MeMGB-Diff

[MedIA] MeMGB-Diff: Memory-Efficient Multivariate Gaussian Bias Diffusion Model for 3D Bias Field Correction

The code project is optimized based on the official code of DDIM, and the specific library version and configuration can be referred to https://github.com/ermongroup/ddim

Running the Experiments

Train a model

python main.py --exp {PROJECT_PATH} --config mydataset.yml --doc {MODEL_NAME} --ni

Sampling from the model

python main.py --config mydataset.yml --exp {PROJECT_PATH} --doc {MODEL_NAME} --sample --sequence --eta 0 --timesteps {STEPS}

where

• STEPS controls how many timesteps used in the process.

• MODEL_NAME finds the pre-trained checkpoint according to its inferred path.

• References and Acknowledgements

@article{qiu2025memgb,
  title={MeMGB-Diff: Memory-Efficient Multivariate Gaussian Bias Diffusion Model for 3D bias field correction},
  author={Qiu, Xingyu and Liang, Dong and Luo, Gongning and Li, Xiangyu and Wang, Wei and Wang, Kuanquan and Li, Shuo},
  journal={Medical Image Analysis},
  volume={102},
  pages={103560},
  year={2025},
  publisher={Elsevier}
}