Optimizing CycleGAN Design for CBCT-to-CT Translation: Insights into 2D vs 3D Modeling, Patch Size, and the Need for Tailored Evaluation Metrics
We used ganslate framework for these experiments.
This repository contains the experiment configuration files and PyTorch dataset loading and preprocessing that we used in our paper.
It is organized into two folders:
-
LungProton- experiments and dataset classes for lung CBCT-to-CT translation for adaptive proton radiotherapy -
CervixPhoton- experiments and dataset classes for cervix CBCT-to-CT translation for adaptive photon radiotherapy
To run the experiments (if you have access to the data):
pip install ganslate
git clone [email protected]:ganslate-team/CBCT-to-CT-cycle-consistent-GANs.git
cd CBCT-to-CT-cycle-consistent-GANs
# change the config file accordingly
ganslate train config="./LungProton/experiments/1_2D_cyclegan_vnet.yaml"