Warn | Devpost

Inspiration

Our entire team was both amazed and enraged when we found out between 10 and 20 percent of all interactions with a physician lead to a misdiagnosis. We were even more upset by the fact that 30% of these diagnostic mistakes become deadly in some way. We were up to the challenge to combat this problem, armed with only our wits, datasets and Azure's Machine Learning power.

What it does

Warn allows users to submit a list of the symptoms they are experiencing, as well as any other medically-related data about them (age, sex, duration of symptoms, etc.). Warn then calls the Azure API and submits their data to an algorithm we developed which was able to diagnose patients correctly 95% of the time. Our algorithm was trained with thousands of rows of patient data, giving it the ability to diagnose a plethora of diseases with an accuracy which rivals the current medical system.

How we built it

Warn's algorithm was developed using Azure's Machine Learning Studio, and trained with anonymous patient data that we collected with the proper permission. Our web app hosted on an Apache server communicates with the Azure REST API through simple http Javascript requests.

Challenges we ran into

Some of us had been to hackathons before, but nobody in our group had used Azure or any of its features before. The tutorial removed some of the hurdles we were experiencing, but we still were pushed to ask the help of the mentors, especially when it came to sending data to Azure and displaying the result. This was also the first time any of us had used servers before, but we're proud we got a final product running

Accomplishments that we're proud of

We are happy to say that Warn correctly warns clients about the condition they are afflicted with 95% of the time (the score of the Azure model). Within 24 hours, with the help of Microsoft's Azure toolkit, we were able to create an algorithm which was able to predict conditions which rivals doctors.

What we learned

Azure allowed us to grasp the basics of machine learning, and became an invaluable tool for teaching us about some of the most breakthrough technology of today. We also learned about HoloLens, even though we didn't end up programming with it. The resources that MSFTHacks provided, the mentors, workshops, and tools, were crucial to learning about machine learning and HoloLens.

What's next for Warn

Warn understands that medicine and diagnosis isn't necessarily binary. We understand that the binary nature of our diagnosis system is a flaw, and we will definitely be fixing that in the future. With the future integration of new datasets and the ability to diagnose more diseases with a greater library of symptoms for inputs, Warn truly has the power to revolutionize and disrupt an industry which is badly in need of a shakedown. Perhaps in the future, we can develop an API for hospitals to call when they are struggling to diagnose a patient.