Inspiration
One of our team members once fractured his arm as a young child, and was taken to the hospital, where over a dozen X-Ray photos were taken of his arm by different doctors overnight. Had just one photo been taken and properly classified in a timely manner, he would not have needed to be subject to so much radiation that night, and he and his family could have left the hospital earlier.
What it does
Our project uses a machine learning model trained on hundreds of X-Ray photos of different types of bone fractures so that it can accurately classify the type of bone fracture in new photos that are fed to it.
How we built it
We built the project using python with the tensorflow and pandas libraries for the machine learning model and data handling. We used google colab as our development workspace, which had built-in collaboration support.
Challenges we ran into
No one on our team had ever done a project like this before, so we were learning as we went. As we tried to figure out how to move forward as we coded, we consulted stack overflow and phind AI, a GPT-4 model for developers. Issues behind stack overflow (or just google searching in general) was that we had to find online posts relevant to our project, and issues with phind AI were that it could generate faulty code, or sometimes suggest revisions that led to new issues, then suggest our old code again when we told it that it's new code had new issues.
Accomplishments that we're proud of
We're proud to have gotten this project working in the end, as this was a monumental accomplishment in and of itself for all of us.
What we learned
We all learned much more about how ML works, specifically how to use tensorflow to generate ML models and train them on data, specifically image data as opposed to something more straightforward like a csv of spreadsheet values.
What's next for Broken Bone Classifier
What's next for our project could include giving it a GUI so that users can easily upload photos to have the model classify, and perhaps we could spend more time fine tuning various configuration variables for the best results and accuracy.
Log in or sign up for Devpost to join the conversation.