Improved Diagnosis of Non-Melanoma Skin Cancer in Resource-Limited Settings

https://doi.org/10.1158/1055-9965.EPI-25-0132

Spencer Ellis*, Steven Song*, Derek Reiman*, Xuan Hui, Renyu Zhang, Mohammad H. Shahriar, Maria Argos, John A. Baron, Mohammed Kamal, Christopher R. Shea, Robert L. Grossman†, Aly A. Khan†, and Habibul Ahsan†

* Equal Contribution

† Equal Contribution

Setup

In order to utilize most of the code in this repo, you must have been approved for access to the UNI, Prov-GigaPath, PRISM, and Virchow V1 models on Huggingface:

• UNI: https://huggingface.co/MahmoodLab/UNI
• Prov-GigaPath: https://huggingface.co/prov-gigapath/prov-gigapath
• PRISM: https://huggingface.co/paige-ai/Prism
• Virchow V1: https://huggingface.co/paige-ai/Virchow

Setting Up Your Local Environment

1. Clone the repo:

   git clone https://github.com/sjne09/akdslab-skin-cancer.git
   

2. Set up a conda environment and install all dependencies:

   conda env create -f environment.yaml
   conda activate nmsc
   

3. Install OpenSlide

4. Clone the Prov-GigaPath repo and install it as a package:

   git clone https://github.com/prov-gigapath/prov-gigapath
   cd prov-gigapath
   pip install -e .
   

5. Update .env with data and output directory paths

Running

To rerun our experiments, see the scripts in scripts.pipeline.