Gleason XAI:

Paper: arxiv

System Requirements

Software dependencies can be found in the conda-env.yml.

The Software was created and tested on a Linux system with a GPU.

Use

Set up:

1. Create conda environment using the conda_env.yml with

   conda env create -f path/to/conda_env.yml
   conda activate GleasonXAI
   

2. Set environment variables:
   • DATASET_LOCATION: location of the images and the dataframe containing the labels
   • EXPERIMENT_LOCATION: location for the log files during experiments
3. Extract and add TissueMicroarray.com Data at DATASET_LOCATION / GleasonXAI / TMA / original
4. Add label data (final_filtered_explanations_df.csv) and hierarchy mapping (label_remapping.json) at DATASET_LOCATION / GleasonXAI
5. Extract and add directory containing model weigths at DATASET_LOCATION / GleasonXAI (with directory structure as is)
6. With download_data.py, download and add the other datasets
7. (if failed due to missing step 4) create the MicronCalibrated data using create_downscaled_dataset.py

ca. 15min (depending on download speed)

---

Use for single image prediction:

1. adjust paths in single_prediction.ipynb
2. run the notebook

ca. 2min

---

Use for generating paper visualization:

1. run test.py to create predictions on the test set (at least for models in GleasonFinal2/label_level1/SoftDiceBalanced-{i}/version_0/).
2. run evaluate_paper_results.ipynb to create the visualizations and figures of the paper.