Novice Track

Inspiration

Skin disease is a major healthcare challenge in the U.S. Rates of melanoma have been rising drastically, while healthrate coverage is often barely affordable. After seeing the statistics of how few Americans choose to visit a dermatologist specialist to get diagnosed with a skin disease, we wonder how we can expand access to affordable diagnosis.

What it does

derm.ai is an app that takes a user submitted image and uses an AI model to classify that image as skin disease. Currently, the model identifies Acne, Melanoma, Warts, and HPV. At low confidence ratings, it not return an identification.

How we built it

We use the Dermnet dataset from kaggle to acquire images of skin diseases. Using Teachable Machine, we create a neural network in Tensorflow that accurately identified the diseases. We integrated the tensorflow model into a Flutter app using the tflite plugin. We would like to thank Kshitjj Ra, whose work on a Waste Classification App helped us tremendously when we were trying to figure out how to integrate the model into Flutter. https://github.com/kshitij-ra/Waste-Classification-App

Challenges we ran into

Creating an adequate model was a challenge, but with some fine tuning we were able to get it to output >80% accuracy.

Accomplishments that we're proud of

That we were able to get this done with a minimum of sleep.

What we learned

The power of AI image recognition is largely untapped within the healthcare space, and we should strive to use AI as a tool to improve diagnostic outcomes.

What's next for derm.ai

Including more skin diseases like Poison Ivy or Vascular Tumors would be a natural step. Also, the app should navigate to a website like WebMd to provide more information about the diseases.