What inspired you (or your team)?
Over the summer I worked at the United States Department of Veterans Affairs as a Data Scientist. My mentor, Dr. Oliver Aalami, told me about various projects he was currently working on which required a machine learning specialist. I chose to work on utilizing machine learning techniques to accurately classify surgical wounds. I spent about 3 months reading current deep learning literature and developing the model. I realized that this algorithm would have a lot of potential as a mobile application. I designed it with the main users (patients and doctors) in mind and handcrafted the elegant user interface. With his consent, I decided to submit the app, Theia, to this competition.
Tell us about the innovation, what materials or toolsets you used, what it does, how it works, why it is important.
Patients currently need to visit the doctor every few days to check up on their wounds after a surgery for the early diagnosis of postoperative issues, such as surgical site infections. Furthermore, these negative ailments are often not caught in time due to low patient compliance, leading to huge expenditures of both time and money from patients, healthcare providers, and insurers. I built Theia, an app that solves this problem. Theia enables patients to easily track postoperative wounds using their mobile devices. Every day, they can take a picture of their wound and have it analyzed by our deep learning algorithm for specific wound afflictions, such as infections, granulation tissue, or drainage, among many others. Patients can also input their daily medication and exercise, alongside the number of times they’ve changed their wound dressing. They can also partake in daily assessments, where they provide an image of their wound, their weight, and current pain rating. Theia creates vivid charts from this data enabling both patients and doctors to better understand trends in health. Finally, patients can create comprehensive wound reports and deliver them to both family and care providers via text or email on a regular basis. The app was built using the Apple iOS Software Development Kit. A core contribution of this project is automated wound assessment through a convolutional deep neural network. It was engineered in the Python programming language on a Tesla K80 GPU running in the cloud.