SpotCheck - A submission to the Responsible AI Hackathon

Team: SpotCheck

Team Members:

• Adam
• Madhav
• Luisa
• Rajiv

Problem Statement

To rectify the discrepancy in AI models underdiagnosing darker-skinned patients with melanoma due to being trained to better recognize cancer on light skin, our goal was to create a model trained on an image dataset which was neutral to skin tone and color to increase diagnostic accuracy.

How to run the code

Training

1. Download both datasets from:
   • https://isic-challenge-data.s3.amazonaws.com/2020/ISIC_2020_Training_JPEG.zip
   • https://www.kaggle.com/datasets/kmader/skin-cancer-mnist-ham10000
2. Unzip both datasets into AIModel/images in the AIModel folder. Extract the images from the ISIC_2020_Training_JPEG.zip into 'AIModel/images' and the images from the HAM10000_images_part_1 and the HAM10000_images_part_2 folders into 'AIModel/images'.
3. Create a new virtual environment called SpotCheck with python3 -m venv SpotCheck
4. Install the requirements with pip install -r requirements.txt
5. cd into the AIModel folder
6. Run the training script with train.sh
7. The model will be saved in the AIModel folder as skin_cancer_diagnosis_model.h5

Running the web app

1. Download https://drive.google.com/drive/folders/1eI2kVQ4_SLDMZpcSdTkHETtxRu4V329X?usp=sharing
2. Create a new virtual environment called SpotCheck with python3 -m venv SpotCheck and install the requirements with pip install -r requirements.txt
3. cd into the AIBackend folder
4. Run the web app with python backend.py
5. Open the web app at localhost:5000

Note: The app will crash if the computer does not have a GPU with enough VRAM to run the model. If this is the case, please run the app on a computer with a GPU with at least 4GB of VRAM.