Unsupervised Discovery of 3D Hierarchical Structure with Generative Diffusion Features

This repo includes the code for Unsupervised Discovery of 3D Hierarchical Structure with Generative Diffusion Features.

We show our method on the Synthetic 3D Dataset from Hsu et al (regular, irregular) and also on BraTS'19

Pipeline

1. Diffusion model training

   python diffusionTrain.py --dataset_path <folder with training images> -output_path <folder to save diffusion models> 
   

2. Unsupervised segmentation/discovery model training

   python segTrain.py --dataset_path <folder with training images> --d_ckpt <folder with saved diffusion models> --output_path <folder to save segmentation models>
   

Citation

If you find this project useful, please cite:

@inproceedings{tursynbek2023unsupervised,
  title={Unsupervised Discovery of 3D Hierarchical Structure with Generative Diffusion Features},
  author={Tursynbek, Nurislam and Niethammer, Marc},
  booktitle={Medical Image Computing and Computer Assisted Intervention – MICCAI 2023},
  year={2023},
  publisher="Springer Nature Switzerland",
  address="Cham",
  pages="320--330",
  isbn="978-3-031-43907-0"
}