[MICCAI 2024] ORCGT: Ollivier-Ricci Curvature-based Graph Model for Lung STAS Prediction

Min Cen*, Zheng Wang*, Zhenfeng Zhuang, Hong Zhang, Dan Su, Zhen Bao, Weiwei Wei, Baptiste Magnier, Lequan Yu, Liansheng Wang†

Installation

• Clone the repo:

- git clone https://github.com/zhengwang9/STAS.git && cd STAS

• Create a conda environment and activate it:

conda create -n env python=3.9
conda activate env
pip install -r requirements.txt

Image Preprocession and Feature Extraction

• We used CLAM to split the slides and extract featurers of patches by Ctranspath

Major Tumor Margin Extraction

• We employ a pretrained HoVerNet to classify tumor patches based on their cell count.
• Then, we construct the tumor density map by tumor_density.py. After that, we derive the mask for the major tumor region using UNet by tumor density map.
• Finally we select patches in ring of major tumor margin by choose_ring.py.

# classify tumor patches
python ./hv_res_post-process/choose_tumor_patch.py
# construct tumor density map
python ./hv_res_post-process/tumor_density.py
# select Ring patches
python ./hv_res_post-process/choose_ring.py

Graph Construction

We construct the graph with curvature by two steps:

• extract feature of patches (nodes) in major tumor margin
• construct graph.

# extract feats in ring.
python ./graph_construction/extract_huandai_feats.py
# construct graph
python ./graph_construction/ToPyG_curva.py

Training

# train the model
cd train
python train_curapooling.py

Citation