Radiology Application (RadApp)

Inspiration

Machine learning appears at first glance to be a perfect match for super-charging a radiologists day to day task but for some reason the tech hasn't made a big debut onto the scene. The goal of the RadApp is to use the blockchain as the bridge to providing anonymity to real time patient data while allowing all of the benefits of outsourcing predictions for double checking and building stronger predictive models.

What it does

RadApp takes a users patient data and a pathology labels and returns a label that either agrees or disagrees with the radiologist and to what degree(effective allows for double checking). The uploaded data is deidentified upon retrieval and the user is also anonymous providing HIPAA compliant data storage. The stored data can thereby be used for retraining and providing more deidentified data to the public.

How we built it

We used docker images for storing the machine learning environment and inference script. The Oasis blockchain would thereby query the image based on the data uploaded and provide the user a prediction.

Challenges we ran into

We ran into trouble integrating the docker image with the oasis api so we weren't able to fully flesh out the app. The frontend also didn't reach completion as we didn't have coverage in the side of the stack and enough time to throw something together.

Accomplishments that we're proud of

We have an initial database as well as strong starting point to build off of for more machine learning prediction scripts as well as docker image for deployment.

What we learned

We learned a lot about the oasis blockchain, IPFS, docker, and a bunch of other technologies that have really got us excited for the next steps of radapp itself. This hackathon has really jumpstarted the app and empowered us with the knowledge for how to move forward and what technologies we can use for different aspects of the project.

What's next for Radiology Application (RadApp)

We plan to spend the winter deploying the first iteration of the radapp through oasis as well as increase the amount of models for different pathology labels we currently take in. A major goal is to provide public access to the data aggregated as well as set up a retraining loop using said data.