNet_preprocessed = join(base, 'nnUNet_preprocessed') # or change to os.environ.get('nnUNet_preprocessed')
nnUNet_results = join(base, 'nnUNet_results') # or change to os.environ.get('nnUNet_results')

Verify that environment parameters are set: execute echo ${nnUNet_raw} etc to print the environment variables. This will return an empty string if they were not set.

Preprocessing

  1. Replace the new file of ACDC dataset conversion. The new file is located here. You need to execute the following lines (adapt the paths to the actual folders you intend to use).
cp /UU-Mamba/ACDC_code/Dataset027_ACDC.py /UU-Mamba/uumamba/nnunetv2/dataset_conversion/Dataset027_ACDC.py
  1. Run the dataset conversion file Dataset027_ACDC.py.
python /UU-Mamba/uumamba/nnunetv2/dataset_conversion/Dataset027_ACDC.py -i /data/ACDC/database
  1. Preprocess the data runing nnUNetv2_plan_and_preprocess, ACDC dataset ID DATASET_ID is 027.
nnUNetv2_plan_and_preprocess -d DATASET_ID --verify_dataset_integrity

Train model

ACDC dataset ID DATASET_ID is 027.

nnUNetv2_train DATASET_ID 3d_fullres all -tr nnUNetTrainerUMambaEnc

If you want to use pretrained model, download pretrain weight here and put it into pretrain_weight folder, then run

nnUNetv2_train DATASET_ID 3d_fullres all -tr nnUNetTrainerUMambaEnc -pretrained_weights /UU-Mamba/pretrain_weight/checkpoint_UU-Mamba.pth

Inference

Predict testing cases with model

Example to parameter setting:
INPUT_FOLDER  -> /UU-Mamba/data/nnUNet_raw/Dataset027_ACDC/imagesTs/
OUTPUT_FOLDER -> /results/
DATASET_ID    -> 027
CHK_NAME      -> checkpoint_best.pth or checkpoint_latest.pth or checkpoint_final.pth ...
GT_LABEL      -> /UU-Mamba/data/nnUNet_raw/Dataset111_IRIS/labelsTs
EVL_FOLDER    -> /results/

nnUNetv2_predict -i INPUT_FOLDER -o /results/seg/niigz -d DATASET_ID -c 3d_fullres -f all -tr nnUNetTrainerUMambaEnc --disable_tta -chk CHK_NAME

After predict testing cases with model, run IRIS_DSC_Eval.py to have the result DSC of each testing cases in csv file.

Example to parameter setting:
GT_LABEL   -> /UU-Mamba/data/nnUNet_raw/Dataset111_IRIS/labelsTs # Testing ground truth data path.
SEG_FOLDER -> /results/ # The output segmentation testing data path. (usually is the same as OUTPUT_FOLDER)
EVL_FOLDER -> /results/ # Where result csv file save.

python /UU-Mamba/evaluation/ACDC_DSC_Eval.py --gt_path GT_LABEL --seg_path SEG_FOLDER --save_path EVL_FOLDER

Citation

Please kindly consider citing our papers in your publications.

@misc{tsai2024uumamba,
      title={UU-Mamba: Uncertainty-aware U-Mamba for Cardiac Image Segmentation}, 
      author={Ting Yu Tsai and Li Lin and Shu Hu and Ming-Ching Chang and Hongtu Zhu and Xin Wang},
      year={2024},
      eprint={2405.17496},
      archivePrefix={arXiv},
      primaryClass={eess.IV}
}
@article{wang2024artificial,
  title={Artificial Intelligence in Image-based Cardiovascular Disease Analysis: A Comprehensive Survey and Future Outlook},
  author={Wang, Xin and Zhu, Hongtu},
  journal={arXiv:2402.03394},
  year={2024}
}