This repository contains code that was used to perform the experiments described in the paper "Efficient High Cone-Angle Artifact Reduction in Circular Cone-Beam CT using Symmetry-Aware Deep Learning with Dimension Reduction" by Minnema et al.
We recommend installing Conda or Miniconda for Python3 to set up the required packages to run this code. A virtual environment in (Mini)conda can be created and activated with:
env_name=*my_name*
conda create --name $env_name python=3.6
conda activate $env_name
To get the source code and install the required pacakges
git clone https://github.com/Jomigi/Cone_angle_artifact_reduction.git
cd Cone_angle_artifact_reduction
python setup.py
Download the Walnut data set from zenodo.
The script generate_training_data.py contains all code that is necessary to reproduce the reconstruction of cone-beam CT scans as performed in the paper. This includes the computation of FDK reconstruction and iterative ground truth reconstruction and the extraction of radial slices from these volumes via interpolation. Note that as described in the paper, this step is repeated 24 times with rotated scan geometries and only the radial slices close to 0° and 90° are extracted in repetition. The script takes up to one day to run for a single Walnut. To run generate_training_data.py, customize the variables in the section "user defined settings" and run:
python generate_training_data
This includes all code necessary to train an MS-D Net or a U-Net to reduce cone-angle artifacts in cone-beam CT scans. To train the networks, specify the datapaths in train.py and then run:
python train.py
This folder also contains the validation scheme (validation.sh) that was used to optimize the depth and dilations of MS-D Net as well as the number of epochs to train both CNNs.
This folder contain a single script which was used to performed to Radial-to-Cartesian re-sampling step.
To run this code:
python radial2axial.py
The code is licensed under the MIT license - see the LICENSE.md file for more details.