PyTorch implementation of the paper "Deep Learning Computed Tomography based on the Defrise and Clack Algorithm". This repository includes the code for a data-driven methodology for reconstructing CBCT projections for a given orbit.

The Defrise and Clack Neural Network code is developed using Python 3.11, PyTorch 2.1.1 and PyTorch-lightning 2.1.2. To ensure compatibility, please install the necessary packages using the following commands to create and activate a conda environment:

conda env create -f environment.yml
conda activate Defrise_and_Clack_Neural_Network

The simulation data set used for training can be generated by executing simulated data/data_gen_2.py.

The pre-trained model is saved under the path checkpoints/checkpoint.ckpt.

This repository is organized as follows:

• simulated data/data_gen_2.py: This script is responsible for generating the dataset.

• dataset.py: This script is responsible for handling the dataset.

• DandCReconstrucion.py: Contains the implementation of the Defrise and Clack Neural Network.

• intermediateFunction.py: Calculation for the Grangeat's intermediate function.

• weight.py: Defines the weight layers required in the reconstruction process.

• train.py: Execute it to train the neural network.

• reference.py: Execute it to test the neural network.

@article{ye2024deep,
  title={Deep Learning Computed Tomography based on the Defrise and Clack Algorithm},
  author={Ye, Chengze and Schneider, Linda-Sophie and Sun, Yipeng and Maier, Andreas},
  journal={arXiv preprint arXiv:2403.00426},
  year={2024}
}

• Thanks to sypsyp97 for his Eagle_Loss, which was a great reference in building this application.