SkinSight

Inspiration

Inspiration for this project came when I saw the scary statistic that over 2 people die per hour due to skin cancer. I was empowered to solve this issue as a way to address the UN Sustainable Development Goal 3 of "ensure healthy lives and promote well-being for all at all ages".

What it does

SkinSight detects 9 different types of skin cancer plus a being able to detect benign skin lesions. It is also able to generate a medical report following detection.

How we built it

I used Tensorflow with MobileNetV3Large to build the machine learning model. For the frontend of the website, I used HTML and CSS. For the backend of the website, I used flask.

Challenges we ran into

Initially I wanted to use Resnet50 as the machine learning model however, training was too slow. I then pivoted to using InceptionV3 but I ran into the issue of it not being able to be saved properly. I then tried MobileNetV2 but accuracy was very poor(<70%). Finally, I settled on MobileNetV3Large as the model of choice.

Accomplishments that we're proud of

I am proud that I was able to build a machine learning model that is able to detect 9 different types of skin cancer with over 80% accuracy.

What we learned

I learned how to effectively plan and design a machine learning project. I also learned how to create an effective pitch presentation.

What's next for SkinSight

We want to be able to scale this project to integrate into a phone app where the user can simply take a picture of a skin lesion and the app will automatically classify it. Currently it is a website and the user has to upload the image into the website for it to detect it. Additionally, we hope to improve the accuracy as the model can give inaccurate results for some test cases.