BenignAI

Inspiration

Skin cancer is one of the most common and deadly forms of cancer, yet early detection can significantly improve survival rates.
However, many people either ignore suspicious blemishes or lack access to dermatologists for timely diagnosis.
We wanted to leverage AI to create an accessible tool that helps individuals assess potential risks and seek professional care as soon as possible.

BenignAI is an AI-powered image analysis tool that scans skin blemishes, rashes, and moles to determine the likelihood of malignancy.
Users upload an image, and the model provides a probability score along with recommendations to consult a specialist.
In addition, users can converse with Clara: an AI model emulated through an ELI5 model to act as a radiologist and assist users in understanding this prediction.

We trained a machine learning, random forest model on a dataset of labeled skin images of benign and malignant tumors, utilizing NumPy to turn visual images into numerical data.
Integrated the model into a user-friendly web app using Flask and React for seamless accessibility.
Utilized GeminiAI and GoogleMapsAPI to better help users understand the model and seek further help if necessary.
Authenticated login implemented through Auth0.
Deployed through AWS and Vercel.

Finding a diverse and well-labeled dataset that covers various skin tones and conditions.
Preventing overfitting due to limited training data.
Optimizing model performance to ensure real-time predictions without compromising accuracy.
Designing a responsible solution that emphasizes medical consultation rather than self-diagnosis.

- Successfully training a high-accuracy model that provides meaningful predictions.
Building an intuitive and accessible web application through React and Flask.
Overcoming technical hurdles in dataset preprocessing and model deployment through AWS.