Inspiration
Tuberculosis is a major public health concern, affecting over 10 million people in 2018 alone. One of the biggest problems in treatment is ensuring that patients stick to their treatment regimen of antibiotics, as forgetfulness and the stigma against having tuberculosis prevents patients from getting better and also leads to antibiotic resistance,
Especially in developing countries, there are typically hundreds of patients per health worker, making it impossible to visit every individual's house to ensure they take their medication on time.
What it does
I developed a duel-training scheme with one prediction network that determines a score from 0-1 about how likely an individual is going to skip their medication that day (based on demographic information and past history of taking the medication using a system called 99Dots) and a differentiable approach to the orienteering problem (shortest path to visit as many weighted destinations as possible given a time constraint) to create a system that can generate a report telling health workers which patients they need to visit on a given day.
How I built it
Built using Pytorch and the CVXPY library (Differentiable Convex Optimization) in Python.
Challenges I ran into
Accomplishments that I'm proud of
What I learned
What's next for TB Prediction
Probably making a front-end to intuitively display the information necessary, maybe figuring out a way to incorporate informal health worker intuition based on other screener questions.
Log in or sign up for Devpost to join the conversation.