This repository contains the code to our paper "Medical Diffusion: Denoising Diffusion Probabilistic Models for 3D Medical Image Synthesis" (see https://arxiv.org/abs/2211.03364).
This code has been tested on Ubuntu 20.04 and an NVIDIA Quadro RTX 6000 GPU. Furthermore it was developed using Python v3.8.
In order to run our model, we suggest you create a virtual environment
conda create -n medicaldiffusion python=3.8
and activate it with
conda activate medicaldiffusion
Subsequently, download and install the required libraries by running
pip install -r requirements.txt
Once all libraries are installed and the datasets have been downloaded, you are ready to train the model:
First, we need to train the three-dimensional VQ-GAN model. To do so in the BraTS dataset, you can run the following command:
PL_TORCH_DISTRIBUTED_BACKEND=gloo python train/train_vqgan.py dataset=brats dataset.root_dir=<INSERT_PATH_TO_BRATS_DATASET> model=vq_gan_3d model.gpus=1 model.default_root_dir_postfix='flair' model.precision=16 model.embedding_dim=8 model.n_hiddens=16 model.downsample=[2,2,2] model.num_workers=32 model.gradient_clip_val=1.0 model.lr=3e-4 model.discriminator_iter_start=10000 model.perceptual_weight=4 model.image_gan_weight=1 model.video_gan_weight=1 model.gan_feat_weight=4 model.batch_size=2 model.n_codes=16384 model.accumulate_grad_batches=1
Note that you need to provide the path to the dataset (e.g. dataset.root_dir='/data/BraTS/BraTS 2020') to successfully run the command.
To train the diffusion model in the latent space of the previously trained VQ-GAN model, you need to run the following command
python train/train_ddpm.py model=ddpm dataset=brats model.results_folder_postfix='flair' model.vqgan_ckpt=<INSERT_PATH_TO_CHECKPOINT> model.diffusion_img_size=32 model.diffusion_depth_size=32 model.diffusion_num_channels=8 model.dim_mults=[1,2,4,8] model.batch_size=10 model.gpus=1
Where you again need to specify the path to the VQ-GAN checkpoint from before (e.g. model.vqgan_ckpt='/home/<user>/Desktop/medicaldiffusion/checkpoints/vq_gan/BRATS/flair/lightning_logs/version_0/checkpoints/latest_checkpoint.ckpt')
To simpify the dataloading for your own dataset, we provide a default dataset that simply requires the path to the folder with your NifTI images inside, i.e.
root_dir/ # Path to the folder that contains the images
├── img1.nii # The name of the NifTI file is not important
├── img2.nii
├── img3.nii
├── ...
All you need to do now is just specify the path to this root directory the way we have dealt with it before, i.e.,
PL_TORCH_DISTRIBUTED_BACKEND=gloo python train/train_vqgan.py dataset=default dataset.root_dir=<INSERT_PATH_TO_ROOT_DIRECTORY> model=vq_gan_3d model.gpus=1 model.default_root_dir_postfix='own_dataset' model.precision=16 model.embedding_dim=8 model.n_hiddens=16 model.downsample=[2,2,2] model.num_workers=32 model.gradient_clip_val=1.0 model.lr=3e-4 model.discriminator_iter_start=10000 model.perceptual_weight=4 model.image_gan_weight=1 model.video_gan_weight=1 model.gan_feat_weight=4 model.batch_size=2 model.n_codes=16384 model.accumulate_grad_batches=1
Note that you need to provide the path to the dataset (e.g. dataset.root_dir='/../../root_dir/') to successfully run the command.
To train the diffusion model in the latent space of the previously trained VQ-GAN model, you need to run the following command
python train/train_ddpm.py model=ddpm dataset=default model.results_folder_postfix='own_dataset' model.vqgan_ckpt=<INSERT_PATH_TO_CHECKPOINT> model.diffusion_img_size=32 model.diffusion_depth_size=32 model.diffusion_num_channels=8 model.dim_mults=[1,2,4,8] model.batch_size=10 model.gpus=1
Where you again need to specify the path to the VQ-GAN checkpoint from before (e.g. model.vqgan_ckpt='/home/<user>/Desktop/medicaldiffusion/checkpoints/vq_gan/DEFAULT/own_dataset/lightning_logs/version_0/checkpoints/latest_checkpoint.ckpt')
To cite our work, please use
@misc{https://doi.org/10.48550/arxiv.2211.03364,
doi = {10.48550/ARXIV.2211.03364},
url = {https://arxiv.org/abs/2211.03364},
author = {Khader, Firas and Mueller-Franzes, Gustav and Arasteh, Soroosh Tayebi and Han, Tianyu and Haarburger, Christoph and Schulze-Hagen, Maximilian and Schad, Philipp and Engelhardt, Sandy and Baessler, Bettina and Foersch, Sebastian and Stegmaier, Johannes and Kuhl, Christiane and Nebelung, Sven and Kather, Jakob Nikolas and Truhn, Daniel},
title = {Medical Diffusion - Denoising Diffusion Probabilistic Models for 3D Medical Image Generation},
publisher = {arXiv},
year = {2022},
}
This code is heavily build on the following repositories:
(1) https://github.com/SongweiGe/TATS
(2) https://github.com/lucidrains/denoising-diffusion-pytorch