Strokely

• 

  logo

Inspiration

According to a study from the University of Buffalo and Vanderbilt University, there is a significant health disparity in stroke diagnosis times and clinical outcomes between low-income countries and high-income countries. Many of the citizens of these low-income countries do not have knowledge of stroke procedeures, or how to identify symptoms.

Additional bottlenecks for doctors in low-income country hospitals are the amount of patients; many of these countries are densely populated. Combined with the problem of low stroke literacy, this means that doctors get many patients (who are normal) who think they have stroke, which takes away time doctors could have used to treat an actual stroke patient.

What it does

Strokely aims to combat these two bottlenecks with a simple questionnaire, coupled with a facial detector. Based off scorings from the American Heart Association (AHA), we have a basic "at home" Stroke App.

How we built it

We used IEEE datasets for training our model to recognize patients with stroke; we used Streamlit for our frontend and tensorflow to build the model. The scoring was derived from the base paper in the Amarican Heart Association.

Challenges we ran into

Finding a medical problem that could practically be solved with programming was a challenge, as we wanted to specifically create an application that could be deployed in low-income demographics. We iterated over our idea after looking through a few papers, as well as after talking to the TerraAPI sponsors. Some technical challenges we ran into, besides training the model, was configuring the web application in Streamlit.

Accomplishments that we're proud of

We got the machine learning model to train and predict reasonably well (90% accurate), all while enjoying the event and socializing.

What we learned

We learned about how machine learning can really be useful in created an automated way to help patients get an at home diagnosis.

While researching this, we also really gained a deeper understanding of the stroke literacy disparity between high and low income regions, and how we can mitigate those disparities by building and app for stroke detection.

What's next for Strokely

For the future, we can make a neuromorphic embedded chip so that we can efficiently run neural networks and provide this kind of stroke detection technology to underdeveloped countries to make healthcare accessible to everyone.

Also, in order to make meaningful conclusions from our app, we have to make and run clinical trials. We will reach out to potential collaborators from academic hospitals in underdeveloped demographics, and get IRB approved. Once we are satisfied with the accuracy of the model, we will then develop a map system that can route the patient to the nearest clinic that provides services for stroke care.

Built With

• pandas
• python
• streamlit
• tensorflow