Graph neural networks for PDAC vs CP in histology

Paper on Journal of Pathology Informatics

Conda environment

Install anaconda/miniconda
Required packages

  $ conda env create --name pyg --file env.yml
  $ conda activate pyg

Prepare patches

Download the dataset (e.g. TMA dataset) from Google Drive
Use the deepzoom_tiler.py from this repository to prepare patches. Otherwise, the patches (normalized) can be downloaded from Google Drive.

Stain normalization in batches

Example usage is shown in normalization.ipynb.
Note that the normalization algorithm is from Paper and the code is from GitHub.

Computing features using pretrained ResNet

See notebook_distributed.ipynb.

Training and cross-validation

See notebook_distributed.ipynb.

Generating region proposals from unannotated slides

Prepare patches following above steps for both the slides and annotated regions.
See notebook_proposals.ipynb