We present ZECO, a ZeroFusion guided 3D MRI conditional generation framework that extracts, compresses, and generates highfidelity MRI images with corresponding 3D segmentation masks to mitigate data scarcity.
# Clone repository
git clone https://github.com/yourusername/zeco.git
cd zeco
# Create environment
conda create -n zeco python=3.9
conda activate zeco
# Install PyTorch (CUDA 11.8)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
# Install dependencies
pip install -r requirements.txt
pip install monai[all]Download BraTS 2020 dataset and update the path in scripts/train_*.py:
train_data_dir = "/path/to/MICCAI_BraTS2020_TrainingData"python train.py --model vqvae --channel 0 # FLAIR
python train.py --model vqvae --channel 1 # T1python train.py --model ldmpython train.py --model 3mcgFor detailed configuration, edit scripts in scripts/ directly.
# Test VQVAE
python test.py --model vqvae --checkpoint /path/to/vqvae.pth --compute_ssim
# Test LDM
python test.py --model ldm \
--checkpoint /path/to/ldm.pth \
--vqvae_checkpoint /path/to/vqvae.pth \
--compute_ssim
# Test 3MCG
python test.py --model 3mcg \
--checkpoint /path/to/controlnet.pth \
--vqvae_checkpoint /path/to/vqvae.pth \
--compute_ssim --visualizeAvailable metrics: --compute_fid, --compute_ssim, --compute_mmd
Visual Comparisons
Quantitative Results
@article{wang2025zeco,
title={ZECO: ZeroFusion Guided 3D MRI Conditional Generation},
author={Wang, Feiran and Duan, Bin and Tao, Jiachen and Sharma, Nikhil and Cai, Dawen and Yan, Yan},
journal={arXiv preprint arXiv:2503.18246},
year={2025}
}
@article{feiran2025zeco,
title={ZECO: ZeroFusion Guided 3D MRI Conditional Generation},
author={Feiran, Wang and Bin, Duan and Jiachen, Tao and Nikhil, Sharma and Dawen, Cai and Yan, Yan},
journal={IEICE Proceedings Series},
volume={93},
number={O1-2-2},
year={2025},
publisher={The Institute of Electronics, Information and Communication Engineers}
}MIT License. See LICENSE for details.