This is the official implementation of the paper "FreeSeed: Frequency-band-aware and Self-guided Network for Sparse-view CT Reconstruction" (arxiv, springer).
dudo_trainer.py
(and corresponding details in basic_wrapper_v2.py
, dudofree.py
, main.py
, train.sh
)The AAPM-Myo dataset can be downloaded from: CT Clinical Innovation Center
(or the box link). Please walk through Please see here for simple data preprocessing../datasets/process_aapm.ipynb
for more details on preparing the dataset.
Please check train.sh
for training script (or test.sh
for inference script) once the data is well prepared. Specify the dataset path and other setting in the script, and simply run it in the terminal.
Notably, it is time-consuming to directly train sinogram-domain sub-network and image-domain sub-network of FreeSeedDUDO using a combination of loss functions simultaneously.
A more efficient way, as in dudo_trainer.py
, is to:
- Linux Platform
- python==3.7.16
- torch==1.7.1+cu110 # depends on the CUDA version of your machine
- torchaudio==0.7.2
- torchvision==0.8.2+cu110
- torch-radon==1.0.0
- monai==1.0.1
- scipy==1.7.3
- einops==0.6.1
- opencv-python==4.7.0.72
- SimpleITK==2.2.1
- numpy==1.21.6
- pandas==1.3.5 # optional
- tensorboard==2.11.2 # optional
- wandb==0.15.2 # optional
- tqdm==4.65.0 # optional
We choose torch-radon toolbox to build our framework because it processes tomography really fast! For those who have problems installing torch-radon toolbox:
python setup.py install
without triggering too many compilation errors🤔.If you find our work and code helpful, please kindly cite the corresponding paper:
@inproceedings{ma2023freeseed,
title={FreeSeed: Frequency-band-aware and Self-guided Network for Sparse-view CT Reconstruction},
author={Ma, Chenglong and Li, Zilong and Zhang, Yi and Zhang, Junping and Shan, Hongming},
booktitle={Medical Image Computing and Computer Assisted Intervention -- MICCAI 2023},
year={2023}
}