- Barış İnandıoğlu @baris-inandi
- Yiğit Kerem Oktay @yigitkeremoktay
- Alp Niksarlı @alpnix
SkinScan is a service aiming to provide affordable solutions to people with dermatological issues. Skin diseases make up a substantial portion of the problems global health is currently facing. To be more precise, according to aad.org, in the US, more than 16 million people are suffering with Rosacea and over 7.5 million people deal with Psoriasis. Hearing stories about extended waits in hospitals and hardly ever getting certain results is sadly not all that uncommon. Not to mention the amount of money spent on hospitals. As team line0, we all kept this specific problem in mind while designing and prototyping our product, SkinScan. Our product offers quick and early diagnosis of dermatological conditions. In order to use our product, the users will be able to download our application to their device. Once they open the app, they will be prompted to take a picture of the skin area they are having problems with. Afterwards, our AI technology will analyze the sample it received and try to diagnose the condition as accurately as possible. If needed, the app will prompt the user to answer follow-through questions such as "Do you feel burning in the area?" or "Is the area itchy?" to get a better understanding of the case. Once the results are received, they will be displayed on the screen in a clear and concise manner. The users will have easy access to pages where they can find information such as symptoms, complications, and treatments of the diagnosed condition. The entire process, from a user downloading the app to getting their results, is to take just minutes. Notably, we are trying to offer the fastest diagnosis possible without needing to go to a hospital. In addition, through collaboration with experts in dermatology, the AI model can be trained, and improved over time to the point that it can become a viable way of formal and accurate diagnosis that can actually result with treatment for the patient. As some of our future plans for our product, we are considering making the API public, opening it up for the use of third parties with a subscription model. In that case, the images/data sent by third parties can be used as source material in future AI training, similar to what is being done with captchas today.
We have a mission to make dermatology easier. We know that health systems across the world can be very painful and expensive. From unacceptable waiting times and expensive doctor visits and/or prescriptions, healthcare has a lot of aspects that can be improved. With the rise of AI technology in the past years, we have the power to enhance every aspect of our lives, now more than ever. That is why SkinScan is a better alternative to the current system when it comes to healthcare. AI has the power to analyze and compare images so why not use it for diagnosis of dermatological conditions? Over time, our AI has the potential to become a way to cheaply and accurately diagnose conditions and help the patients get help immediately, getting rid of all the tedious and time-consuming parts when trying to get access to proper healthcare. Our service targets those who seek immediate help and early diagnosis as our product will provide much quicker results than one can get by going to hospital. In addition, our target audience will include the ones who are looking for cheaper solutions to dermatological issues since we will always be offering cheap, and affordable solutions for a better future in global health.
How will the service be beneficial to line0? The AI model that will power this app will initially be trained via thousands of samples of dermatological diseases from the internet. As the scans are completed and the app turns into profit, our team could hire dermatologists who could review the AI’s decisions after a while to be periodically contributed to the neural network as training data. This means our app will get more intelligent with time. As the app grows, we will be able to improve our model.
Our team has started preparing the app’s UI and UX design using the prototyping tool Figma using the UI principles loosely based on Google’s Material UI 3. This UI prototype contains the flow of the user from the start screen of the app to the submission screen and the follow-through questions. We have also gone ahead and scaffolded a simple backend, namely our API that can get these images submitted from the app which is currently only in the prototyping phase, and convert it into an OpenCV compatible format to be further used in conjunction with PyTorch/Tensorflow and then return a dummy result of a predefined disease with predefined questions.
// Example API output, subject to change
{
"results": {
"Psoriasis": {
"id": 0,
"out": 0.53, // 53% accuracy
"wiki": {
...
} // more info for insights page
},
"Rosacea": {
"id": 1,
"out": 0.16
}
},
"follow-through": {
"Do you feel burning in the area?": 0,
"Is the area itchy?": 1 // in favor of result with ID 1
}
}A simplified version of the data set that will be used to identify images has been generated using 200 images each for 2 diseases and a neutral case. Currently the images have been grabbed by google via automations and then hand-picked manually from images that aren’t confirmed to be the cases of these diseases. The production data will use more reliable samples of the diseases which have been previously confirmed by experts. In addition, the planning of the tech stack and the API methods and parameters as well as JSON response structure have been completed which we will be updating with minor changes and touching up as the app is built.
There are two major aspects in regard to the development of our service: the mobile app and the AI API. We chose Flutter (subject to change, replaced with ionic react) as our framework of choice for our mobile app. On the client side, after letting users sign in with SSO using the OAuth protocol, they will be able to send requests to our API through a HTTP POST request with the OAuth token being the authentication header. The data returned will be presented to the user through our app. Effectively, the frontend will let the users submit a picture of their skin. The second aspect is our Python/FastAPI API. After the server verifies the OAuth token through the identity provider, it will temporarily save the sent image and add it to an SQL database. A separate Python process will be dedicated to running the image classification. When a new image is added, this process will load the image, and then use OpenCV and TensorFlow/PyTorch (subject to change) to classify the image using our AI model. When the classification is complete, the API will return JSON with match percentages of images for use in the frontend. The API will also implement features to improve on the end user experience. For example, in cases where the percentage of accuracy is low or the gap between percentage of accuracies are too low, the API will return follow up-questions as well for the client-side to use in order to increase the accuracy of the result. In practice, the user will be able to take a photo of their sample, answer a few questions if needed, and see their results. In addition, they will be able to gain insight through the condition’s dedicated page with info such as symptoms, complications, and treatments.
- Project Drive link: https://drive.google.com/drive/folders/1bj3r-M8NyBCrgWyyBnRNpZGI_wCkUgCo
- GitHub repository: https://github.com/baris-inandi/skinscan
- Figma design link: https://www.figma.com/file/wOYpIffc3C8dTDzUTwUhiY/SkinScan
Most of our design choices were loosely inspired by Googles Material v3 design guidelines. The design of the app is both intended to look professional with a modern twist. The design puts legibility and usability first. User experience (UX) is a huge part of our design system. We are keeping the user interface (UI) as simple as possible for users to find each resource easy on the interface. Similarly, the interface we used for taking a photo allows our users to comfortably tap anywhere on the screen whenever they capture the skin area they want to examine. Our color scheme relies on the study of color psychology. Colors like tan and pink play a huge role for evoking feelings such as warmth and calmness in our users. In addition, the usage of tan color tones similar to tones of the human skin helps convey the nature of our service and implies that it is human at its core. In short, these design principles make up the prototyped we designed. The designs and renders are linked in the document under the “More Resources” section.
By the end of the hackathon, we aim to have a cross platform mobile app which our users will interact with. The app will have functionalities such as uploading an image, prompting users’ questions related to their possible disease, and displaying the results of users. Another component we are working on is an application programming interface (API) which will constantly communicate back and forth with our mobile app. The API will be responsible for analyzing the image submitted by the user to figure out which disease it is that they are more likely to possess.
In the current stage of our development, we believe our chances of being successful is around 6 out of 10. We are aware that our success will depend on multiple factors, some of which are in our hands, some of which aren't. Even though we are willing to put in the hard work for our product to be successful, we will also need medical experts' collaboration and financial support in order to pull this project to a level of our desired success.