This is an official implementation of AAAI-2025 paper "Dynamic Entity-Masked Graph Diffusion Model for Histopathology Image Representation Learning". We introduce H-MGDM novel self-supervised Histopathology image representation learning method through the Dynamic Entity-Masked Graph Diffusion Model. Specifically, we propose to use complementary subgraphs as latent diffusion conditions and self-supervised targets respectively.
✅ Networks
✅ Training
✅ Downstream tasks
The codebase is tested on Python 3.9. For additional python libraries, please install by:
pip install requirements.txt
Graph construction is following by Histocartography using tissue graph generation. We show the processed graph example in pyg format in ./data.
For VAE training, we refer to the implementation of AutoencoderKL in LDM.
For H-MGDM training:
bash train.sh
For classification downstream tasks:
bash tuning.sh
For survival tasks, we following CLAM for WSI processing, and training scripts:
cd survival
bash survival.sh
@inproceedings{zhuang2025dynamic,
title={Dynamic Entity-Masked Graph Diffusion Model for Histopathology Image Representation Learning},
author={Zhuang, Zhenfeng and Cen, Min and Li, Yanfeng and Zhou, Fangyu and Yu, Lequan and Magnier, Baptiste and Wang, Liansheng},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={39},
number={10},
pages={11058--11066},
year={2025}
}