SCOPE

This repository contains the PyTorch implementation of our manuscript "Self-Supervised Coordinate Projection Network for Sparse-View Computed Tomography". [ArXiv] [IEEE Xplore]

1. Main Running Environment

To run this project, you will need the following packages:

• PyTorch
• tinycudann
• SimpleITK, tqdm, numpy, and other packages.

2. File Tree

SCOPE
│  config.json          # configuration file.
│  dataset.pyc
│  eval.py              # evaluates the reconstruted CT result.
│  README.md
│  reprojection.py      # generates DV sinogram via the fine-trained MLP.
│  scope.pyc
│  train.py             # trains the MLP network.
│  utils.py
│
├─data
│      90_img.nii       # CT image by FBP on the SV sinogram (90_sino.nii).
│      90_sino.nii      # SV sinogram (input data).
│      gt_img.nii       # GT CT image by FBP on the GT DV sinogram (gt_sino.nii).
│      gt_sino.nii      # GT DV sinogram (reference data).
│
├─model
│      checkpoint.pth   # pre-trained model for SV sinogram (90_sino.nii).
│
├─output
│  ├─img
│  │      scope_recon.nii     # Our reconstructed result.
│  │
│  └─sino
│          720_sino_pre.nii   # DV sinogram generated by SCOPE.
│
└─script_matlab
        gene_angle.m
        gene_img.m      # matlab script for FBP algorithm.

3. Training and Re-projection

To train the model from scratch, navigate to ./ and run the following command in your terminal:

python train.py

This will train the model for the input sinogram (90_sino.nii). The pre-trained model will be stored in ./model.

Next, go to ./ and run the following command in your terminal for reprojting DV sinogram:

python reprojection.py

This will generate the DV sinogram, which will be stored in output/sino.

Finally, navigate to ./script_matlab and use MATLAB to run gene_img.m to recover the final CT image, which will be stored in ./output/img.

4. Evaluation

To qualitatively evalute the result, navigate to ./ and run the following comman in your terminal:

python eval.py

This will compute PSNR and SSIM values for the reconstruced image (./output/img/scope_recon.nii). PSNR and SSIM are respectively 40.45 dB and 0.9794 for our provied result.

5. License

This code is available for non-commercial research and education purposes only. It is not allowed to be reproduced, exchanged, sold, or used for profit.

6. Citation

If you find our work useful in your research, please cite:

@ARTICLE{10143286,
  author={Wu, Qing and Feng, Ruimin and Wei, Hongjiang and Yu, Jingyi and Zhang, Yuyao},
  journal={IEEE Transactions on Computational Imaging}, 
  title={Self-Supervised Coordinate Projection Network for Sparse-View Computed Tomography}, 
  year={2023},
  volume={9},
  number={},
  pages={517-529},
  doi={10.1109/TCI.2023.3281196}}