We were inspired by the high rates of death and severe illness based on skin problems.

What it does

Skin Skan is an AI-powered skin lesion diagnosis that can accurately identify & classify skin lesions based on dermoscopic images. It uses a deep learning algorithm to analyze images and provide diagnosis w/ high accuracy. It also has a very elegant, user-friendly interface. It can recognize 8 different types of cancers and skin lesions and categorize 1st, 2nd, 3rd degree burns. It gives advice for all skin diagnoses.

How we built it

First, we acquired skin cancer and burn datasets from Kaggle. Then, in PyTorch, we used transfer learning with the ResNet-18 Convolutional Neural Network. We finetuned our model based on the datasets. Then, we created a Flask framework, combining Python and HTML. On the backend, we called our model to develop predictions based on inputted images. On the frontend, we employed Bootstrap and JavaScript to create a simple yet elegant interface for users.

Challenges we ran into

Challenges we ran into were inputting images based on forms with flask and properly handling them to be processed by our machine learning model. We were also challenged by the high amount of JavaScript required throughout the app.

Accomplishments that we're proud of

We are very proud of the frontend of our website, as it is the most complex we have created so far. We are also proud of the high accuracy rate of the diagnoses by the machine learning model.

What we learned

We learned a better understanding of JavaScript functionalities as well as the Pillow library to handle our images

What's next for Skin Skan

We plan to add more types of skin problems, such as classifying the different types of acne (yay teenage life).

+ 6 more
Share this project: