LesionLearning

Made by Hussein A.
Freshman @ WNNS

Tested on macOS. Might have issues on other operating systems.

Labels (dx)

• akiec — actinic keratoses / Bowen’s disease
• bcc — basal cell carcinoma
• bkl — benign keratosis-like lesions
• df — dermatofibroma
• mel — melanoma
• nv — melanocytic nevi
• vasc — vascular lesions

Dependencies

python3 -m pip install torch torchvision pandas pillow scikit-learn tqdm

Dataset

This project uses the HAM10000 skin lesion dataset.

You must add the following to the project directory:

• HAM10000_images_part_1/
• HAM10000_images_part_2/
• HAM10000_metadata.csv

These files and datasets were not created by me.

The HAM10000 dataset can be downloaded from the
Harvard Dataverse – HAM10000

Please take the results with a huge grain of salt. You can edit the epoch however you want. Default is 15. Longer epoch --> More accurate/resource heavy Shorter epoch --> Less accurate/not as resource heavy

The json file created can be plugged into a website that displays all the pictures and results.