Meet Skin Friend, the app which will save thousands of lives. At Skin Friend, we aim to democratize and accelerate healthcare, by providing free cancer detection resources, in the palm of your hand. One in every three cancers is skin cancer. In the United States, the country with the highest number of cases (by a long shot), one in every five people gets skin cancer. That is 9,500 cases daily! This year, 196,060 Americans will be diagnosed with skin cancer, out of whom 6,850 will die. Throughout the entire world, 2-3 million people will be diagnosed with skin cancer and many will die because they aren’t diagnosed and don’t get help. But it doesn't have to be this way. So far, we have only looked at people who are able to see doctors. 46 million Americans lack health care resources and are unable to access frequent medical checkups. That is 46 million people and counting under the radar! All diagnosed cases come from those who are able to meet medical professionals regularly, but what about people who do not have these resources? They won’t find out they have a disease and they won’t get the treatment they need as a result. That’s where we come in. Skin Friend builds bridges towards democratizing healthcare, meaning that people of all backgrounds, incomes, and situations, will be able to diagnose critical diseases easily, affordably, and rapidly. Considering the increasing severity of skin cancer, it is more important than ever to prevent deaths from this otherwise trivial condition. If skin cancer is diagnosed early, there is a 99% survival rate, and Skin Friend provides exactly that: rapid and convenient diagnosis. No longer do you have to wonder whether you should see a doctor or worry about your lack of healthcare for the random dark spot on your skin: simply scan your problem spot through our app, and receive a diagnosis in seconds. For free. Skin Friend allows the user to diagnose their skin problems early on, so if it is skin cancer, they can still have a high survival rate and get the help they need ASAP.
Narrated "pitch" audio further explaining the app: https://drive.google.com/file/d/19Acb5ofDtEYd4W0IXOnQDLC-nKh5fjd-/view?usp=sharing
The Skin Friends team was inspired by personal experiences, whether it was sheer laziness to visit the doctors office when something minor happened (such as irregularities in the skin), and the shock that came with learning how many people are unable to regularly visit doctors. Our team acknowledges the privilege we have by being able to afford healthcare, so we decided to use this privilege to help others who lack basic necessities, especially healthcare. Health care is a basic human need according to UN SDGs and we are glad to make an impact in it.
What it does
Skin Friend only requires a working phone with a camera (something which almost everyone has, 45% of world population including elderly and young kids). The introductory page features an intuitive step-by-step guide on the steps that need to be taken to use our app. Users are then given three options (in flowchart form for the suggested order). The first and most important feature is the diagnosis feature. Users are prompted to take a picture of their problematic skin area in the app. After taking the picture, users are quickly diagnosed with their skin-related issue. The other pages include an “information” page and a “contact” page, meant to provide easy access to doctors and hospitals if needed, and great customer service.
How it was built
After extensive coding and testing, we used a Convolutional Neural Network in TensorFlow to create a custom model with high accuracy for diagnosing skin cancer. Unfortunately, our ML engineers lacked a GPU, so we used Google Colab, which allowed us to test and edit much faster. We used pandas and numpy for reading and storing images. We also used scikit-learn for a few functions such as train test splitting. We used transfer learning with Xception as the base and added model layers on top. Our preprocessing and model training was all done in the cloud and we collaborated through Google Meet to discuss things live. This hackathon was also the first time our frontend developer programmed a full web app, which made SET hacks a front-to-end learning experience! We also used Canva to design our UI and screen recording software to record demo videos.
Challenges we ran into
There were multiple challenges, such as the model's long training time (taking hours at best) which reduced how much we could tweak and fiddle with parameters to find the best accuracy. We worked around them by doing most of the guesswork at the beginning. Other challenges included the preprocessing, which took a lot of googling to fully materialize and it took a while due to our limited processing power at home. The 12 hour time zone difference between our members was another huge challenge. We are proud that we managed to overcome our challenges.
Accomplishments that we're proud of
We're proud that throughout the numerous obstacles, we persevered and finally ended up with a product we can proudly call our own, and something that is incredibly socially beneficial. We are also proud of the knowledge we acquired along the way, which makes us better suited for future projects. We also went through some really mentally taxing periods, especially while pulling all-nighters due to time difference, and we're glad we didn't give up.
Also, the accuracy is very high for something done in such a short time and we are proud that we managed to bring it up to that level and we are sure we can improve it even further with more time.
What we learned
Throughout the whole process, there were countless learning moments, ranging from errors during frontend app development to learning the best model architecture and reading up on intense deep learning theory. Overall we now have a much stronger understanding of key concepts than we did when we started.
What's next for SkinFriend
As a team, we strive to code our way to a better healthcare system and a better future. In fact, our team plans to expand our app to not only skin cancer detection but also other external diseases such as eye conditions! We will use our experience gained from SetHacks to continue further. We hope to distribute our web app and get it used by hundreds of people in the near future.
Thank you so much to the organizers and judges for volunteering and making this event possible!