SkinIQ

Diagnosing skin cancer from images taken/stored on the phone, and then recommending possible treatment

Comment 3

Inspiration

Mr. Chang, a teacher we've had in the past, was diagnosed with skin cancer. Fortunately, he was diagnosed in time and survived radiation therapy. However, the majority of people are not as lucky as Mr. Chang. The survival rate of skin cancer at a late stage is under 20%. We want to help people get help early like Mr. Chang to receive an early diagnosis, which led us to create SkinIQ.

What it does

SkinIQ is an android app that can detect if you have skin cancer. You can use your computer's camera or upload a photo of your skin, and the app can classify if you have skin cancer or not. If you have skin cancer, then the classifier will also tell you the type of skin cancer you have. Lastly, we inform you of the nearest medical facilities from your current location if you have any immediate medical concerns.

How we built it

We used Android studio to make the frontend of the app. We used Jupyter Notebook, Keras, and Tensorflow to build a Convolutional Neural Network with roughly 94% accuracy after training it for 4 hours with 20,000 images. We used the Google Maps API to locate the nearest treatment center if necessary.

Challenges we ran into

Converting the model from Keras to Tensorflow was a difficult task. In addition, training the data took FOREVER and required us to tweak the Convolutional Neural Network code layers and set up the Tensorflow GPU library.

Accomplishments that we're proud of

We finished our Convolutional Neural Network, and able to achieve 94% accuracy. We also have a website and a presentation showcasing our project.

What we learned

Machine Learning, Keras, Tensorflow, and implementing all of that in an Android app.

What's next for SkinIQ

Higher accuracy and more types of skin diseases that can be classified using our CNN.

Built With

Share this project:

Updates