CoCycleReg

This is the official implementation of the paper CoCycleReg: Collaborative Cycle-consistency Method for Multi-modal Medical Image Registration.

Some code of this repository is borrowed from Voxelmorph and NeMAR.

Getting started

1 Requirements

If you are using conda, you can continue with

  conda env create -f environment.yaml

2 Data preparation

• Preparing the data for training like

  ├── /the/path/of/training/data/

        ├── img1_modality1.npy

        ├── img1_modality2.npy

        ├── img2_modality1.npy

        ├── img2_modality2.npy

          ......

• Preparing the data for validating like

  ├── /the/path/of/validating/data/

        ├── img1_modality1.npy

        ├── img1_modality2.npy

        ├── img1_modality1_seg.npy

        ├── img1_modality2_seg.npy

        ├── img2_modality1.npy

        ├── img2_modality2.npy

        ├── img2_modality1_seg.npy

        ├── img2_modality2_seg.npy

          ......

3 Training

• Set the data path, GPU ID, batch size and other parameters in config.yaml.

• Start training by running

  python train.py
  

• Tensorboard is supported, the log files are in /the/path/of/output/log/.

• The weights are saved in /the/path/of/output/pth/.

Feedback

If you have any problem, please feel free to report it in the issue, thank you!