Inspiration
Every year 5.4 million people are diagnosed with skin cancer. Of those, most deaths are caused by late diagnosis. We believe that if skin cancer screening was more widely available and free, the disease would be more treatable.
What it does
This is a web app that takes in a user-uploaded image of melanoma and returns a value representing the likelihood of the melanoma being malignant based on color.
How we built it
The algorithm utilizes opencv to analyze the color content of images for specific colors that we wanted to select for. We downloaded a subset of images from the ISIC database, but we didn’t end up having enough time to preprocess the data and then train the model on these. However, we envision that this can be easily achieved for future development.
Challenges we ran into
There were a lot of errors that we ran into when trying to use new modules or trying to code things we never did before. Some included loading in the ISIC database, splitting and writing data to csv files, and reading an image from a path.
Accomplishments that we're proud of
We’re proud of collaborating efficiently and creating a tangible, although primitive, product for social food. Also, this was all of our team members’ first hackathon, so we’re definitely proud of just getting through this weekend!
What we learned
Web-based IDEs like repl.it are susceptible to timing out. In terms of coding, we learned a lot about different python modules like pandas, opencv, numpy, and os. In general, we all think we’ve been challenged and we’ve improved as engineers solving a problem!
What's next for Melanosite
The UI can be improved and the code decluttered. The model should be trained on a more diverse set of images, including images of darker skin. The functionality can be increased. We also want the model to take into account border and asymmetry as well as patient conditions. Right now, the site is very primitive, and there’s a long way to go in terms of improvements!
Log in or sign up for Devpost to join the conversation.