Skindex - Skin Condition Analyzer

Inspiration

Studies have found that 80% of adults experience skin concerns, with one-third citing an inability to afford dermatological care. For many, visiting a dermatologist feels unnecessary for the cost or inconvenience. But what looks harmless isn’t always safe, and knowing when to seek help can make all the difference; So we wanted to revolutionize the age-old process of googling your symptoms and seeing unclear results on Mayo Clinic by turning it into a simple photo upload that assesses your skin. That’s why we’re building an accessible, user-friendly app that empowers people to take control of their skin health. With just a photo, users receive an instant analysis and guidance on whether their condition is likely benign or worth medical follow-up. It’s simple, affordable, and reliable. Because everyone deserves peace of mind when it comes to their skin.

What it does

With 85% accuracy, Skindex is able to look at pictures of your skin and identify what condition you have using trained machine learning models.
Allows for integrated frontend-backend communication through REST API’s. Users can seamlessly upload a picture of their skin and receive an analysis in seconds.
Stores ML models in the application to prevent having to retrain data yourself.

How we built it

We used Pytorch to train multiple different neural networks and were able to choose the best performing model.
We used a Flask backend, with a Vite frontend, using REST API’s in order to communicate between the two.
We used React in conjunction with HTML/CSS for frontend interactivity.
ONNX to store our neural networks that we pretrained for our application.

Challenges we ran into

We struggled with finding a valid data set to train our data with as we didn’t have access to modern medical datasets.
We ended up training with various datasets and models to see which gave the most accurate results.
Training models took a significant amount of time, so we had to be intentional with every tweak to the models.

Accomplishments that we're proud of

85% Accuracy on ML-trained Skin Model
Training and developing our first ML models
We were able to develop a full-stack application that allowed for user file upload and interaction.
Successful backend integration

What we learned

How to work and collaborate as a team.
How to develop Pytorch models
How to train a neural network for image classification.
How to store machine learning models using ONNX.
How to create modern and interactive UIs

What's next for Skindex

Working with medical institutions in order to gain access to clinincal databases for better training
Integrating AI into diagnoses
Implementing Computer Vision to identify areas of interest so users can follow along with our models