Inter-Intra Hypergraph Computation for Survival Prediction on Whole Slide Images

Xiangmin Han, Huijian Zhou, Zhiqiang Tian, Shaoyi Du, Yue Gao^*

IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), 47(7): 6006--6021, 2025

Click the link to access the paper.

Introduction

In this repository, we provide the training code for Intra-Hypergraph and Inter-Hypergraph models, along with various methods for hypergraph structure modeling. The dataset includes a sample list from publicly available datasets, which can be downloaded directly from TCGA.

(a) Multi-level medical information and corresponding correlations. Each subject contains multi-level medical information, such as intra-correlation at the cell and tumor level, and the inter-correlation at the group level. (b) Existing graph-based and MIL-based WSI analysis methods. (c) Different hypergraph modeling methods for WSI, the boundary-wise topological hypergraph models the boundary of WSI as hyperedges, the Spatial-wise Topological hypergraph considers the location interaction of patches as hyperedges, and the global-feature semantic hypergraph computes the feature distance to model the hyperedges.

The pipeline of inter-intra hypergraph computation framework.

Training Data Structure

• DIR: config
  • xx.yaml (your train/test config file)
• DIR: get_feature
  • sampled_vis (sampled patches, only for visualization)
  • patch_ft (deep features extracted via CNN models)
  • patch_coor (coordinates of the sampled patches, only for visualization)

Training

1. Feature Extraction

This script will generate three types of files: sampled_vis, patch_ft, and patch_coor.

WSI_sample_patch.py

2. Training Intra-HGNN

You can train the Intra-HGNN model to obtain intra-embeddings and intra-risk.

Note that this module can be used independently.

python train_stage1_intra.py  

3. Training Inter-HGNN

You can train the Inter-HGNN model to fuse intra- and inter-risks for the final result.

Note that if you have defined the feature vectors of inter-vertices in the inter-hypergraph, you can train this module without the first stage.

python train_stage2_inter.py

Citation

If you find our work useful in your research, please consider citing:

@article{han_2025_inter,
  title = {Inter-intra hypergraph computation for survival prediction on whole slide images},
  author = {Han, Xiangmin and Zhou, Huijian and Tian, Zhiqiang and Du, Shaoyi and Gao, Yue},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2025},
  month = jul,
  volume = {47},
  number = {7},
  pages = {6006--6021},
  publisher = {IEEE},
  doi = {10.1109/TPAMI.2025.3557391},
}

Contact

IIHGC is maintained by iMoon-Lab, Tsinghua University. If you have any questions, please feel free to contact us via email: Xiangmin Han.