Cortex-Grounded Diffusion Models for Brain Image Generation

Official Pytorch Implementation of Paper - Cortex-Grounded Diffusion Models for Brain Image Generation

Installation

1. Create environment: conda env create -n c2v --file c2v_env.yml
2. Activate environment: conda activate c2v

Data preparation

For dataset the path should be formatted as:

dataset_path/{format}Tr  # training data
dataset_path/{format}Val  # validation data
dataset_path/{format}Ts  # test data

for different formats of input.

After that, the dataset configuration should be specified in config file as:

dataset_name: 'dataset_name'
dataset_config:
  img_folder: 'dataset_path/mri'
  shape_folder: 'dataset_path/shape'
  condition_folder: 'dataset_path/other_condition'

Usage

After specifying the config file in configs/c2v.yaml, simply start training/evaluation by:

sh train_c2v.sh

for training, and

sh test_c2v.sh

for evaluation.

Acknowledgement

Our code is implemented based on the BBDM, thanks!

Citation

If you find this repository useful, please consider giving a star 🌟 and citing the paper:

@inproceedings{bongratz20253d,
  title={3D Shape-to-Image Brownian Bridge Diffusion for Brain MRI Synthesis from Cortical Surfaces},
  author={Bongratz, Fabian and Li, Yitong and Elbaroudy, Sama and Wachinger, Christian},
  booktitle={International Conference on Information Processing in Medical Imaging},
  pages={187--202},
  year={2025},
  organization={Springer}
}