Inspiration
With more than 5M people suffering from Crohn's Disease worldwide both directly and indirectly imposes new standards when it comes to managing symptoms for lifelong remission. Despite being a notorious medical condition where there is no cure available, expensive biologics treatment, disease making a comeback that desperately needs a solution where one could track his symptoms daily and conclude whether to seek medical attention or not. People with IBD do require staying in touch with their doctors to keep their physical health in check.
As a Crohn's Disease patient myself, I do make regular visits to the hospital to keep track of my IBD activity. Having a first hand experience with dreaded life threatening symptoms that requires me to watch on my bowel activity so that one could avert fatalities. I do believe having a first of a kind IBD symptoms tracker along with severity activity predictor could solve some of these issues which could provide long term remission for the patients.
What it does
A dedicated app for both doctors and patients.
Our mobile application allows eminent Gastroenterologist to provide input data from every visited patient who has been diagnosed with Crohn's Disease.
A TabNet Classifier model is trained with this input data upon which an inference is made from patients input data.
Prediction is made on the patients symptoms data, and thereby initiate them to take timely medical assistance.
Disease severity is classified as: very well, fair to good, poor and very poor.
A severity score is computed which ranges from 100 to 450. Those below 100 are classified as doing very well and those above 200 as doing very poor.
Detect Early Failure and seek urgent medical attention before it becomes fatal.
To save money by cutting down expensive treatment options like biologics. (Annual cost: $50K)
Daily log of symptoms related to Crohn’s Disease.
Disease activity indicated by severity score.
How we built it
Flutter for building cross platform mobile application.
Firebase Firestore to store both doctors and patient input data.
Firebase Authentication to authenticate new users.
FastAPI to expose endpoint.
AWS EC2 for deployment infrastructure
Pandas & Sci-Kit Learn for data manipulation and categorical encoding
Challenges we ran into
No ground truth data was available therefore generated synthetic data, modelled after Gastroenterologists input.
TabNet's built-in "save model" seems to be broken, we decided to completely expose our train-test-eval method through FastAPI to make predictions.
Accomplishments that we're proud of
From ideation to implementation of working proof of concept is a great source of pride for us.
It requires a good knowledge of all systems involved and, Time management principles for successful execution of the project.
Despite having issues saving the model we still made exposing its endpoint possible.
Learning Flutter itself was a big learning curve.
What we learned
Time management principles.
Medical terminologies.
Concept Ideation to implementation.
Training tabular data on deep learning framework like Pytorch
What's next for Crohn's Disease Activity Tracker
Federated Learning (HIPAA Compliant)
Elemental Diet Protocol (Promising Results)
Include Colon Cancer
Like Minded Community Support from fellow Crohnie's.
Log in or sign up for Devpost to join the conversation.