This is the official implementation of the paper "CoreDiff: Contextual Error-Modulated Generalized Diffusion Model for Low-Dose CT Denoising and Generalization". The pre-print version can be found in arxiv; the early access version can be found in TMI.

• Dec, 2023: Updated the code for preprocessing the original Mayo 2016 "DICOM" format data (data_preporcess/prep_mayo2016.py) and its corresponding training demo (train_mayo2016.sh).
• Oct, 2023: initial commit.

• The AAPM-Mayo dataset can be found from: Mayo 2016.
• The "Low Dose CT Image and Projection Data" can be found from Mayo 2020.
• The Piglet Dataset can be found from: SAGAN.
• The Phantom Dataset can be found from: XNAT.

Please check train.sh for training script (or test.sh for inference script) once the data is well prepared. Specify the setting in the script, and simply run it in the terminal.

For one-shot learning framework，please check train_osl_framework_training.sh for training script (or test_osl_framework.sh for inference script)

These curves are calculated based on our simulated 5% dose data.

- Linux Platform
- python==3.8.13
- cuda==10.2
- torch==1.10.1
- torchvision=0.11.2
- numpy=1.23.1
- scipy==1.10.1
- h5py=3.7.0
- pydicom=2.3.1
- natsort=8.2.0
- scikit-image=0.21.0
- einops=0.4.1
- tqdm=4.64.1
- wandb=0.13.3

• Our codebase builds heavily on DU-GAN and Cold Diffusion. Thanks for open-sourcing!
• Low-dose CT data simulation refers to LD-CT-simulation. Thanks for open-sourcing!

If you find our work and code helpful, please kindly cite the corresponding paper:

@article{gao2023corediff,
  title={CoreDiff: Contextual Error-Modulated Generalized Diffusion Model for Low-Dose CT Denoising and Generalization},
  author={Gao, Qi and Li, Zilong and Zhang, Junping and Zhang, Yi and Shan, Hongming},
  journal={IEEE Transactions on Medical Imaging},
  year={2023}
}