Inspiration

Big data analytics in healthcare field enables researchers to find insightful solutions to improve the health outcome, which will benefit millions of people who may suffer from potential health risks. Inspired by the prospective future of artificial intelligence applications used in healthcare field, we participated in the topic of ‘Hacking Healthcare with AI’ for SheHacksBoston.

What it does

We designed an application that can be used by both patients and physician to predict the probability of a patient getting diseases based on current health conditions and basic personal information. This AI tool could help medical professionals in designing treatment plans and suggesting actionable plans for reducing the risk of getting the potential diseases.

How we built it

We built a user interface to collect data from a patient, and it can display the predicted result. In the machine learning part, we used Medicare data (beneficiary and inpatient claims) from CMS in 2008. We trained feedforward neural network and multi-label classification models on Google Cloud Platform to predict the probability of getting additional diseases.

Challenges we ran into

The data is too big to process on a local machine. We moved to Google Cloud Platform and initiated a VM instance to enable the computing power. Also, it allows collaboration efficiently. Predicting the probability of getting additional diseases is a multi-label classification problem. It's our first time to work on such problem. By training different multi-label classification models and feedforward neural network, the predicting results are good.

Accomplishments that we're proud of

Thanks to the opportunity provided by SheHacksBoston, we were exposed to large-scale of healthcare data set and plunged deeply into the healthcare field. We used brainstorm to get insight from the data and came up with many ideas that can be applied in our future research.

What we learned

We learned how to use Google Cloud Platform and collaborate on it. In addition, we learned multi-label classification models and trained them with healthcare data.

What's next for Disease Prediction

Due to the limit of time and lack of domain knowledge in healthcare insurance, we cannot implement all the ideas in our minds. We will definitely continue our research and design more AI tools that help improve the efficiency of the medical system and balance the medical resource throughout the whole country.

Built With

Share this project:
×

Updates