diagnosis.ai

There are 37 million American people who are classified as diabetic and this number continues to rise. With diagnosis.ai we can now fight to bring this number down by pre diagnosing those at risk.

Comment

Inspiration

When I, Abhiram Pendela, was in the second grade my mother had been diagnosed with type two diabetes. For the next three years, she had been using tablets that were of no use. This is because she was misdiagnosed. In fact, she had type one diabetes, which is a condition that cannot be improved by tablets. Now, she uses insulin injections to control her blood glucose. I have firsthand experience of the effects of an incorrect diagnosis.

What it does

Our application takes in medical features as an input, returns a diagnosis of whether a patient is diabetic or not, and also provides a percentage of how confident the model is. This confidence percentage is important to determine whether its a close call or not. This website can be scaled at a massive level and is very simple to use. Its a practical application of a tool which doctors can use to double check their diagnosis is.

How we built it

We created a machine learning pipeline on the Microsoft Azure Machine Learning Studio. Using a reputable database from the Kaggle website, we trained a Two-Class Logistic Regression model to make predictions. Then, we created a real time inference pipeline on Azure and connected it's REST endpoint to our Flask application. Our application takes in input from the user and sends requests to the REST endpoint. Finally, it displays the value returned from the model to show the diagnosis.

Challenges we ran into and ehat we learned

Our main challenge, one that we predicted before the competition even started, is integrating front end and backend. Our team doesn't have much experience using a REST API in an application and thats something our group learned.

Accomplishments that we're proud of

As this is our group's first hackathon, we are very proud to have a fully functioning application that takes in an input and returns an output from a machine learning model.

What's next for diagnosis.ai

We want to improve our model by using more high-quality data from various age ranges. Our database currently has a lack of data from younger patients (aged one to twenty )in particular. Getting data from younger patients and training our model will result in much more valuable diagnosis for younger patients to detect diabetes on its onset. Finally, we also want to test different algorithms to further improve the accuracy of our model

Our Video (A drive link just in case)

https://drive.google.com/file/d/1ciWtUfng7Y7w_rONWwI1NbftmQ61ydsz/view?usp=sharin

Share this project:

Updates