MICCAI 2023: FSDiffReg: Feature-wise and Score-wise Diffusion-guided Unsupervised Deformable Image Registration for Cardiac Images

This repository is the official implementation of "FSDiffReg: Feature-wise and Score-wise Diffusion-guided Unsupervised Deformable Image Registration for Cardiac Images".

Requirements

Please use command

pip install -r requirements.txt

to install the environment. We used PyTorch 1.12.0, Python 3.8.10 for training.

Data

• We used 3D cardiac MR images for training: ACDC dataset

Training

To run training process:

python train.py -c config/train_3D.json

Test

To run testing process:

python3 test.py -c config/test_3D.json -w [YOUR TRAINED WEIGHTS]

Trained model can be found here

Acknowledgement

We would like to thank the great work of the following open-source project: DiffuseMorph.

Citation