Awesome

This repository provides the Pytorch implementations for "Histopathology Whole Slide Image Analysis with Heterogeneous Graph Representation Learning"

Paper can be found here and video walkthrough is here.

Download the WSIs

The WSIs can be found in the TCGA project:

https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga

Patch Extraction

To extract the patches from the downloaded WSIs, users need to first modify the parameters in get_patches.py (including the WSI paths) and extract the patches by running the following command:

python get_patches.py

Graph Construction

After the patch extraction is finished, users can obtain homogeneous and heterogeneous graphs by first edit the configurations in ./configs/GraphConstruction, and specify the correct yaml configuration file in get_graph.py, then run the following command

python get_graph.py

Training HEAT Model

The configurations yaml files for each benchmarking dataset is grouped in respective subfolders. Users may first modify the respective config files for hyper-parameter settings, and update the path to training config in main.py.

python main.py

The training pipeline is mainly written in ./trainer/train_gnn.py. Evaluation is performed after every epoch on validation sets and testing sets. The codes can be find in ./evaluator/eval_homo_graph.py.

Load checkpoints

The trained checkpoints will be saved in ./chekpoints, including the GNN model. Users can perform evaluation using the saved weights inside the checkpoint.