To use the code in this repository, you will need to set up two different environments: one for WSI preprocessing and another for model training.
For WSI preprocessing, use the wsi_pipeline.yml environment.
For model training and evaluation, use the dermato.yml environment.
As we are not sharing the whole slide images required to reproduce the results, please note that you cannot simply run the code provided in this repository to replicate the exact results. Certain files contain hard-coded paths that are specific to our setup. However, we have included essential components of the pipeline that can serve as a reference or starting point for your own research.
To perform WSI segmentation, refer to the segmentation/TissueExtractionToolkit_demo.ipynb notebook.
The code for stain-specific labeling can be found in dermato/stain_labeling.py. After labeling, the dataset is created in CSV format using dermato/train.py. We also used dermato/labelmap.py to create the label map to visually check the labels.
To train the model, use dermato/train.py along with the dermato/train_config.yaml file. Please note that there are several hard-coded paths in the code for loading CSV files and saving weights.
To evaluate the model's performance, including metrics such as AUROC and AUPRC, use dermato/eval.py with the dermato/eval_config.yaml file. Similar to training, there are hard-coded paths in the code.
To generate prediction confidence heatmaps, use dermato/heatmap.py with the dermato/heatmap_config.yaml file. The script utilizes a sliding window approach and PyTorch Lightning for scalability.
For saliency map generation, refer to the saliency/generate_saliency_visualizations.ipynb notebook.