This repository contains the source code associated with our paper titled "Volumetric-Conditional-Score-based-Residual-Diffusion-Model-for-PET-MR-Denoising" which has been accepted at MICCAI 2024.
Ensure all the necessary packages listed.
numpy matplotlib scikit-learn scikit-image click requests psutil tqdm imageio imageio-ffmpeg pyspng pillow nibabel Pytorch Version: torch torchvision --index-url https://download.pytorch.org/whl/cu118
Diffusion implementation : PatchDiffusion_dnPET
U-net implementation : PatchDiffusion_simpleunet
Transformer implemntation : PatchDiffusion_transformer
To conduct experiments, please build adn run docker image using the command below. Note that you should adjust the paths and hyperparameters according to your specific requirements:
- move to the location of source code (where dockerfile is located).
- Build docker image
sudo docker build -f ./dockerfile_train -t pet_train ./- Run docker image
sudo docker run --shm-size=8G --rm --gpus all -v /home/example/:/external/ pet_trainNote; here "/home/example/" is where source code and dockerfile are located in your GPU server.
- To perform experimentsn, please build adn run docker image using the command below.
sudo docker build -f ./dockerfile_test -t pet_test ./sudo docker run --shm-size=8G --rm --gpus all -v /home/example/:/external/ pet_testPretrained model weights : https://drive.google.com/drive/folders/1jbyC63eMJE51Pz-bRBl1D28D7dqgNjfT?usp=sharing Please contact to author or leave the issue in github, if you have any question on model weights.