This is the code for 4D-CT Lung Deformable Image Registration.
You can see the overall framework named FFC3DLungDIR which has been used in this study. Inspired by, Fast Fourier Convolution paper, we utilized the 3D version of this operator (FFC3D) to increase the receptive field of typical convolution operators to capture global and local information in feature maps simultaneously.
In order to install the required packages, run this command in your cmd or terminal:
pip install -r requirements.txt
For training the model, we have used CREATIS dataset consisting 6 sets of 4D-CT lung images. As a test set, we utilized publicly available DIRLAB dataset including 10 sets of 4D-CT lung images each containing images of ten respiratory cycle phases.
Configure FFCResnetGenerator_settings.yaml to build your desired architecture. Also, for setting the training parameters e.g., hyperparameters, modify training_settings.yaml. Then, for training it, run train.py using the following command in your terminal or cmd:
python train.py --exp [experiment number] --training_config_path path/to/model/config --model_config_path [path/to/model/config] --training_config_path [path/to/training/config]
