search

jmaces / aapm-ct-challenge

AAPM DL Sparse-View Computed Tomography Image Reconstruction Challenge - Team Robust-And-Stable
MIT License
44 stars 6 forks source link

readme

AAPM DL-Sparse-View CT Challenge Submission - Designing an Iterative Network for Fanbeam-CT with Unknown Geometry

GitHub license code-style black made-with-python made-with-pytorch

Team: robust-and-stable

This repository provides the implementation of our submission to the AAPM DL-Sparse-View CT Challenge.

More details can be found in our ICML2022 paper Near-Exact Recovery for Tomographic Inverse Problems via Deep Learning (by M. Genzel, I. Gühring, J. Macdonald, and M. März) and our short challenge submission report Designing an Iterative Network for Fanbeam-CT with Unknown Geometry.

Usage

The repository contains code to train the complete pipeline (Operator -> UNet -> ItNet -> ItNet-post) of our proposed reconstruction method, as well as for two comparison networks (Tiramisu & Learned Primal Dual).

The challenge data is not contained in this repository and needs to be obtained directly from the challenge website.

  1. Check (and modify if necessary) the configuration file config.py. It specifies the directory paths for the data and results. By default, the data should be stored in the subdirectory raw_data and results and model weights are stored in the subdirectory results.
  2. Identify the forward operator using the scripts named script_radon_indentify.py and script_radon_learn_inv.py
  3. You can check and evaluate the identified operator using the script named script_evaluate_operator.py.
  4. Train networks using the scripts named script_train_*.py.
  5. You can evaluate the trained networks on the test data using the scripts named script_evaluate_test_*.py

Requirements

The package versions are the ones we used. Other versions might work as well.

cudatoolkit (v10.1.243)
matplotlib (v3.1.3)
numpy (v1.18.1)
pandas (v1.0.5)
python (v3.8.3)
pytorch (v1.6.0)
torchvision (v0.7.0)
tqdm (v4.46.0)

Acknowledgements

Our implementation of the U-Net is based on and adapted from https://github.com/mateuszbuda/brain-segmentation-pytorch/.
Our implementation of the Tiramisu network is based on and adapted from https://github.com/bfortuner/pytorch_tiramisu/.
Our implementation of the Learned Primal Dual network is inspired by https://github.com/adler-j/learned_primal_dual/.

Thank you for making your code available.

License

This repository is MIT licensed, as found in the LICENSE file.