[CVPR 2025 (Oral)] TopoCellGen: Generating Histopathology Cell Topology with a Diffusion Model

Meilong Xu¹, Saumya Gupta¹, Xiaoling Hu², Chen Li¹, Shahira Abousamra³,
Dimitris Samaras¹, Prateek Prasanna¹, Chao Chen¹

¹Stony Brook University, NY, USA
²Harvard Medical School and Massachusetts General Hospital, MA, USA
³Department of Biomedical Data Science, Stanford University, CA, USA

Environment

Training and evaluation environment: Python 3.8.19, Pytorch 1.11.0, CUDA 11.3. Please run the following command to install required packages.

conda create -n topocellgen python=3.8
conda activate topocellgen

git clone https://github.com/Melon-Xu/TopoCellGen.git
cd /path/to/TopoCellGen
pip install -r requirements.txt

Generating Cell Layouts

Use the following code to generate synthetic cell layouts for the BRCA-M2C testset:

python3 generate_layout_brca.py

After running it, the results are structured as follows:

/path/to/results/%Y-%M-%D/%H-%M-%S
  img/
  npy/
  cell_counts.json

Evaluation

For the evaluation, you need to firstly set the paths to the reference and synthetic layouts, and then run:

cd /path/to/TopoCellGen/evaluate
python3 evaluate_BRCA.py

For the TopoFD evaluation, you can run:

python3 eval_TopoFD.py

Qualitative Results

Topological Fréchet Distance

Download

• TopoCellGen pre-trained weights: Google Drive
• BRCA-M2C dataset: You can assess the data from official website.
• Lizard dataset: You can assess the data from official website

Citation

If you found this work useful, please consider citing our paper:

@inproceedings{xu2024topocellgen,
  title={TopoCellGen: Generating Histopathology Cell Topology with a Diffusion Model},
  author={Xu, Meilong and Gupta, Saumya and Hu, Xiaoling and Li, Chen and Abousamra, Shahira and Samaras, Dimitris and Prasanna, Prateek and Chen, Chao},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2025}
}