Inspiration
The sponsor presentation by Prudential about non-invasive diagnosis, impressed us with its utility. We were inspired to think of further applications of this type of technology that could help to efficiently triage patients for GP appointments as well as diagnose patients in areas with limited access to medical care.
What it does
The Rash Identifier is a website application that takes an input image from the user and identities if it is 1 of 4 skin conditions; Basal Cell Carcinoma, Dyshidrotic Eczema, Melanoma or Urticaria. Each input is compared to a model generated by a machine learning algorithm to identify which rash the input image shares the most characteristics with.
How we built it
To generate a database to train our model, we used a bot that we found on Github to download the Google Image search result for each type of rash. We also included images of normal skin. Next, we used a Tensorflow library to implement a machine learning algorithm to generate a model of the types of rashes and normal skin in our database. Finally, we created a website interface for the user to upload a photo of their rash to the server which returns the result of the comparison. This is hosted on a local server. For an accuracy of below 60%, the application says that it does not identify the image as a skin condition.
Challenges we ran into
Initially, we attempted to use Microsoft Azure Machine Learning but ran into some difficulties.
Accomplishments that we're proud of
Using Machine Learning through Tensorflow for the first time!
What we learned
How to make the best use of libraries! We learnt how to use Tensorflow and increased our understanding of backend website development, in particular, using Flask.
What's next for Rash Identifier
Ideally, the model we use would be deployed and hosted on a live server. To improve accuracy, we would like to regenerate the model using each new submission by the user. Furthermore, we hope to extend this project to a wider variety of skin conditions, including but not limited to insect bites and other types of rashes.
Log in or sign up for Devpost to join the conversation.